Development of bifacial n-type solar cells for industrial application

Transcription

1 Development of bifacial n-type solar cells for industrial application Dissertation zur Erlangung des akademischen Grades des Doktors der Naturwissenschaften (Dr. rer. nat.) an der Universität Konstanz Mathematisch-Naturwissenschaftliche Sektion Fachbereich Physik vorgelegt von Alexander Edler Tag der mündlichen Prüfung: Referent: Prof. Dr. E. Bucher 2. Referent: Prof. Dr. M. Fonin

2

3 This thesis will get an A There are no passengers on spaceship earth. We are all crew. Richard Buckminster Fuller ( ), American inventor and architect, one of the first strong supporters of renewable energy We re killing people in foreign lands in order to extract 200-million-year-old sunlight. Then we burn it in order to boil water to create steam to drive a turbine to generate electricity. We frack our own backyards and pollute our rivers, or we blow up our mountaintops just miles from our nations capital for an hour of electricity, when we could just take what is falling free from the sky. Danny Kennedy (1971-), clean-technology entrepreneur and environmental activist

4

5 Contents 1 Introduction Background and Motivation The Zoo of n-type cell concepts Aluminium rear emitter cells PERC, PERL and PERT cell architectures Interdigitated back-contact cells Heterojunction device concepts Outline of the thesis Manufacturing technology and relevant concepts Cell processes and Cell Manufacturing Material Wet Chemical Processing High Temperature Processes Passivation Metallization Solar Cell Characterization and Definitions Cell parameters and IV characterization Optical losses Recombination losses and lifetime measurements Resistive losses and resistance measurements Wafer characterization Boron emitter diffusion and passivation Diffusion of boron in silicon Bulk lifetimes of silicon wafers after high temperature processing Experiment Results and Discussion Conclusion Homogeneity of emitter diffusion and BSG layer Conclusion Passivation of highly boron doped surfaces Introduction Overview of passivation methods Experimental results iii

6 iv CONTENTS Conclusion Cell processes Cell processes with homogeneous diffusions Potential of cell precursors in PC1D device simulations Cell processing of full ingot range Cell process with lowly doped emitters Cell process with optimized BSF diffusions Concepts for selective cell doping Motivation Laser doping for selective emitter and BSF formation Screen-printable doping pastes Chemical etch-back for selectively doped structures Conclusion Rear emitter cell concept Motivation Experiment Conclusion Metallization of bifacial solar cells Introduction Experimental Details Results and Discussion Identification of main losses Losses in open-circuit voltage Quantifying the metallization losses Modelling the metallization losses Conclusion Measurement of high efficiency solar cells Motivation Measurement uncertainties of bifacial solar cells Flasher measurements of high efficiency n-type cells Introduction Observation Explanation Maximum power point scan Conclusion Bifacial measurements of solar cells Motivation Experimental setup Experimental evaluation of bifacial properties Conclusion Summary and Outlook 107

7 CONTENTS v Zusammenfassung 111 List of Acronyms, Symbols and Constants 115 References 130 List of Publications 131 Acknowledgement 133

8

9 1 Introduction 1.1 Background and Motivation The first practical solar cell was invented in 1954 at the Bell Laboratories and was a rear contact technology and based on arsenic doped silicon [1]. With an efficiency of 6% it presented an efficiency leap from former device architectures. Their invention was praised by the New York Times to mark the beginning of a new era, leading eventually to the realization of one of mankind s most cherished dreams-the harnessing of the almost limitless energy of the sun for the uses of civilization. For the following two decades energy generation from solar cells was extremely expensive. Consequentially solar power was applied mainly in space applications, where the higher power-toweight ratio over depleting energy sources mattered. These early applications drove the inventions in the solar field. It was then demonstrated that p-type based cells showed higher radiation hardness than n-type based solar cells in the space radiation environment. Accordingly the research focus shifted towards cells based on boron or gallium doped p-type silicon. It is this early focus on manufacturing techniques for p-type cells that is at least partly responsible for the dominating position of p-type solar cells in the industry today. At the current stage of development it seems that p-type efficiencies have saturated. Over the years many researchers have acknowledged the higher suitability of n-type Cz-material for solar cell processing [2 4]. A shift from p-type to n-type based cell concepts could consequentially be the next evolutionary step. However, this requires the adoption of new fabrication methods and sequences. Although many of those have been successfully demonstrated in lab scale, manufacturers are only reluctantly moving in that direction. It can be assumed that the booming expansion in the first decade of the century and the preoccupation with p-type manufacturing, has lead to an inherent inertia of the industry. Self-explaining, any new emerging technology has to prove its competitiveness compared to current technology. If a technology has evolved over many decades, like the p-type cell technology, it is possible that adapting a new technology at one point becomes prohibitively costly. This can be the case, even if the inherent properties argue for that new technology in the long run. In this case economists speak about path-dependency or lock-in phenomena to describe this barrier. Several hints indicate that such a scenario exists in the photovoltaic industry today. Examples can be found all along the standard silicon photovoltaic production sequence. State-of-the-art phosphorous diffusion in tube furnaces is far more advanced than boron tube diffusions, n + surface passivation using PECVD SiN x is well established, whereas suitable boron emitter passivation techniques have accumulated only little production experience. The metallization of n-type cells, and boron emitters in particular, requires new paste formulations and state-of-the-art products are not nearly as advanced as their p- 1

10 2 Chapter 1: Introduction type counterparts. Nevertheless, even with current production methods, the attainable efficiencies rival the best p-type production results, indicating that n-type cell technology is ready for the transition to the industry. Due to the novelties and innovations involved, the easiest way to introduce new n-type based cells will be to maintain as far as possible production methods well known from p-type manufacturing. This approach comprises using tube diffusions for the formation of emitter and/or back surface field, PECVD deposition of dielectrics for passivation and screen-printing and firing for the formation of the contacts. The reluctance of the industry, due to decades of p-type manufacturing experience, must be overcome today to enable the continuation of the development of PV production efficiencies. It is the scope of this work to understand the limitations and improve the efficiency of one of the closest neighbours of the p-type PERC cell, the n-pert concept. Figure 1.1: Learning curve for the price of c-si module as a funtion of cumulative installed capacity, from Ref. [5] Solar cell market situation Although the solar cell market has been growing rapidly over the last 10 years, the market is in turmoil. In a market that was largely driven by government incentives and is therefore highly fluctuating it is very difficult to anticipate market development in the long term. For several years in a row the market growth exceeded all expectations. The massive ramp up of production capacity, especially in China, was the result. As a consequence the industry has built up a massive oversupply of production capacity. In 2012 solar installations with a rated power output of 31.1 GW have been installed, this was met by global PV production capacity of around 57 GW [6]. As a result the price for solar modules has plummeted below the manufacturing costs for many suppliers. The price decline can be represented by the famous learning curve for PV (Fig. 1.1), which is the result of thorough research and development work that transitioned from laboratories to production lines while perfecting the manufacturing techniques and enabling laws of scale. The oversupply has lead to an accelerated price reduction, which can be seen from the kinks in the learning curve. While this makes solar modules currently as cheap

11 1.2. The Zoo of n-type cell concepts 3 as 0.69 USD/W peak for the end user, it is at the same time a dangerous situation for an emerging industry. The low prices impede the ability of manufacturers to invest in R&D or new production equipment. Many large manufacturers have already been driven into insolvency. In order to reduce manufacturing costs to progress along the learning curve it is therefore most likely to rely on industry proven processing that enables new cell concepts. 1.2 The Zoo of n-type cell concepts It has been demonstrated that current p-type solar cell architectures are increasingly limited by inherent bulk recombination from the metastable boron oxygen defect [7]. For advanced cell concepts the bulk limitation has to be overcome. The substitution of boron by gallium is discussed, but leads to a very large resistivity spread over the ingot and interstitial iron is still an issue causing degradation. The magnetic Cz (MCz) approach is another option to reduce the influence of this defect by reducing the oxygen concentration in the crystal. It is however very expensive and therefore not industrially viable today. Even if the boron oxygen defect could be suppressed, the sensitivity of p-type for common metal impurities is higher than for n-type material due to asymmetric capture cross sections of their defects [2, 4, 8]. This could still be the limiting factor for the bulk lifetime. However this highly quoted assumption should be taken with care. On n- type cells with boron emitter even higher cleanliness requirements might apply since the boron diffusion is extremely sensitive towards emitter bulk contamination. Even though the bulk might enable high efficiency limits, the recombination within a contaminated p + emitter can easily destroy the cell performance. In case of high impurity concentrations in the bulk also the limited gettering efficiency of boron diffusions [9] can become a drawback for example for cell processes on multi-crystalline n-type material [10]. Every p-type module installed today suffers from light induced degradation (LID), which reduces power output substantially over the working life of a module. This effect is fully negligible in cells made from non-compensated n-type silicon. Today it is taken as normal by an installer, that the energy output of a Si-module diminishes over its working life, simply because p-type modules have been the only type available. The stable performance will certainly be a major advantage for all kinds of n-type devices, but it will require marketing efforts. Many n-type cell concepts moreover are inherently bifacial. The module assembly can then be adjusted to make use of the bifacial properties, enabling new ways of installation in the field but also for building integrated photovoltaic. Last but not least there is evidence that n-type devices are more sensitive to low light intensity as well as having a lower temperature coefficient mostly due to the higher V OC. All these effects can lead to a remarkably higher kwh energy harvesting and reduction of BOS (balance of system) costs due to lower area needs for the same installation capacity. The following list gives a non-exhaustive overview of existing n-type cell concepts.

12 4 Chapter 1: Introduction Aluminium rear emitter cells The closest relative of the aluminium BSF p-type cell employing n-type material, is the aluminium rear emitter cell. Basically it can be fabricated using the process sequence of standard p-type Al-BSF cells simply by starting with n-type Cz material. This makes the process appealing at first, since existing production lines can be re-fitted to produce this cell. The potential of the cell concept has been evaluated by Rüdiger et al. [11] and efficiencies of 19.4% on 6 inch wafers have been demonstrated by Book et al. [12]. Ultimately, the benefits over its p-type equivalent are limited. The poor electrical properties of a full area aluminium emitter limit the cell V OC and the internal optical properties put an upper bound on the achievable cell current density J SC. Also the module interconnection imposes challenges, due to the aluminium alloyed rear emitter PERC, PERL and PERT cell architectures An advanced line of cell concepts deals with improving the rear surface compared to aluminium BSF cells. As a hint about the time of their conception their acronyms indicate what these concepts have in common, namely a passivated emitter. Today, a modern cell cannot be imagined without emitter passivation, the rear surface, however, much looks like it did since the aluminium BSF was invented. The PERC (passivated emitter and rear cell) concept, which is gradually being adopted by the industry, adds rear side passivation to the structure. As rear side dielectric commonly Al 2 O 3 is employed. The rear side is opened locally, usually by laser ablation, to allow for local alloying of aluminium. This forms the BSF and contacts, thus improving the recombination and optical properties. Accordingly the PERC concept has been promoted for many years as the next candidate for widespread industrial implementation. By now initial efficiencies exceeding 20% have been demonstrated by many manufacturers. This approach can not easily be transferred to n-type cells. In case of a PERL (passivated emitter, rear locally diffused) cell this is refined even further by diffusing a local BSF, which is contacted by evaporated and plated metal contacts. The long-standing efficiency world record for silicon devices was established using such a structure. Zhao et al. [13] reported 24.7% on a designated area of 4 cm 2 on FZ p-type wafers. This is achieved by combining all cell design ingredients necessary for achieving highest efficiencies. Those include amongst others selective front doping, ideal inverted pyramid texture with double layer anti-reflection coating, an evaporation coated rear side for improved light trapping and front contacts structured by photolithography. Using a process of similar complexity 23.4% efficiency has been shown for 4 cm 2 n-type PERL devices by Benick et al. [14, 15]. Needless to say that these processes are far too complicated and costly for a production environment. The direct contender of the p-type PERC cell in the race for industrial implementation can be found in the n-type PERT (passivated emitter, rear totally diffused) cell discussed in this work. These can be fabricated with front emitter or as back junction cells with an emitter on the rear side. The best lab results for front junction n-pert devices have been reported by Zhao et al. [16] with 21.1% on small area Cz and 21.9% on FZ material. Newer experimental results using Al 2 O 3 -based passivation were reported by Richter et al. [17], who could demonstrate 20.5% on 4 cm 2 FZ cells with a p + nn + structure using advanced metallization techniques. On 5 inch Cz-material still 19.6% has been shown by

13 1.2. The Zoo of n-type cell concepts 5 the same group using industrial diffusions but still evaporated contacts. Developments on back junction devices with rear boron emitter have been pioneered by Q-Cells/Hanhwa as reported from Bordihn et al. [18]. In a recent update by Mertens et al. [19] efficiencies up to 21.3% on large area Cz based solar cells have been shown, using PVD aluminium metallization on the rear side. Several research groups with a focus on industrial implementation are currently working on the front emitter version of this cell concept. ECN (Energieonderzoek Centrum Nederland) is promoting this concept under the name n- Pasha (towards 20% [20]) and attempts to merge the concepts with metal wrap through (MWT) technology have been successfully demonstrated [21]. Also in France, at CEA- INES (Institut national de l énergy solaire) these cell types are being developed, recently the efficiency of 20.1% [22] was reported. Both share the formation of doped layers with conventional tube diffusions, passivation by aluminium oxide or nitric acid oxide based passivation stacks and industrial screen-printing. The advent of ion implantation for solar cell manufacturing also has implications for the process sequences of n-pert cells. At CEA-INES phosphorous implanted or fully implanted n-pert cells achieving 19.9% and 19.5% maximum efficiency respectively have been fabricated. Recently also BOSCH Solar Energy AG reported n-pert results from pilot line production of up to 20.6% using implanted phosphorous BSF structures [23]. These results have been achieved in a collaborative project between ISC-Konstanz and BOSCH Solar Energy AG also including work presented in this thesis. P-type PERT cells with full area boron BSF have also been demonstrated [24] Interdigitated back-contact cells The interdigitated back contact (IBC) cell concept is characterized in that all metal contacts can be found on the rear side of the cell. This enables a front side optimized entirely for higher light absorption without the requirements of contact formation. Since the contacts are all on the rear side, the contact layout can be optimized for transport properties irrespective of shading considerations. These advantages stand in opposition to higher process complexity due the requirement of patterned diffusions and the risk of fatal cell shunting due to the proximity of n + and p + doped regions. At ISC-Konstanz a process for IBC cells based on 6 inch Cz n-type wafers has been developed. By using only industrial processing techniques an efficiency of 21% could already be demonstrated. For back-contact back junction designs high bulk lifetime is further required to achieve highest efficiencies as analysed by Granek [25]. This structure can also be fabricated using the aforementioned aluminium emitter, but will then suffer from the same limitations. A report on the status of development for this cell is given in Ref. [26]. The most prominent example for IBC cells comes from the company Sunpower Corp. With a rated maximum efficiency of 24.2% and a reported average production efficiency of 23.6% their implementation of the IBC concept stands out from the range of commercially available silicon solar cells [27]. These results are achieved on 5 inch wafers and employing a very complex production sequence.

14 6 Chapter 1: Introduction Heterojunction device concepts A new category of devices that combines the advantages of n-type bulk material and semiconductor grade passivation are heterojunction structures. Here a high quality n- type substrate is coated with stacks of intrinsic and doped amorphous silicon to give a p + i n i n + structure. The wafer is coated on either side with indium tin oxide (ITO) to enhance current collection from the contacts. The a-si enables superb passivation of the bulk silicon and enables a tunneling junction to the doped emitter layer. This both results in extremely high voltage potential and low temperature coefficients ( 0.23%/ C). The cost benefit of low temperature deposition for the amorphous layers leads to the requirement of exceptionally high bulk lifetime. Additionally the low temperature processing forbids the use of firing through metallization. The commonly employed low temperature contact pastes show significantly larger resistances compared to their sintered counterparts. Despite these drawbacks very high efficiencies have been demonstrated by several companies. The most famous example certainly is the concept of heterojunction cells with intrinsic thin layer (HIT) from Panasonic, earlier Sanyo. The certified peak efficiency was just raised by one percent absolute to 24.7% for a 100 µm thick cell on cm 2. One competitor, Kaneka, reported a certified efficiency of 24.2% on large area cells, using copper plated contacts at the same time. These results demonstrate the very high efficiency potential of this concept. It is expected that average production efficiencies range 1-2% lower than the best cell efficiencies in the research environment. Other companies, like Silevo, a California based start-up, and Tetrasun, a subsidiary of First Solar, are commercializing slightly different proprietary cell technologies. Their cell concepts combine an n-type silicon bulk, with an amorphous silicon emitter. This heterojunction architecture enables over 730 mv in V OC. A semiconductor grade tunneling oxide is deposited on top, forming a tunneling contact to the copper plated metallization. Peak efficiencies of 22.1% have been demonstrated on 239 cm 2 wafer size. An average production efficiency of 21.3% was also reported recently [28]. 1.3 Outline of the thesis This introductory Chapter 1 gives a summary of the general motivation for this research. In the previous section many cell concepts have been introduced. Some of them are being applied commercially, but generally they are adopted only by few and very specialized companies (e.g. Panasonic, Sunpower, Silevo). The largest group of cell or module manufacturers is still producing mainly p-type aluminium BSF cells. It is assumed that several technological challenges have so far prevented a widespread adoption of next generation n-type cells and it is the scope of this work to understand and improve these processing and measurement steps. In Chapter 2 we introduce the employed manufacturing tools and concepts of solar cell characterization. The facilities at ISC Konstanz embrace a pilot scale solar cell manufacturing line which is shown here. It can be employed to process various different cell types and wafer sizes. The extensive characterization equipment is needed to allow comprehensive loss analysis of solar cells and wafers.

15 1.3. Outline of the thesis 7 The analyses in Chapter 3 are intended to tune the properties of a boron emitter and its interface. Tube diffusion of boron is commonly perceived to be more complicated than phosphorous diffusion. We have reviewed the properties of boron diffusions and focused on the formation of homogeneous and stable boron emitters of different sheet resistance. The concerns about associated lifetime degradation could be invalidated. A different important aspect about n-type cell technology is the passivation for p + surfaces. Over a long period only thermal oxide was suited to reliably passivate boron emitters. In the course of growing research interest in n-type technology Al 2 O 3 -based passivation has received much attention. Here we have investigated the properties of an alternative passivation stack based on boron silicate glass (BSG) and compared it to state-of-the-art methods. Chapter 4 deals with the cell processes for n-pert cells. The tolerance of the cell process towards utilization of base material with different resistivity is an important aspect in view of the large resistivity spread found for n-type ingot. The impact of doping profiles of emitter and BSF has been evaluated in experiment and simulation. Especially the BSF doping can be adjusted to benefit from new developments of Ag contacting pastes. Also the implications of two consecutive high temperature steps for the cell potential have been tested. The high doping concentration and especially metal contact recombination has been demonstrated to constitute main loss mechanisms. Therefore processes for the selective doping of emitter or BSF have been evaluated. In Chapter 5 we analyse the metal related recombination losses in-depth. A new technique using printing layouts with different metallization fractions is employed to distinguish between the contributions from front and rear side contacts. The losses are quantified in terms of J 0e(met), to demonstrate that metal induced recombination is indeed the primary loss mechanism for this concept. It is found that the damage at p + contacts is far more detrimental for minority carriers than the n + BSF contact. Moreover the effect is also stronger than for n + emitters of conventional p-type cells. A simulation model was built to explain the effects underneath the contacts in detail. Chapter 6 deals with the adequate IV characterization of bifacial high-efficiency cells. The high V OC of advanced cell concepts brings up the topic of hysteresis errors in fast IV measurements. A strategy to cope with this problem with limited flash durations is suggested. The open rear architecture, found in many n-type cells, leads to uncertainties in the J SC generation depending on the measurement background. The standard testing conditions (STC), so successful in allowing comparability for all kinds of devices, do not encompass bifaciality. We have constructed a sample holder to measure the cell performance under varying bifacial illumination conditions. It is suggested that STC should be reviewed to enable a fair assessment of this property, in order to allow marketing of the higher energy yield of such modules.

16

17 2 Manufacturing technology and relevant concepts In this chapter we will introduce the manufacturing steps that are involved in the advanced solar cell process, basic solar cell parameters, as well as the characterization methods employed throughout this work. 2.1 Cell processes and Cell Manufacturing Material The starting material are phosphorous doped wafers from Czochralski grown ingots (Cz), that were slurry sawn to a thickness of 180 µm. In an n-type ingot of typical length the resistivity range is larger compared to a p-type crystal doped with boron. This is due to the lower segregation coefficient of phosphorous (0.35) compared to boron (0.8). The doping concentration along the crystal length for perfectly mixed melts is described by the Scheil equation 2.1, wherein k is the segregation coefficient, C 0 the total doping concentration in the melt and f s the fraction solidified. c( f s ) = k C 0 (1 f s ) k 1 (2.1) Fig. 2.1 shows a typical resistivity distribution, which was measured in an n-type production ingot as an example. We have mainly employed wafers from the middle regions of the ingots, showing a resistivity from 3 5 Ωcm. The doping concentration in Figure 2.1: Exemplary resistivity distribution along a Cz n-type crystal. 9

18 10 Chapter 2: Manufacturing technology and relevant concepts the wafer is an important quantity in solar cell design and the tolerance of a cell process towards a variation in resistivity needs to be tested. For Cz ingots the usable fraction solidified for a given cell process is a key parameter which determines the wafer cost. An interesting approach to get around this limitation is the continuous feeding method. Here the dopant concentration in the melt is held at a constant level by continuously adding doped material during crystal growth. Ingots with very uniform resistivity and high bulk lifetimes over the length of the whole crystal have already been demonstrated Wet Chemical Processing At the beginning of the cell process the saw damage is removed with alkaline (NaOH) etching of 5 7 µm of silicon on each side. After saw damage removal and before every high temperature or passivation step either one of two cleaning sequences is required. The cleaning procedures employed at ISC-Konstanz are listed below, whereas the first one being called standard cleaning and the second one IMEC cleaning hereafter. Standard cleaning Batch process, max. 50 Wafers/carrier, DI H 2 O rinse, 3% HCl (5min) bath, DI H 2 O rinse, 2% HF (2min) bath, DI H 2 O rinse, drying at 110 C IMEC or Piranha cleaning Batch process, max. 50 Wafers/carrier, comprises the standard cleaning followed by H 2 SO 4(conc) /H 2 O 2 (80 C, 10min) bath, DI H 2 O rinse, 2% HF (2min) bath, DI H 2 O rinse For the laboratory process we perform a full IMEC cleaning sequence before each high temperature or passivation step. This sequence is too complicated to be implemented in a production environment and the cleaning efficiency of different cleaning steps has also been tested and is presented by Buchholz et al. [29]. It could be demonstrated that also less sophisticated cleaning sequences can reduce the surface contamination sufficiently to achieve high lifetimes in the solar cell process. In order to reduce the reflectivity surface texturing in a potassium hydroxide (KOH) and isopropyl alcohol (IPA) bath is employed. For monocrystalline wafers with (100) surface orientation the anisotropic etching of (111) over (100) surfaces results in a random pyramid texture. This reduces the reflection of the surface to about 11%. The reflection is further minimized by the deposition of an anti-reflection coating (ARC), such as silicon nitride (SiN x ). In the standard solar cell process 70 nm of SiN x are deposited, for passivation stacks comprising different dielectrics the thickness of deposited SiN x has to be adjusted High Temperature Processes For the bifacial n-type solar cell two high-temperature steps are required to form the emitter and BSF regions by diffusion of dopant atoms. At ISC-Konstanz the high temperature diffusion steps are carried out in an industrial quartz tube furnace from Centrotherm. The furnace features four separate tubes, whereas wet oxidation (O 2 DCE), POCl 3 -based phosphorous diffusion and BBr 3 -based boron diffusion can be carried out. The wafers are loaded into quartz holders which are driven into the tube. At 4 mm spacing up to 200 wafers can be diffused simultaneously. The gas inlet for the precursor gases is situated on the left side of the tube, while the exhaust outlet is close to the loading

19 2.1. Cell processes and Cell Manufacturing 11 door. The wafers are always positioned standing upright with the surface to be diffused facing the gas inlet. Since the boron silicate glass deposition is more homogeneous on the side facing the gas inlet, lower variation in sheet resistance is measured on that surface. Throughout this work diffusions are performed with a capacity of wafers. The homogeneity of the sheet resistance over the wafer and over the tube position may vary, depending on the amount of wafers. The second high temperature step also influences the dopant distribution of the initial diffusion step. The doping distribution is hence defined by the combined thermal budget as well as the surface capping layer. Throughout this work the formation of the boron emitter has been performed as the second high temperature step. There are several reasons for that. First, the risk of cross contamination of the boron emitter with phosphorous atoms during diffusion makes the sequence where boron is diffused second favourable. It has been found that minimal contamination of phosphorous, as might leak through pinholes or other imperfections in the protection layer, can degrade the cells through shunting effects. In any process sequence with an initial boron emitter diffusion a diffusion barrier would be required. The protection against cross contamination with amorphous dielectric layers is difficult especially for textured surfaces. Such layers would have to be easy to remove even after high temperature densification and at the same time dense enough to prevent any phosphorous penetration. PECVD deposition of SiO/SiN x stacks is currently the only viable method for this application. The second motivation to start with the phosphorous diffusion is the potential utilization of BSG as a passivation layer. Since it was shown that the passivation quality degrades during further high temperature diffusion steps, the only option is to employ the emitter forming step at the end of the process Passivation At the surfaces of a semiconductor wafer the crystal structure is abruptly interrupted. This discontinuity in the crystal order can be imagined as a big crystal defect. The formation of discrete energy bands is hence disturbed and multiple defect states exist within the bandgap. These defect states allow for multistep recombination of minority carriers, which drastically limits the lifetime. Therefore all surfaces need to be passivated in order to reduce the recombination activity. This can be done in two ways. Chemical passivation reduces the density of defect states in the bandgap, i.e. by saturating dangling bonds at the surface with hydrogen. In the case of field effect passivation electric fields keep the region close to the surface depleted of one type of carriers to avoid their recombination. Generally different dielectrics are required to passivate surfaces with different dopant type and surface concentration. Especially the industrial passivation of the p + surfaces required extensive development as is reported in Section 3.3. Available methods at ISC- Konstanz for the passivation of boron emitters are thermal oxidation followed by PECVD deposition of SiN x and a stack system of BSG/SiN x. Phosphorous surfaces are best passivated by thermally grown SiO 2 and SiN x, but pure SiN x can also effectively passivate the surface. As for all components of solar cells it is required to demonstrate not only high initial passivation quality but also to confirm its stability in the interplay with other manufacturing steps and the durability in the long operating life of the cell. Special requirements in this respect would be firing stability during contact formation, the long

20 12 Chapter 2: Manufacturing technology and relevant concepts term stability, e.g. against UV radiation and compatibility with common anti-reflection layers Metallization Metal contacts are required on surfaces of either polarity to extract the generated current from a solar cell. These contacts are widely formed using screen-printing of metal pastes. This technique has a long history in solar cell manufacturing due to its favourable properties, such as high throughput capability and reliability. In this step the grid design is defined by printing a mesh of thin metal lines (fingers) interconnected with broader conductive paths (busbars) on the light sensitive side. While the rear side of a p-type cell is fully covered with aluminium paste, the diffused and passivated rear surface of the n- type cell also allows for the application of an open grid on the rear side. The metal pastes required for solar cell contact formation differ widely depending on the type of cell and contacting purpose. For bulk silicon solar cells in general silver based pastes are used to form the front grid pattern due to the high conductivity of the sintered silver. The basic ingredients of contacting pastes, beside the metal powder, are glass frits, binders and an organic vehicle. The contact formation to p + doped layers further requires the addition of aluminium in the percentage range to the silver paste. All commercial products contain aluminium since it is currently the only option to achieve low contact resistivities. At the same time, however, the aluminium reduces the line conductivity and leads to serious surface damage, representing a major drawback for n-type based cells. After the printing step the paste is dried in a drying furnace at 200 C where the solvents evaporate. The cells are then transferred to an IR furnace, where the contact to the silicon is actually formed. In the IR furnace the solvents and binders are further burned off at temperatures up to 550 C. Above that temperature the glass frit liquefies and etches the passivation layer while the silver powder starts to dissolve. The glass frit oxidizes the silicon surface and the liquid glass lead phase is saturated by silver. Upon rapid cooling from the peak temperature around 800 C, silver crystallites grow epitaxially into etch pits at the silicon surface. The sintered silver bulk is separated from the silicon by a thin SiO 2 layer, with embedded silver precipitates. This glass layer plays a vital role in establishing mechanical adhesion of the contacts. It is assumed, that especially direct current paths via silver imprints connecting the silicon and silver bulk are responsible for a low contact resistivity. Growth of such silver crystallites is promoted by sharp surface features, as well as excess surface doping concentration. During this work different generations of commercially available contacting paste have been employed. Each was fired at the best known firing settings, which were individually determined during firing optimization test runs. The firing step moreover fulfils the function of enabling hydrogen passivation, which diffuses from the SiN x to the interfaces during the temperature step. The metallization scheme has a great influence on the cell performance as will be shown in Chapter 5.

21 2.2. Solar Cell Characterization and Definitions Solar Cell Characterization and Definitions The following section explains influences that limit the conversion efficiency of solar cells and methods that have been employed to characterize wafers and solar cell performance during this work. Like any other solar cell, devices made from silicon in principal have to fulfil two very basic tasks. The first is light absorption in order to generate electron hole pairs and the second is the separation of these generated electron hole pairs. Silicon is an indirect semiconductor with a bandgap energy of 1.12 ev. In the presence of lattice phonons, photons of energy portions larger than the bandgap energy can be absorbed and thus contribute to the photo-generated current. Excess energy of photons is lost due to rapid thermalization of charge carriers to their respective band edges. Photons of lower energy than the bandgap are not absorbed. These two effects fundamentally limit the conversion efficiency of single-junction devices made from silicon to around 44% [30]. Further influences that limit the efficiency of solar cells will be discussed in this chapter and can be sorted in either one of the following categories: Optical losses Recombinative losses Resistive losses Cell parameters and IV characterization For the characterization of the finished device the cell parameters are extracted from current-voltage (IV) curves, measured under dark or illuminated conditions. During an illuminated IV measurement a voltage ramp is applied in forward bias direction to change the voltage at the cell terminals stepwise, while the generated current is measured for each step. More details on IV measurements are reported in Chapter 6. From this IV curve the cell parameters that constitute the power conversion efficiency (η) are extracted. The most important parameters are the fill factor (FF), the open circuit voltage (V OC ) and the short circuit current density (J SC ). The power conversion efficiency η is defined as the ratio between the maximum electrical power density p el and the intensity of the incident light I L. It is measured according to Standard Testing Conditions which are described in DIN EN Herein the spectrum of incident light (AM 1.5G) as well as an incident intensity of 1000 W/m 2 at 25 C are specified. η = p el,max I L = V OC j sc FF I L (2.2) Fig. 2.2 gives a graphical representation of the main cell parameters. Instead of absolute current and incident power usually the current density and illumination intensity is stated to yield comparable numbers independent of the cell geometry. The short circuit current density is the maximum current density that can be extracted from the cell under 1 sun illumination. The shape of that curve is established by Shockley s ideal diode law, wherein J ph is the photo-generated current density, J 01 is the saturation current density, k B is the Boltzmann constant and T the temperature in Kelvin. By convention the signs are chosen so that the power generating points of the curve lie in the first quadrant.

22 14 Chapter 2: Manufacturing technology and relevant concepts I S C 1 0 C u rre n t [A ] a n d P o w e r [W ] F F = Illu m in a te d I-V c u rv e P o w e r c u rv e P M P P I S C V O C M P P V o lta g e [V ] V O C Figure 2.2: Typical IV characteristic and power curve with indication of main cell parameters ] J(V) = J ph J 01 [e qv/kbt 1 (2.3) The efficiency is determined at the point of maximum power generation, indicated by the maximum of the product of current and voltage along this curve. The ratio of this maximum power point (MPP) and the product of J SC and V OC is called the fill factor: FF = J MPP V MPP J SC V OC (2.4) For an ideal diode the maximum FF can be approximated according to Green [31] from the normalized voltage v oc = V OC /(k b T/q) by: FF 0 = v oc ln(v oc ) v oc + 1 The light sensitive diode is in reality afflicted with leakage currents through the junction, represented in the equivalent circuit by a parallel or shunt resistance R p. This is accounted for by the third term in Eq. 2.6 and thus reduces the externally measured current. The voltage is diminished by the influence of series resistances R s within the cell, which distort the shape of the curve. The series resistance, encountered by the current along its path, is constituted of the bulk resistance R bulk, the lateral resistance in the emitter or BSF layer R sheet, the contact resistance R contact and the resistance of the metal grid R line. The influence of typical series resistances (few mω) and shunt resistances (several kω) distorts the shape even more and reduces the FF to around 79% down from its ideal value of around FF 0 of 84%: (2.5) ] J(V) = J L J 01 [e q(v+jrs)/kt 1 V + JR s (2.6) R p The dark saturation current density J 01 is generally the sum of the recombination losses in the emitter and base regions according to:

23 2.2. Solar Cell Characterization and Definitions 15 J 01 = J 0e + J 0b (2.7) For the metallized cell J 0e and J 0b can be subdivided into contributions from the metallized (J 0e(met) ) and passivated (J 0e(pass) ) areas, weighted according to their area fractions F met as: J 0e = (1 F met ) J 0e(pass) + F met J 0e(met) (2.8) The contribution of an n-type base J 0b can be expressed after [31] by the hole diffusion coefficient D p, the intrinsic doping concentration n 2 i and the diffusion length L p as: J 0b = qd pn 2 i L p N D (2.9) The influence of finite cell dimensions modifies this term by a geometry factor F N which relates the bulk diffusion length L p to the dimensions of the cell, represented by the base thickness W and depends on the rear surface recombination velocity S rear. It is given by: F N = S rearcosh( W L p ) + D p L p sinh( W L p ) D L p cosh( W L p ) + S rear sinh( W L p ) (2.10) When no current is extracted from the illuminated cell, the open circuit voltage is measured between its terminals. In this case the external current is zero and the entire light generated current is consumed by recombination within the cell. For ideal diodes the V OC is found to be: V OC = k [ ] BT Jph q ln + 1 J 01 (2.11) This J 01 takes into account the radiative, Auger and SRH recombination. It was found, however, that deviations from this shape can apply whenever recombination in the space charge region (SCR) is present. Also a large variation in series resistance over the cell area can distort the IV curve. These influences are best described by an extended double exponential model analogue to a second diode with ideality factor 2 connected in parallel. According to Mcintosh [32], a dark saturation current density J 02 in the range of na/cm 2 yields the best description of the IV characteristic in this case. This J 02 determines the shape of the IV curve at low voltages and up to the maximum power point, while at open circuit voltage the influence of J 01 still dominates. Alternatively a parametrization with a single exponential function with variable local ideality factor can be used. In the thesis of McIntosh it is shown how the value of the local ideality in the low or high voltage regions of the IV curve can be used to distinguish between dominant non-srh recombination mechanisms. ] [ ] J(V) = J L J 01 [e q(v+jr s)/m 1 kt 1 J 02 e q(v+jr s)/m 2 kt 1 V + JR s (2.12) R p Aside from current-carrying IV measurements the SunsVoc technique can be used to measure a V OC over intensity curve of the finished device [33]. The cell voltage is

24 16 Chapter 2: Manufacturing technology and relevant concepts measured between a conductive chuck and contact pins on the front side, while the cell is kept at open circuit condition at all times. A slowly decaying light pulse from a photo flash is monitored by a reference cell close-by. When the actual J SC under 1 sun illumination is known, the I SC V OC curve can be constructed from the SunsVoc curve. Since no current is extracted from the cell during the measurement, the resulting I-V diagram is free of series resistance contributions and therefore easy to fit. The evaluation of the FF of this curve shows the influence of the shunt resistance R P and the junction quality in terms of J 02 contribution. It is hence called the pseudo fill factor (pff). Spectral response and quantum efficiency The spectral response of a solar cell can be defined as the ratio between short circuit current density J SC (λ) and the monochromatic light intensity I(λ) of the illumination. SR(λ) = J SC(λ) I(λ) (2.13) It is a wavelength-dependant quantity in units of A/W. A more comprehensive figure can be calculated when the current is expressed in units of elementary charges and the light intensity in number of incident photons. This quantity is called the external quantum efficiency (EQE) and represents the conversion efficiency of incident photons of a specific wavelength into externally detected electrons. EQE(λ) = J SC q hc λi(λ) = hc SR(λ) (2.14) qλ From this expression the internal quantum efficiency (IQE) can be calculated by taking into account the reflectivity of the front surface R(λ) and the metal covered area F met as: IQE(λ) = EQE(λ) (1 R(λ))(1 F met ) (2.15) This term therefore represents the conversion efficiency of photons inside the device. Measurement of the cell IQE is a vastly useful, non-destructive analysis tool since the different absorption lengths of light of different wavelengths allow to discriminate between various recombination effects. The cell IQE in the blue and green wavelength range carries information about emitter recombination and surface passivation. The IQE towards near infrared radiation instead carries information about bulk lifetime and rear surface recombination, as well as free-carrier absorption. From the IQE the attainable J SC of the cell can be calculated by integration over the spectral intensity of the AM1.5G spectrum up to the cut-off wavelength, which is around 1107 nm for silicon: Optical losses λ 2 J sc = q EQE(λ)I(λ)λdλ (2.16) hc λ 1 The maximum current generated in a silicon wafer of 160 µm thickness illuminated under the AM1.5G spectrum at 1 sun intensity is around 43.6 ma/cm 2. In reality this theoretical

25 2.2. Solar Cell Characterization and Definitions 17 maximum value is reduced by optical losses which can have several origins. The biggest factor for double side contacted cells is the shading due to the front metal contacts. In the case of 6.5% metal coverage, this reduces the maximum attainable cell current to 40.8 ma/cm 2. Another influence is the non-ideal reflectivity of the surface. Bare silicon reflects more than 30% of the light, and therefore surface texturing is required to reduce this figure. In case of monocrystalline silicon a random pyramid texture is created in a KOH/IPA bath, which reduces the reflectivity to around 11%. On top of that an antireflection coating (ARC) is deposited to enable even lower reflectivity due to destructive interference of waves, which are either reflected at the ARC surface or the wafer surface, respectively. By tuning the refractive index and thickness of the layer, the reflectivity is minimized for a specific wavelength. Typically it is adjusted for minimal reflection around 600 nm, where the solar photon flux peaks. Further optical losses occur due to parasitic absorption in the dielectric layers, free-carrier absorption in highly doped silicon and escape light due to insufficient light trapping. All of these effects limit the current generation currently to around 39.0 ma/cm 2 for a bifacial n-pert device Recombination losses and lifetime measurements Recombination refers to the annihilation of electron hole pairs, the reverse process of carrier generation. The average time for this reverse process in a population of excess carriers is called carrier lifetime. In solar cells high lifetime is required to fulfil the second task of each cell, which is separation of charges. In order to contribute to an external current, electron hole pairs need to be separated at the p-n junction. The minority charge carrier needs to reach the junction by diffusion. The mean distance that is covered in the silicon bulk is called the diffusion length, which is related to the lifetime via L p = D p τ (for holes). As a rule of thumb, the diffusion length needs to be at least 3 times the wafer thickness to allow for high collection probability. In the following we present several processes that limit carrier lifetime. A general relation between the carrier lifetime τ and the recombination rate U for a specific recombination process is: τ = n U (2.17) Two mechanisms are inherent to semiconductors and can not be suppressed technically. The first is radiative recombination, the directly opposite process of carrier generation which is therefore proportional to the carrier concentration product. The total recombination rate can be written with the coefficient for radiative recombination B(300K) = cm 3 /s [34] as: U rad = Bnp (2.18) Since silicon is an indirect bandgap semiconductor, radiative recombination plays a minor role in the balance of charge carriers. The second mechanism is the Auger recombination which is the reversed process of impact ionization. The recombination rate of this three particle process can be expressed by: U aug = C n n 2 p + C p np 2 (2.19)

26 18 Chapter 2: Manufacturing technology and relevant concepts The associated Auger lifetime (for n-type silicon) in low injection ( n N D ) or high injection conditions ( n N D ) is given by: τ aug,low = 1 C n N 2 D and τ aug,high = 1 (C n + C p ) p 2 (2.20) The auger coefficients take values of C n = cm 6 /s and C p = cm 6 /s as determined by Dziewior et al. [35]. A revised model has been suggested by Kerr et al. [36] The presence of imperfections in the crystal matrix gives rise to a different category of recombination. Impurities or dislocations result in discrete energy levels within the energy bandgap. The recombination of carriers due to these trap levels is greatly enhanced because they allow for recombination via a two-step process. The following description of defect related recombination is based on the description in the thesis of Kerr [36]. The first formal description was given by Shockley, Read and Hall [37, 38] after whom this mechanism is hence called (SRH). They described the recombination rate U SHR for a single defect level by: U SRH = np n 2 i τ p0 (n + n 1 ) + τ n0 (p + p 1 ) (2.21) Herein τ p0 and τ n0 are the fundamental lifetimes related to the probability to encounter a defect of the density N t and capture cross sections for holes σ p and electrons σ n, when moving at the thermal velocity v th. τ p0 = 1 σ p v th N t and τ n0 = 1 σ n v th N t (2.22) n 1 and p 1 express the contribution to the carrier densities due to the occupation of the trap level. n 1 = N C exp( E t E C kt ) and p 1 = N V exp( E C E G E t ) (2.23) kt Here N C and N V are the state densities at the band edges and E C and E V the energy levels of the band edges of conduction and valence band. E t is the defect energy level. Accordingly the SRH lifetime can be expressed by: τ SRH = τ n0(p 0 + p 1 + n) + τ p0 (n 0 + n 1 + n) n 0 + p 0 + n (2.24) From this term it can be shown that defect levels close to the center of the bandgap are the most recombination active. It should be noted that a variety of interstitial defects due to common metal impurities (Fe, Ti, Mo) shows higher capture cross sections for electrons than for holes [4]. This leads to lower impact of SRH recombination for a common degree of contamination in Cz n-type wafers and accordingly high diffusion length of holes in such wafers. For silicon it is usually the dominating recombination for low injection conditions, eventually overtaken by Auger recombination in the higher injection regimes. Analogous to the formalism for bulk defects the lifetime due to defects at the wafer surface can be described. The variables however, need to be adjusted to

27 2.2. Solar Cell Characterization and Definitions 19 express defect densities and recombination events averaged over surface area instead of volume. The surface recombination rate thus becomes analogue to 2.21: U S = n sp s n 2 i n s +n 1 S p0 + p s+p 1 (2.25) S n0 Now n s and p s are the densities of carriers at the surface while, S n0 and S p0 are the surface recombination velocities for electrons and holes according to: S p0 = σ p v th N ts and S n0 = σ n v th N ts (2.26) In reality the surface produces a whole distribution of defect levels within the bandgap D it (E) so the actual U S follows from integration over the whole bandgap. It can be concluded that two fundamental approaches exist for reducing the surface recombination rate. On the one hand the density of interface states D it (E) should be reduced (principle of chemical passivation), while on the other hand the surface concentration of either type of charge carrier should be reduced (field effect passivation) (refer 2.1.4). Lifetime measurement techniques In the cell or wafer the lifetime is influenced by all of the mechanisms presented above. The resulting lifetime is called effective lifetime. Since the recombination mechanisms occur independently the effective recombination rate is formed as the sum of the individual rates. U e f f ective = U rad + U Aug + U SRH + U emitter + U S (2.27) The effective lifetime then calculates as: 1 = (2.28) τ e f f ective τ rad τ Aug τ SRH τ emitter τ S = 1 τ bulk + 1 τ emitter + 1 τ S (2.29) Effective lifetime is also the only quantity that is accessible experimentally. Most often we are also interested in the contributions of individual recombination processes. A key technique to distinguish between those, is to measure the effective lifetime while knowing as much as possible about the other recombination mechanisms. This can be done either by suppressing a recombination path, e.g. by taking high bulk lifetime material or by calculating and subtracting known contributions, e.g. the Auger recombination. Since the efficiency of the solar cell is determined by different limiting influences, lifetime measurements are an important tool in solar cell characterization. In order to evaluate the efficiency potential of the finished device we measure the effective lifetime on cell precursors, which are fully processed wafers without any metallization. Also dedicated symmetric lifetime structures (p + np + or n + nn + ) are produced to separate out the contributions from the different surfaces. Since the recombination properties at the surface or in the bulk generally depend on the injection level a technique should be able to determine the effective lifetime over a wide range of injection levels. This is accomplished

28 20 Chapter 2: Manufacturing technology and relevant concepts in the quasi-steady state photoconductance (QSSPC) method. Also the spatial resolution of the effective lifetime can reveal information about the origin of the recombination. We therefore employ the microwave detected photoconductance decay (µwpcd) technique which has a high spatial resolution. This method however works with pulsed laser radiation resulting in arbitrary intensity levels, leaving the exact injection level unknown. The techniques are therefore complementing each other. Both are applied throughout this work and will be briefly discussed in the following section, while more extensive explanation can be found in the literature [39 41]. Quasi-steady state photoconductance The most widely used instrument for lifetime testing in research and development is the quasi-steady state photoconductance (QSSPC) tool from Sinton Instruments. A photo flash is employed to generate carriers in a semiconductor sample, which is inductively coupled to an RC circuit. A sketch of the setup is seen in Fig The photo flash exhibits an exponential decay and its intensity is constantly monitored by a reference cell. Typically the flash decay time is much longer than the measured lifetimes, so that the excess carrier distribution can be considered in steady-state condition at all times. The excess conductance is measured by a coil underneath the sample. The carriers in the sample couple to the electromagnetic field in the vicinity of the coil, leading to eddy currents which induce a current of opposing direction in the coil. The voltage in the RC circuit is then directly linearly related to the conductance, which makes the calibration of the device using reference wafers of known conductivity straight forward. The measurement of the conductance is averaged over the area of the coil, which results in a measurement spot of 2 cm in diameter. This measurement of the excess photoconductance σ L is related to the average excess carrier density via: W σ L = q ( nµ n + pµ p )dx (2.30) 0 q n av (µ n + µ p )W (2.31) This term is valid for homogeneous carrier distributions throughout the sample and equal generation of holes and electrons ( n = p) which requires that this balance is not disturbed by trap states. The evaluation of the excess carrier density therefore requires knowledge of the sample thickness W, which can be easily measured and the carrier mobilities which are well known from literature [42]. In steady state condition the photo-generated current J ph is always balanced by the recombination current J rec. The latter can be calculated as: Combining 2.30 and 2.32 returns: J rec = q W 0 n dx τ e f f = q n avw τ e f f (2.32) τ e f f = σ L J ph (µ n + µ p ) (2.33)

29 2.2. Solar Cell Characterization and Definitions 21 Figure 2.3: Drawing of QSSPC measurement setup from [40] The measurement of excess photoconductance is therefore a direct way to measure τ e f f, given the photo-generated current in the sample is known. Therefore the current of the calibrated reference cell is measured and the result is converted into the average photo-generated current J ph in the test sample by applying an optical factor. This factor relates the optical properties of the reference cell to the measured sample in order to correct for different surfaces. From the excess carrier density an implied voltage can be calculated according to: V OC = kt q ln[( p(n D + n)/n 2 i ) + 1] (2.34) This implied V OC can be evaluated at all intensity levels. When evaluated at the intensity equivalent of 1 sun illumination, it can be a meaningful figure of merit. Implied V OC takes into account all the recombination effects in the bulk and at the surfaces as well as base doping, which is why it is more suitable to compare the performance of cell precursors than the effective lifetime as such. Since the implied V OC is determined on cell precursors without any metal it represents an upper limit to the cell V OC, without metallization induced losses. Another application of photoconductance measurements is the determination of emitter saturation currents J 0E for the characterization of highly doped surfaces. Kane and Swanson have suggested the method to be presented here [43]. The measured effective lifetime can be separated into the contributions coming from the bulk and the two surfaces via: 1 = 1 τ e f f τ bulk + 2 J 0 qn 2 n (2.35) i W This is valid for two identically processed surfaces as in symmetrically diffused and passivated test structures. For this method homogeneous carrier generation and distribution throughout the wafer is required. This can be attained by near infrared excitation which shows sufficiently high absorption length. For samples with passivated surfaces

30 22 Chapter 2: Manufacturing technology and relevant concepts and a bulk diffusion length that is long compared to the wafer thickness, a homogeneous distribution of minority carriers can be assumed. Under high injection conditions ( n N D ) the bulk lifetime is determined by intrinsic recombination channels which can be described formally. This allows to separate the bulk and surface recombination 1 even though they can be of the same magnitude. The plot of τ e f f 1 τ bulk over n allows to extract the emitter recombination current density from the slope of a linear fit. J 0e has been found to be a valuable figure to express all recombination effects within the emitter volume and at the surface in a single number. Microwave detected photoconductance decay The microwave detected photoconductance decay method (µwpcd) probes the conductance of a samples by measuring the transient of the microwave reflectance upon pulsed excitation. The carriers are excited by a short laser pulse and an antenna over the sample detects the change in the reflectance signal. The reflectance is not necessarily linear with the conductance, but depends on the geometry of the setup. This signal exhibits a mono-exponential decay which can be interpreted as an effective lifetime. Additionally the injection level is undefined which makes it difficult to interpret the lifetime data. The benefit lies in the high spatial resolution which enables lifetime mapping (examples are shown in Section 3.2). Only recently improvements to the existing setup were presented which enable measurements at defined injection levels. A steady state bias light is employed to specify different background injection levels, while the laser pulse probes the conductance. These quasi-steady state microwave detected photoconductance decay measurements (QSS µpcd) have been shown to correlate well to effective lifetimes as measured with QSSPC [44, 45] Resistive losses and resistance measurements The difference between ideal fill factor FF 0 and the real FF is governed by series and shunt resistance losses, as well as second diode recombination. The influence of the latter two at MPP can be extracted using the difference between FF 0 and the pff as determined by the SunsVoc technique. For the bifacial n-type cell this gives a gap of 84%(FF 0 ) 83%(pFF) = 1% abs. The remaining 83%(pFF) 79%(FF) = 4% power loss can mainly be explained by series resistances in the cell. It is therefore important to understand the impact of the different series resistance contributions. Most influences, namely sheet resistance, contact- and line resistance can be measured using a 4 pointprobe setup. This is achieved by a 4 probe design, wherein the contacts that feed a test current to the sample are separated from the voltage probes that actually measure the voltage drop in that sample due to that test current. The 4 probe connection ensures that the measurement is free of the resistance of feed cables. Contact- and sheet resistance are determined according to the transfer length method (TLM) [46].

31 2.2. Solar Cell Characterization and Definitions Wafer characterization Electrochemical Capacitance Voltage (ECV) measurements The ECV method was used extensively during this study and shall be presented in some detail. It is employed for accurate depth profiling of active dopant concentrations in the wafer. The technique consists of an alternating sequence of capacitance-voltage measurements and electrochemical etching steps. One advantage of the method over SIMS measurements is the simple sample preparation which consists only of a short HF-dip for oxide removal. The sample to be measured is placed in contact with an NH 5 F 2 electrolyte (0.1M) and contacted by metal pins on the same surface. (refer Fig. 2.4). Due to free charge carriers a reverse-biased electrolyte/silicon interface behaves like a Schottky-diode. The depletion capacitance of the interface can be expressed as: C dep = A dq dv (2.36) where dq is the charge variation for the voltage step dv and A is the measurement area. The thickness of the layer depleted of carriers (depletion width W d ) extends into the silicon by: W d = 2(Φ V)ɛ 0 ɛ r qn (2.37) At the same time a depletion layer of only a few nanometers extends into the electrolyte solution, because of the high concentration of freely moving ions (Helmholtzdouble layer). The capacitance of this depletion region is determined for discrete steps of reverse bias voltages between an electrode in the electrolyte and the contacted piece of silicon. For each bias voltage an additional small AC (1-4 khz) signal is applied to vary the width of the depletion layer. From the oscillating current the depletion capac- Figure 2.4: Schematic of electrochemical cell from Ref. [47]

32 24 Chapter 2: Manufacturing technology and relevant concepts itance can be calculated for each bias voltage. This procedure yields the C-V plot for a given voltage range. From this plot the doping concentration can be calculated using the Mott-Schottky relation (for homogeneous doping and sharp borders of the depletion layer) N = 1 C 3 qɛ 0 ɛ r dc/dv (2.38) It should be noted that the depletion capacitance is connected in a circuit comprising the series resistance of electrolyte and cable connections and a parallel admittance due to leakage currents around the actual measurement area. The capacitance of the Helmholtz layer, which is connected in series, is large compared to the depletion layer capacitance and can therefore be neglected. In this procedure the active dopant concentration at the silicon surface can be measured. For depth profiling a second step, the electrochemical etching of silicon becomes important. Nowadays the same electrolyte is used to dissolve a silicon sample under forward bias. The choice of electrolyte (0.1M NH 5 F 2 is used for silicon) is very important to ensure the formation of a Schottky barrier and to yield equal etching rates for different doping concentrations. The dissolution current is measured and integrated over time to evaluate the actual etching depth according to Faraday s law. The dissolution of silicon requires holes, which are present in in p-type material, whereas for n-type material holes have to be provided by illuminating the surface during the etching step. x etch = M zfρa Idt (2.39) Here M represents the molecular weight, while ρ represents the density of the sample. F is Faraday s constant and A the etching area as defined by the sample holder. The variable z stands for the charge amount transferred per sample molecule in units of the elementary charge. The sequential repetition of probing and etching steps allows for accurate depth profiling. The etching area is obviously an important input parameter not only for the determination of the profiling depth 1 A, but also in the doping concentration calculation 1 A 2. The surface enlargement of a textured sample has to be accounted for by application of an area factor. The geometrical measurement area is defined by the sealing ring of the sample holder but for random pyramid textured samples this area is multiplied by a correction factor of 1.6. Limitations of the method arise from the accuracy in the area measurement. For long etching times the edge of the etching crater increases the surface area or the comparison of samples after laser diffusion is difficult due to the varying degree of surface deformation. Also the growth of hydrogen bubbles on the sample surface distorts the etching front. A pumping step to wet the surface is recommended after each etching interval.

33 3 Boron emitter diffusion and passivation The following chapter summarizes the optimizations of boron diffusions from a liquid BBr 3 source in a tube furnace. Other doping sources such as doping paste and spin-on precursors have been considered as well. The boron diffusion is required to form the emitter of the n-type cell and is therefore one of the most crucial steps in the solar cell process. A variety of properties needs to be established for industrially diffused high efficiency emitters. The goal of this optimization was to fulfill the following requirements: Confirm the stability of the silicon bulk lifetime after boron diffusion Achieve sufficiently homogeneous sheet resistances over the wafer and boat positions Achieve homogeneous thickness of the BSG layer for passivation purposes Reduce processing time In the first part of this chapter general properties of boron diffusions will be reviewed. Alternative sources to BBr 3 based diffusion such as BCl 3, boron doping paste and a spin-on precursor are also presented. In the second part of this chapter the properties of passivation stacks for p + surfaces will be reviewed. A recently proposed method to enable p + passivation will be investigated and compared to state-of-the-art passivation stacks. One of the reasons for the predominant position of p-type cells in production was that no affordable solution for the passivation of p + surfaces was available. The single best solution for many years was thermal oxidation of the emitter surfaces, which involves additional costly tools and a high temperature step. In the last years the Al 2 O 3 based passivation has received much attention and is considered today as the first choice for industrial boron emitter passivation, even though it is still very costly. The BSG based passivation is therefore suggested to enable a cheap and commercially applicable alternative. 3.1 Diffusion of boron in silicon There are many boron diffusion sources available with a long research record. For this work we rely mostly on boron diffusion based on a boron tribromide (BBr 3 ) source. Consequently the process description is focussed around this diffusion method. Other potential precursors such as boron chloride (BCl 3 ), boron nitride (BN), diborane (B 2 H 6 ) or boron doped oxides can be considered and the same diffusion mechanisms apply. In the diffusion furnace the wafers are positioned standing upright in a quartz boat 25

34 26 Chapter 3: Boron emitter diffusion and passivation Figure 3.1: Schematic of diffusion tube with indication of wafer positions, gas-, center- and load zone (GZ, CZ, LZ). (compare Fig. 3.1) with the side to be diffused facing the gas inlet side. A spacing of 4 mm between the wafers enables diffusion of up to 200 wafers at the same time. In the first step (deposition) the N 2 carrier gas is injected into a bottle of liquid BBr 3 to transport the BBr 3 gas into the diffusion tube. Additionally O 2 is injected into the tube and immediately reacts with BBr 3 upon mixing according to (3.1). 4 BBr O 2 2 B 2 O Br 2 (3.1) Since the vapour pressure of boron oxide (B 2 O 3 ) is very small (only 10 8 atm at 920 C) the liquid condenses onto the wafer surfaces and furnace walls at common process temperatures. In order to ensure a homogeneous diffusion over the wafer surface and along the length of the diffusion tube the formation boron oxide needs to be precisely controlled. On the wafer surface elemental boron is formed by the reduction of boron oxide following reaction (3.2). Also the Si surface gets oxidized according to (3.3) This applies similarly for spin-on or APCVD deposited glass sources. 2 B 2 O Si 4 B + 3 SiO 2 (3.2) Si + O 2 SiO 2 (3.3) The silicon oxide is partly dissolved in the boron oxide to form the boron silicate glass system (BSG). The growth rate of the glass layer is determined by the supply of oxygen to the BSG/Si interface either by diffusion of B 2 O 3 through the BSG (3.2) or by diffusion of O 2 through BSG (3.3). For high B 2 O 3 content in the glass Arai et al. [48] have demonstrated that the diffusion of O 2 in glass and thereby the growth rate of the BSG is greatly enhanced. Elemental boron diffuses into the silicon wafer but also distributes within to oxide layer. Since the boron is produced at the BSG/Si interface an accumulation of boron can favour the formation of an SiB compound via the reaction (3.4). The complex crystal structure of SiB x has been analysed by several authors [49 51] while x has been determined to be in the range of 5-6. This compound is known to be insoluble in hydrofluoric acid and can induce lifetime degrading defects if developed to a thickness greater than 5-10 nm [52]. This so called boron rich layer (BRL) can easily be detected because it renders the surface hydrophilic even after glass removal. Si + xb SiB x (3.4) Arai et al. [48] have described the conditions for BRL formation as follows. Whenever the production rate of elemental boron at the interface exceeds the boron diffusion

35 3.2. Bulk lifetimes of silicon wafers after high temperature processing 27 current J(t) = N 0 DB in Si /πt into the wafer, boron can accumulate at the interface and form a BRL. The production rate is governed by the supply of B 2 O 3 to the interface. Diffusion under oxygen atmosphere has been shown to suppress the growth of BRL by oxidation of excess boron and dissolution of existing SiB complexes according to (3.5). With increasing oxygen concentration the oxidation of the silicon surface however leads to an overall reduction in the amount of incorporated boron. 2 SiB O 2 2 SiO B 2 O 3 (3.5) Pignatel et al. [50] have demonstrated that even oxidation during cool-down can dissolve the BRL while maintaining a high boron doping concentration in the silicon. As long as a BRL is present it serves as an infinite doping source and the amount of boron in the wafer only depends on the diffusivity of boron in silicon. In this case the surface concentration can exceed the solid solubility limit [49]. When the BRL is dissolved the surface concentration will be reduced to the solid solubility limit or even below and lead to a surface depletion. Many authors [49, 53, 54] have also reported an anomalous diffusion mechanism of boron in the presence of a BRL. It has been shown that the amount of boron diffused into a wafer decreases and accordingly the sheet resistance increases with increasing BRL thickness. It is suggested that the boron diffusion is enhanced by the presence of silicon interstitials. Kurachi et al. have now suggested that the compressive stress resulting from a BRL can increase the silicon interstitial concentration to a level where the affected layer becomes effectively a diffusion barrier for boron atoms. In this case it is also likely to observe lifetime degradation after boron diffusion as reported by different groups [52, 54, 55]. 3.2 Bulk lifetimes of silicon wafers after high temperature processing It is a common interpretation that boron diffusion from glass sources is associated with bulk lifetime degradation. For example Cousins et al. [55] have reported on diffusioninduced misfit dislocations as a potential mechanism for bulk lifetime degradation. Other researchers [52] describe a mechanism due to the thermal expansion coefficient mismatch between the SiB compound and silicon as the root cause of extended dislocation defects which can migrate easily during high temperature processing. On the other side the BRL can be used as a gettering layer for wafers with high concentrations of metal impurities. It can be formed to exhibit a gettering effectiveness in the range of phosphorous diffusions [9, 56 58]. It is assumed that in a mechanism analogous to gettering of a phosphorous dead layer the boron precipitates act as a sink for impurity atoms. In this case special care has to be taken when dissolving the BRL in a post-oxidation step. This can release the trapped impurities and accordingly degrade the bulk lifetime. Alternatively the BRL can be removed by a chemical etching step using a mixture of nitric acid (HNO 3 ), glacial acetic acid (CH 3 COOH) and HF, which drastically improves the passivation quality [59]. Throughout this work we have carefully avoided the formation of BRL during BBr 3 diffusion by adjusting the gas composition during the deposition phase and by in-situ oxidation. The material quality has been shown to be high enough so that

36 28 Chapter 3: Boron emitter diffusion and passivation extensive gettering is not needed and that a BRL can only be associated with lifetime degrading effects. The presence of BRL is routinely checked by removing the BSG after diffusion and observing the wetting behaviour of the diffused wafers. For the formation of emitters or back surface fields by means of diffusion the influence of high temperature processes on the bulk lifetime of Cz-Si wafers has to be understood. In this section we demonstrate some examples of high temperature processes, such as boron diffusion, that actually improve the bulk properties. Alongside with boron diffusion we compared the effects of pure nitrogen annealing and phosphorous diffusion in the relevant temperature regime (850 C 960 C). Lifetime samples of neighbouring n-type and p-type wafers from ingots grown at BOSCH Solar Energy AG were employed. In a second experiment we investigated the influence of the wafer origin within the ingot and compared two Cz-ingots from different feedstock material side by side, using the same process conditions as before. The results reported in this section have been presented at the 25 th EUPVSEC in Valencia (7) Experiment In a first experiment neighbouring 6 inch wafers were selected from an n-type ingot and from a p-type ingot grown under industrial conditions at BOSCH Solar Energy AG. The as-cut base resistivity was determined to be in the range of 2.5 Ωcm for n-type wafers and 8.5 Ωcm for wafers of the p-type ingot. These values were measured before any processing to include the influence of thermal donors. All wafers were saw damage etched and were subsequently divided into three groups. The first one was the reference group and did not get any high temperature treatment. The second and third group were submitted to either boron diffusion or annealing under nitrogen atmosphere respectively. Process temperatures in the industrial type quartz tube furnace were set to 850 C, 900 C, 930 C and 960 C. Afterwards the wafers have undergone another etching sequence of HF and NaOH in order to remove both, the oxides and diffused layers. Hydrogenated silicon nitride (SiN x ) layers were deposited by PECVD on both sides of the wafers as a passivation layer followed by a firing step. Figure 3.2 shows the process sequences for both groups respectively. Figure 3.2: Process scheme for the preparation of lifetime samples from as-cut wafers.

37 3.2. Bulk lifetimes of silicon wafers after high temperature processing Results and Discussion n-type and p-type comparison The effective bulk lifetime was analysed using µw PCD mapping and QSSPC measurements in the center of each wafer. Fig. 3.3 illustrates the lifetime results. For simplicity reasons only the results from the center zone are shown. The lifetime improvement over the whole area becomes obvious for the n-type wafer when looking at the lifetime maps. The p-type wafers on the other hand tend to be negatively affected with increasing process temperature. We see a correlation of the lifetime behaviour with the change in the resistivity as it is also predicted by the SRH theory for low injection levels. From the lateral patterns in Fig. 3.3 we deduct that effects of the simple passivation method used here come into play and might have limited the lifetimes. The lifetime of p-type minority charge carriers are already initially lower, but especially on the processed samples. The scale of Fig.3.4 has been adjusted to yield comparable colour scale to the n-type samples in Fig. 3.3 since the electron mobility in the p-type bulk is a factor 3 higher than the hole mobility in the n-type bulk. Even if we compensate for the difference in the diffusion coefficient of the electrons and holes, significantly higher diffusion lengths of minority charge carriers are measured on n-type wafers after the high temperature process, as shown in Fig These results indicate that boron diffusion can be performed without degrading bulk lifetime in an industrial solar cell process, and it demonstrates high lifetime and high temperature Figure 3.3: µw PCD lifetime mapping of the n-type wafers. Reference wafer without diffusion (left) and wafers that received boron diffusion with 850 C, 900 C and 960 C (following left to right) Figure 3.4: µw PCD lifetime mapping of the p-type wafers. Reference wafer without diffusion (left) and wafers that received boron diffusion with 850 C, 900 C and 960 C (following left to right), scale adjusted to account for higher electron diffusivity

38 30 Chapter 3: Boron emitter diffusion and passivation B u lk life tim e (a v e ra g e ) [ s ] n o t r e a t m e n t N 2 d r i v e - i n, C n -ty p e, life tim e & d iffu s io n le n g th p -ty p e, life tim e & d iffu s io n le n g th B B r 3, C P O C l 3, C D iffu s io n le n g th (a v e ra g e ) [ m ] Figure 3.5: Average values of effective lifetime of the passivated wafer and corresponding diffusion lengths, deducted from µw PCD mapping of the n- type (circles) and p-type (squares) wafers that received an annealing or either BBr 3 or POCl 3 diffusion step and reference group. stability of n-type wafers as needed for a solar cell application. The substantial increase in lifetime on n-type substrates is observed for both, samples with boron diffusion and tempered samples, and can therefore not be attributed to an impurity gettering effect of boron doped layers. The results for all temperatures are summarized in Fig. 3.5 in form of effective lifetime values and effective diffusion lengths. Comparison of ingot quality Here we selected wafers from the top, middle and bottom region of two different phosphorous doped Cz-crystals. Both crystals were intentionally doped to attain a base resistivity of about 2.5 Ωcm. The difference between the two is the quality of their feedstock. One ingot was made from electronic grade feedstock (hereafter E1) while the second one was made from feedstock of comparable quality as feedstock commonly used in todays industry processes, called solar grade feedstock (hereafter S1). We were interested in a side by side comparison of the high temperature effects on lifetime samples given the different quality of the base material. The influence of the origin of the wafer within the ingot was also of interest. Groups of wafers from the different positions were processed under the same conditions. In Fig. 3.6 we give a set of representative results from both ingots and all high temperature processes. Here we limit ourselves to the the depiction of the middle region of both ingots but corresponding results have been found in samples from other parts of the respective ingot. The samples were prepared following the process sequence outlined above. As can be seen from Fig. 3.6 the ingots perform comparably well after these high temperature steps. What is remarkable is that even the production ingot S1 yielded high lifetimes after phosphorous diffusion at 850 C and also after boron diffusion at 960 C, which are common processing temperatures. The

39 3.2. Bulk lifetimes of silicon wafers after high temperature processing 31 Figure 3.6: µw PCD lifetime mapping of samples from the middle region of ingot E1 (up) and S1 ingot (down) that received high temperature processing with the lifetime maps of a reference wafer (no high temperature process). observations from the first part of the experiment are widely reproduced for the case of n-type base material which emphasizes our earlier findings. Again we suspect that the reason for the substantial increase in bulk lifetime is the increase of base resistivity. The top part of an ingot usually contains high amounts of incorporated oxygen. The large shift of the resistivity in the top part of the ingots E1 and S1 can be explained by annealing of oxygen related defects (thermal donors TD). B a s e re s is tiv ity, Q S S P C [ c m ] E 1 (b e fo re p ro c e s s ) E 1 (a fte r p ro c e s s ) S 1 (b e fo re p ro c e s s ) S 1 (a fte r p ro c e s s ) T D a n n e a l B o tto m M id d le T o p Figure 3.7: Shift of base resistivity for lifetime samples from the test group in the process sequence as measured by QSSPC in the center of the wafers.

40 32 Chapter 3: Boron emitter diffusion and passivation Crystal defects triggering lifetime degradation On few samples originating from the very top of the E1 crystal we observed a pattern of concentric rings of significantly reduced lifetime (down to 5 µs) after high temperature processing. A graphical representation of this effect is given in Fig Cells made from such material would consequentially yield very poor performance. Figure 3.8: µw PCD lifetime mapping of samples from top region of the A1 ingot in the initial state (left) and after nitrogen annealing at 850 C, 900 C and 960 C (following from left to right). We observed this phenomenon on similarly processed top-region Cz-wafers from different manufacturers. Therefore we conclude its great relevance for the suppliers of such material. According to several sources [60 64], we suspect a configuration of grown-in intrinsic defects and oxygen defect clusters at the core of the problem. It has been shown in this literature that certain constellations of crystal growth parameters can cause such characteristic ring shaped defects (OiSF-ring). These defects are known to act as a sink for impurity atoms during prolonged high temperature steps. Samples from the same region which underwent POCl 3 or BBr 3 diffusion in the same temperature range showed similar trends however the fully degraded state occurred at higher temperatures respectively. This underlines that impurity atoms gather at the grown in defect sites. The gettering effects of POCl 3 or BBr 3 diffusion compared to the nitrogen annealing might delay the full development of the degraded state in this case. The formed defect clusters are reported to be very stable, which is why the lifetime of the affected wafers cannot be recovered by any practical means Conclusion No degradation of the bulk lifetime was observed after boron diffusion or annealing in nitrogen on n-type Cz-Si wafers. Furthermore, an improvement of the bulk lifetime could be observed after high temperature processing as compared to the initial lifetime. This improvement is not attributed to a possible metal gettering effect of boron since neighbouring wafers that received only the tempering step in nitrogen ambience exhibit the same behaviour. For p-type Cz-Si wafers, a different behaviour was observed. High temperature steps have a moderate impact on the bulk lifetime regardless of the type of process performed. Nevertheless, the lifetime values are significantly lower than those measured on n-type Cz-Si wafers resulting in significantly higher diffusion lengths of minority charge carriers on n-type wafers after the high temperature process. Together with the results of the second experiment which widely reproduced the observations

41 3.3. Homogeneity of emitter diffusion and BSG layer 33 made for n-type wafers, the superiority of n-type Cz-substrates over p-type as a base material for this application with diffused boron emitters could be shown. The experiments have also shown that large area (241cm 2 ) wafers with high lifetime over a large ingot range could be grown from conventional PV industry feedstock. Higher purity is not required for achieving wafer qualities sufficient for processing of high efficiency solar cells. 3.3 Homogeneity of emitter diffusion and BSG layer The boron diffusion for emitter formation of n-type solar cells has to fulfil several requirements. The present section deals with the goals of tuning the sheet resistance and its homogeneity over the wafer and boat position. Factors that influence these properties and also the thickness of the BSG layer were investigated in a screening experiment employing a design-of-experiment (DOE) approach. The experiments were performed in an industrial tube furnace from Centrotherm, but since the absolute values may depend on the individual processing tool, results will vary for different studies. The observed trends however can mostly be explained from the principles of the gas phase diffusion and are thus transferable to different equipment as well. The factors of interest were chosen from the available tool input parameters and by taking into account the mechanisms of boron tube diffusions as outlined earlier (see Section 3.1). They include the following list of factors and their direct interactions in the value range stated: BBr 3 flow, sccm range O 2 flow, sccm range N 2 flow, slm range duration of BBr 3 flow duration of O 2 flow It is found that high flow rates of O 2 increase the sheet resistance, probably since the available elemental boron is oxidized, preventing it from contributing to the emitter doping. High flow rates of BBr 3 reduce the sheet resistance as expected, and so does a high flow of the carrier gas N 2. The effect of increasing BBr 3 flow is clearly positive, since the overall amount of boron available is increased. For excessively high flow rates this boron supply can promote the formation of a BRL. Therefore the BBr 3 flow has to be balanced with the O 2 flow to level out the amount of elemental boron. Accordingly it is observed experimentally that the ratio of these two inputs plays an important role in achieving low sheet resistance in the minimum amount of time. The effect of high carrier gas flow, which also reduces R sheet, is interpreted as to promote the reaction of O 2 and BBr 3 by mixing the precursor gases and distributing the reaction products within the tube. Especially for the center and load zone (CZ and LZ) of the tube this is a key factor. Also important for the sheet resistance in those zones is the overall duration of the precursor gas flow. Our tube is made in a single inlet/single exhaust design, which is a configuration found in many industrial furnaces. This implies that the precursors react

42 34 Chapter 3: Boron emitter diffusion and passivation S h e e t re s is ta n c e [ /s q.] in flu e n c e o f O flo w 2 in flu e n c e o f B B r 3 flo w lo w O a n d B B r h ig h flo w [a.u.] R s h e e t c e n te r z o n e S h e e t re s is ta n c e [ /s q.] lo w N h ig h 2 flo w [a.u.] Figure 3.9: Example of influence of gas flows on the sheet resistance and its homogeneity based on DOE model for center zone position, using default parameters. immediately upon mixing in the gas inlet volume of the tube and the reaction product B 2 O 3 is transported slowly to the other end of the tube in the N 2 flow. The B 2 O 3 however condenses on all surfaces due to its low vapour pressure, leaning the gas mixture while flowing towards the exhaust end of the tube. Accordingly, if the N 2 flow is to low, or the deposition time to short, there is not enough boron oxide to saturate the wafers situated at the exhaust end of the diffusion tube. In terms of homogeneity the picture looks different. To achieve homogeneous emitter doping, as determined by the standard deviation of the sheet resistance over the wafer, a high flow in BBr 3 and overall N 2 gas is required. Again the influence of these factors increases going from gas zone to load zone, due to the aforementioned reasons. A high O 2 flow however works against the goal of homogeneity. This reveals a conflict of goals, since on the one hand O 2 is needed to form B 2 O 3, but on the other hand high O 2 flow leads to inhomogeneities. This trade-off is possibly a down side of the specific furnace design. It is demonstrated by Miyoshi et al. [65] that a reactor design with multiple

43 3.3. Homogeneity of emitter diffusion and BSG layer 35 S h e e t re s is ta n c e [ /s q.] R s h e e t (fla t) R s h e e t (fla t) fit R s h e e t (te x tu re d ) D iffu s io n te m p e ra tu re [ o C ] Figure 3.10: Temperature dependance of the sheet resistance for baseline boron diffusion gas inlets along the quartz boat lead to significantly improved wafer to wafer and inwafer homogeneity. They have also reported that low deposition temperatures ( 900 C) promote homogeneity, which was also confirmed in our study. This might be due to the lower surface oxidation rate, which leaves a more homogeneous supply of elemental boron available for diffusion into the silicon. As a guideline for conventional furnaces the deposition has to be adjusted by choosing the BBr 3 flow high enough to provide sufficient boron to the tube. Then this supply has to be balanced with just sufficient O 2 gas to convert the entire BBr 3 precursor gas. Again the flow of BBr 3 has an upper limit by the possibility of BRL formation. This is in line with findings from Kessler et al. [52], and the flow of O 2 is restricted by promoting sheet resistance deviations. When this balance is found the level of sheet resistance can be tuned in a range required for solar cell processing by adjusting the time and temperature of the drive-in plateau. A set of acquired diffusion sheet resistances with this scheme is shown in Fig So far we have given guidelines taking into account only the optimization of sheet resistance and its standard deviation. For our special case the situation is more complicated since the BSG from the diffusion is utilized as a passivating oxide. This application additionally requires a minimum BSG thickness ( 15 nm) and a homogeneous distribution of its thickness. The BSG layer is applied as part of a passivation stack in combination with PECVD SiN x so that minor thickness variations ( 10 nm) become immediately apparent even for the naked eye. This puts additional constraints on the BBr 3 flow and time, both of which promote an uneven build up of the BSG layer. Other than reducing the BBr 3 flow, the only option to alleviate the effect is again to increase the flow of the carrier gas N 2. The carrier gas flow has a significant effect on the homogeneity in the center and load zone. The insufficient mixing and inhomogeneous distribution of the B 2 O 3 in the volume around the gas inlet [66] can adversely affect the homogeneity of the sheet resistance and BSG layer thickness. This reveals the shortcomings of a single inlet/single exhaust tube design. However a good balance of the factors was found as shown in Fig It should be noted that these recipes were tested with a wafer spacing

44 36 Chapter 3: Boron emitter diffusion and passivation Figure 3.11: Sheet resistance distribution over wafer surface and tube position for reference diffusion, measured with SHR mapping of 4 mm and loaded with up to 200 wafers. The diffusion outcome may change when diffusion is carried at double capacity (400 wafer) using less wafer spacing Conclusion Boron tube diffusion has been investigated for boron emitter formation. The method of pre-deposition and drive-in in a single high temperature step has been applied for a long time. The goals of sheet resistance homogeneity on one wafer and on a wafer-to-wafer basis can be achieved even with a conventional tube design. The application of BSG as a passivation layer adds further requirements for the BSG thickness and homogeneity, which complicates the process. We have found and evaluated the determining factors and could explain their influence on the monitored parameters. This enabled us to develop diffusion recipes that result in diffused layers of great homogeneity ( 5 Ω/sq.) and repeatability. The application of boron diffusion for solar cell production is often perceived to be problematic. This may be an unfair judgement since stable diffusion results with short process time and high lifetimes can be obtained when the parameters are properly controlled. For industrial diffusion processes however additional factors, such as maintenance schedules or high capacity behaviour come into play, that can not be embraced here due to the low throughput. 3.4 Passivation of highly boron doped surfaces Introduction Until today a variety of options exists to effectively passivate p + surfaces. However only few of them are industrially viable. Among other factors it has been the lack of passi-

45 3.4. Passivation of highly boron doped surfaces 37 vation methods for p + surfaces, simple and cheap enough to be implemented in high volume manufacturing, that has prevented the industrial adoption of n-type cell concepts. The capabilities of the respective passivation techniques are briefly reviewed and stand as a reference to evaluate the passivation quality achieved in this work for the bifacial n-type solar cell. A selection of passivation techniques suitable for p + passivation, that are available today is listed below: Silicon nitride deposition Thermal oxidation Wet chemical oxidation Aluminum oxide deposition (ALD or PECVD) Amorphous silicon (a-si) deposition BSG passivation (investigated here) Depending on the passivation mechanism (field effect and higher interface quality) there are different ways to evaluate the passivation quality. The best method for an application in solar cells is the measurement of the emitter saturation current density J 0E. One has to keep in mind however that this is lumped number which includes all the recombination effects that occur simultaneously in the highly doped regions and at the surface. It is often used to compare passivation quality of different dielectrics and between different research groups. In the latter case extra care has to be taken when comparing these values since the passivated boron profiles usually vary. Additionally the surface orientation plays a big role since (111) surfaces and (100) surfaces exhibit different passivation quality Overview of passivation methods Passivation by silicon nitride The PECVD deposition of silicon nitride is a simple low temperature passivation method for n + surfaces. It combines near ideal anti-reflection properties and excellent surface passivation, which makes it the most widely applied passivation technique for phosphorous emitters. The deposition of SiN x on highly doped p + -surfaces however was shown to lead to increased SRV and therefore higher J 0E values [67]. It is assumed from Schmidt et al. [68] that the reason behind this are defect types which show considerably higher cross sections for electron capture compared to the defects at Si SiO 2 interfaces. Also the amount of positive fixed charges within the SiN x is given as a common explanation for its poor passivation quality. Positive charges in the vicinity of a p + -doped surface induce an inversion layer by attracting electrons, thereby enhancing SRV. Altermatt et al. [69] have simulated the impact of the fixed charges on SRV and deduce that the fixed charges alone are unlikely to be the reason for the poor passivation. In the same line of arguments findings by Chen et al. [70] show that the passivation quality of PECVD deposited Si-rich SiN x layers on boron emitters can be improved significantly by annealing at 450 C. This could again be attributed to a reduction in the fixed charge density,

46 38 Chapter 3: Boron emitter diffusion and passivation the exact mechanisms however are yet under investigation. The deposition of a single layer of SiN x alone does not lead to a reasonable passivation, however the deposition as part of a passivation stack has many application already today. This is often required to enhance the anti-reflection properties and also to provide additional hydrogen. This is crucial to enable the passivation, because the hydrogen diffuses to dangling bonds and saturates interface states during contact firing. Passivation by thermal oxides Thermal oxidation of silicon at elevated temperatures ( C) is employed to form state-of-the-art passivation layers on lowly and highly doped silicon of either dopant type [71 73]. This is mainly attributed to a great reduction in the interface state density, but also to a field effect contribution from a small amount of positive fixed charges of about cm 2 [74]. During thermal oxidation the oxide grows into, rather than ontop of the surface which makes the interface quality rather independent of the initial surface quality. Annealing at 450 C under forming gas has been shown to improve the passivation even more, which is due to the saturation of interface states with hydrogen. The deposition of hydrogenated SiN x and subsequent firing steps shows a similar, yet less strong improvement of thermal oxide passivation. The data presented by King et al. [72] and complemented by Altermatt et al. [69] show the lowest reported J 0E values for highly doped p + -surfaces after thermal oxidation. It should be noted that the boron profiles in the cited studies have been acquired after extended drive-in heat treatments, and are further driven in during oxidation above 1000 C. This leads to an exceptionally low surface concentration which is not representative for industrial boron diffusion profiles. The samples have additionally received a forming gas anneal. The acquired values should be understood as the best passivation quality in the lab and are not representative for passivated industrial type emitters. Another important issue is the long term stability of such passivation layers, which can be a problem for oxidized samples as reported by various authors [16, 69, 75]. It has been shown that for deep diffusions ( 1µm) with extensive surface depletion due to the long oxidation the passivation quality can degrade significantly over a period of 2 years. This effect was shown to be reversible under forming gas anneal for samples oxidized with trichloroethane (TCA) [69] while Zhao et al. [16] demonstrated recovery of the passivation quality after corona charging, indicating an electrostatic origin of the degradation. Passivation by wet chemical oxides It is known from thermal oxides that already thin oxide layers in combination with hydrogenated silicon nitride can provide high passivation quality on p + and n + doped layers. Chemical oxidation of the wafer surface can also lead to dense oxide layers and is widely employed in the semiconductor industry as part of cleaning procedures. The oxide growth incorporates surface contaminants and in a subsequent step the entire oxide layer is removed in buffered HF. The attainable oxide thickness is very limited due to the limited diffusion of oxygen at low temperatures. The passivation effectiveness of such oxide layers has been investigated by Mihailetchi et al. [76] for p + layers. They demonstrated a J 0E as low as 23 f A/cm 2 on polished p + np + test structures with symmetrical

47 3.4. Passivation of highly boron doped surfaces Ω/sq. boron emitter. The ultra-thin passivating oxide of 1.5 nm thickness was grown by nitric acid oxidation of silicon (NAOS method) at room temperature prior to SiN x deposition. In the publication the stability of the passivation layer after contact firing and light soaking for 1000 hrs. has been demonstrated. This patented method was further improved at ECN [77] and is believed to be applied as boron emitter passivation layer of Yinglis Panda cell technology on a large scale. Passivation by aluminum oxide For many years silicon oxide based thin films were the only method to reliably passivate highly p-doped layers. Only recently the advent of aluminium oxide based passivation has attracted considerable interest of research institutes and manufacturers. Amorphous aluminium oxide (Al 2 O 3 ) can be deposited either by atomic layer deposition (ALD) techniques or also by means of PECVD, as was recently introduced [78]. Atomic layer deposition is a technique wherein the dielectric is grown layer by layer in a sequence of self-limiting surface reactions. The wafer is exposed to the precursor gases, commonly trimethyl-aluminium (TMA, Al(CH 3 ) 3 ) and H 2 O are used in thermal ALD reactors respectively. In between the half reactions, the reaction chamber is purged with a noble gas, such as argon. For each cycle a mono layer of Al 2 O 3 is grown, which results in unrivaled deposition homogeneity and thickness control [79]. Alternatively the oxidizing reaction can be achieved by pure oxygen or ozone gas, in which case the molecules are dissociated in a plasma, hence this technique is called plasma-assisted ALD instead of thermal ALD. The passivation quality has been shown to be better than that of thermal oxide with emitter saturation current densities as low as 30 f A/cm 2 for a 54 Ω/sq. boron emitter. Again this holds true for very deep diffusions, with relatively low surface concentrations. The outstanding passivation is owing to a high amount of negative fixed charges (up to 1e10 13 cm 2 ) in the Al 2 O 3 leading to a pronounced field effect, which shields the electrons from the surface, preventing their recombination. Low deposition rates, on the order of 1 nm/min and high costs for tools and consumables are the main drawbacks preventing the large scale employment of ALD techniques in the industry. Lately novel reactor designs (spatial ALD) have been developed to allow for up to 100 times higher deposition rates. This could enable high throughput manufacturing, as was suggested by Werner et al. [80]. Yet a different approach is the PECVD based deposition as presented by Saint-Cast et al. [81] and Sperlich et al. [78, 82]. They have shown that Al 2 O 3 layers of comparable quality in terms of homogeneity and passivation can also be produced by modified inline PECVD tools. For ALD deposited Al 2 O 3 layers and Al 2 O 3 /SiN x stacks the stability after 6 months storage, after contact firing and against UV radiation has been demonstrated by Dingemans et al. [83]. The stability of PECVD deposited layers however is still topic of various investigations. Passivation by amorphous silicon Amorphous silicon can also be deposited by means of PECVD at low temperatures. The achievable passivation quality lies in the range of the best thermal oxides [69]. This cumulates in exceptionally high implied V OC values as seen, for example, on heterojunction structures such as the HIT cell from Panasonic [84]. The high passivation quality

48 40 Chapter 3: Boron emitter diffusion and passivation of amorphous silicon layers comes at a high cost however. On the one hand the poor tolerance against high temperature processing of amorphous layers restricts the use of conventional firing-through metallization. On the other hand amorphous layers show a strong absorption in the UV and blue wavelength range, which produces a trade-off in the layer thickness between passivation quality and light transmission Experimental results Comparison of BSG and aluminium oxide based passivation for different sheet resistances During the boron diffusion the processes described in subsection 3.1 lead to the growth of the boron silicate glass system (BSG). This layer is a mixture of SiO 2, with partly dissolved B 2 O 3 and amounts of elemental boron. We have investigated the passivating properties of such layers for different boron emitters. Obviously the utilization of a boron oxide as a passivation layer can not be evaluated independently from the emitter profile. The profiles, shown in Fig. 3.12, were thus measured by ECV on polished wafers diffused together with the lifetime samples. The first set of diffusion profiles was prepared with identical BBr 3 deposition parameters and drive-in temperatures ranging from 920 C to 960 C for about 25 min. The glass layer is here between nm thick and has a refractive index of about 1.6. For all diffusions the formation of a BRL has been avoided by in-situ oxidation (compare 3.1). This was routinely checked by etching the BSG in 2% HF and confirming that the wafer surface turns hydrophobic. Since the BSG itself is highly hydroscopic it absorbs water quickly from the ambience atmosphere leading to to formation of boride crystals on the BSG layer. This can be avoided by capping the surface with a SiN x layer. We therefore deposit 65 nm of PECVD SiN x (refractive index n = 2.05) immediately after the diffusion, either directly onto the BSG or after a short HF dip used for BSG thinning. In the experimental data shown below we evaluate the passivation quality of the BSG/SiN x stack in comparison with PECVD deposition of an ] C a rrie r c o n c e n tra tio n [c m B o r o n 6 7 / s q. B o r o n 8 8 / s q. B o r o n / s q. B o r o n 5 0 / s q D e p t h [ µ m ] Figure 3.12: ECV profiles of passivated boron samples.

49 3.4. Passivation of highly boron doped surfaces 41 Al 2 O 3 /SiN x stack. The latter process was performed in an industrial deposition tool at BOSCH Solar Energy. Initially the wafers were textured and afterwards cleaned with the IMEC sequence. Different boron emitters were then diffused onto the double-side textured lifetime samples. This represents the actual surface configuration of the cell. After boron diffusion the wafers either received immediately the SiN x capping layer or the BSG was removed in HF and 10 nm of PECVD Al 2 O nm of SiN x were deposited. Both groups underwent a simulated firing step at 830 C peak firing temperature. The lifetimes are measured with the QSSPC setup and implied V OC values are extracted at 1 sun intensity. The J 0E values are determined using the Kane and Swanson method [43]. Since BBr 3 diffusion does not yield fully symmetric samples, care has to be taken when comparing the values with the literature. The surface facing the exhaust side of the tube is usually more lightly doped and the calculated implied V OC can therefore be overestimated. The recorded ECV profiles are measured on diffused samples after the BSG layer had been removed by HF etching. Any other different passivation method involves removal of the BSG followed by additional cleaning steps prior to deposition. This might involve chemical oxidation and etching of the diffused layers (IMEC cleaning) and therefore changes in the surface concentration. The passivated profiles are not necessarily identical, even if they were diffused simultaneously. This is especially true for thermal oxidation. The oxidation consumes the silicon surface, so that the oxide grows on top, as well as into the wafer. For bare silicon 2.17 units of SiO 2 are grown for each unit of silicon. Due to higher solubility in the glass compared to silicon, the boron accumulates in the SiO 2, rather than in the top silicon layer, leading to further surface depletion. The measured implied V OC and emitter saturation currents J 0e are shown in Fig In this range of diffusions the BSG passivation, does not reach the same passivation quality as PECVD-based Al 2 O 3 passivation stacks. The passivation mechanism of Al 2 O 3 is known to be based on a field effect due to a high amount of negative fixed charges. The BSG layer, in comparison, is a mixed phase with an SiO 2 layer growing at the B 2 O 3 -Si interface. The interface oxide is therefore comparable to a thermally grown oxide at the diffusion temperatures, e.g. usually exceeding 900 C. The diffusion reaction predicts that it contains fractions of dissolved B 2 O 3 and is doped with elemental boron. The interface charge of thermal oxide is very low, and even of positive sign ( cm 2 ), but the interface quality is very high (low D it ). The passivation mechanism of BSG is therefore assumed to be comparable to pure thermal oxide. It is found in this study that the emitter surfaces are strongly depleted, indicating a limited boron supply during the diffusion. The significantly lower level of passivation for the BSG stack can maybe partly be explained by the strong surface depletion. The depletion of dopant atoms results in an electric field component which drives minority charge carriers towards the recombination active interface. A field effect passivation, such as from Al 2 O 3 however, repels minority charge carriers and therefore compensates the field component towards the surface. The minority carriers are sufficiently shielded from the interface, resulting in superior passivation of Al 2 O 3 for depleted boron emitters.

50 42 Chapter 3: Boron emitter diffusion and passivation i m p. V O C [ m V ] B S G /S in A lo x /S in J 0 E [ f A / c m ² ] B S G /S in A lo x /S in S h e e t re s is ta n c e [ /s q.] Figure 3.13: Passivation quality of BSG/SiN x stack in comparison with all- PECVD AlO x /SiN x stack. Comparison of different passivation methods for reference boron emitter For emitter diffusions showing less depletion an additional comparison of all available passivation stacks was made. Aside from the two previously mentioned double layers Al 2 O 3 /SiN x and BSG/SiN x, a wet thermal oxidation at 815 C for 20 min, and nitric acid oxidation as passivation also capped with PECVD SiN x, was applied. Only the 70 Ω/sq. reference diffusion, as shown in Fig was used here. This comparison was done at a later point in time and the reference diffusion profile had thus improved. The optimized BBr 3 diffusion, which is the baseline recipe as of today, shows a factor 2-3 higher surface concentration, and less surface depletion. The layer slightly depleted of boron extends only up to 50 nm into the silicon whereas the depletion in the earlier comparison was much more pronounced and the peak concentrations were reached only at 100 nm into the silicon. The data from Fig is also stated in tabulated form in Table 3.1. The passivation quality of the BSG/SiN x stack is found to be on the same level as wet thermal oxidation and modern aluminium oxide based passivation. The passivation is demonstrated to be suitable for (100) as well as (111) surfaces, whereas the application on (111) surfaces, as required on a textured solar cell emitter leads to slightly enhanced recombination, compared to wet thermal oxidized (111) surfaces.

51 3.4. Passivation of highly boron doped surfaces 43 ] C a rrie r c o n c e n tra tio n [c m /s q. p ro file D e p th [ m ] Figure 3.14: Std. solar cell emitter profile of 70 Ω/sq. i m p. V O C [ m V ] im p.v O C J 0 E A lo x W e to x B S G N A O S J 0 E [ f A / c m ² ] Figure 3.15: Passivation quality of different stacks on textured symmetric samples It is found that the passivation of the BSG layer alone is initially poor. Implied V OC values of only mv have been measured on textured, symmetric lifetime samples. It is found that the SiN x deposition and a firing step is required to achieve a high level of surface passivation, while firing of BSG alone does not improve the passivation significantly. This is further investigated in the analysis of cells fired at different peak temperatures. The IQE data for cells passivated with the BSG/SiN x stack is demonstrated in Fig As can be seen in the graph, the blue response of these cells, which is directly related to the passivation quality, increases for increasing firing temperature. It is concluded that the passivation is improved by saturation of interface dangling bonds, by hydrogen diffusion from the SiN x layer. The tempering step is therefore crucial to activate this passivation. Despite many reports about the long term stability of oxide passivation on p + silicon, no degradation of stored samples over time has been observed. Even under accelerated LID test conditions no degradation of the passivation

52 44 Chapter 3: Boron emitter diffusion and passivation Passivation method Surface implied V OC J 0e per side [mv] [fa/cm 2 ] Al 2 O 3 /SiN x textured ± ± 4 thermal SiO 2 /SiN x textured ± ± 6 thermal SiO 2 /SiN x polished ± ± 3 BSG/SiN x textured ± ± 7 BSG/SiN x polished ± ± 3 NAOS/SiN x textured ± ± 7 Table 3.1: Passivation benchmark for symmetric 70 Ω/sq. boron emitter (p + np + ) samples passivated by different methods. quality occurred. The BSG passivation was quickly adopted for the solar cell process and in the following a comparison of this new method against Al 2 O 3 based passivation on cell level is shown. Aluminium oxide passivation was applied at BOSCH Solar Energy AG using state-of-the-art production equipment. All cells were fabricated together until after the boron diffusion, when the BSG group received immediate PECVD SiN x capping, whereas the aluminium oxide group was shipped, with the BSG as protection layer. At the site of the project partner the BSG was stripped and the samples re-passivated. The samples were finalized simultaneously at ISC-Konstanz, with screen-printing of metal pastes and contact firing. The cell data confirms the findings from lifetime samples in that a comparable efficiency is achieved for both passivation stacks. Although the average efficiency is even higher for the new BSG passivation, this is largely attributed to a difference in the FF. The same firing procedure without optimization was applied for both groups. A firing optimization for the aluminium oxide group might lead to improved FF. Also shipping and handling can have lead to increased variation in the cell IQ E [% ] I Q E r e f l. d a t a T p e a k = C T p e a k = C T p e a k = C W a v e l e n g t h [ n m ] Figure 3.16: IQE and reflection data of cells fired at different peak temperatures.

53 3.4. Passivation of highly boron doped surfaces 45 Passivation J SC V OC FF η method [ma/cm 2 ] [mv] [%] [%] Al 2 O 3 /SiN x average of ± ± ± ±0.20 Al 2 O 3 /SiN x best cell BSG/SiN x average of ± ± ± ±0.10 BSG/SiN x best cell Table 3.2: IV data of cells passivated with the BSG method or PECVD Al 2 O 3 /SiN x stacks, averaged and best cell data, error is STD within group. V OC of the aluminium oxide group. It is further noted that Al 2 O 3 passivated cells can show a higher current (see best cell results), due to more homogeneous deposition of the anti-reflection layer for the industrial passivation stack. This comparison has proven the high potential of the BSG/SiN x passivation stack for solar cell application. Optimizing the BSG passivation The boron glass method yields already satisfactory passivation quality but the optical properties of the stack are not homogeneous enough over the wafer and over the positions in the quartz boat. This can be a major drawback for the adoption of this technique in an industrial application. Towards the edges of the cell the BSG layer grows thicker than in the center. Although this variation is in the range of nanometers it becomes visible to the naked eye, when the anti-reflection coating is applied. The variation in the thickness leads to changes in the reflectivity and thus to non homogeneous absorption. This might be the reason for the higher current observed for aluminium oxide passivated cells. The optical impression of an assembly of an array of cells side by side in a module is therefore also quite poor. This is why we tried to reduce the BSG thickness by a short HF etching step. It was found that the BSG is usually fully removed after 2 min etching in 2% HF, whereas after 1 min first spots on the wafer surface turn hydrophobic. In order to make sure that the entire surface is still sufficiently covered by oxide the etching time accordingly has to be shorter than 1 min. A comparison of cell batches with an additional BSG thinning step, before SiN x deposition and without was carried out. The test group was submitted to an etching step of precisely 20 s duration followed by DI-water rinsing. The wafers were processed according to the baseline sequence. During BBr 3 diffusion three boat positions (GZ, CZ, LZ) were loaded, since it is known that the BSG deposition and composition varies throughout the diffusion tube. The results of that experiment are demonstrated in Table 3.3. The differences observed here are significant. Enhanced light absorption (J SC ), improved junction (pff) and passivation quality (V OC ) is found for all cell groups after the short HF etching step. The increase in J SC can be explained by improved anti-reflection properties and reduced parasitic absorption in the glass layer. This partly explains also the enhanced V OC. It can be concluded that the short HF dip improves the passivation, although it is not fully understood why the pff and passivation are affected so strongly. It is assumed that the SiN x deposition on the freshly etched surface enables a better interface quality.

54 46 Chapter 3: Boron emitter diffusion and passivation Position HF dip J SC V OC FF η pff [ma/cm 2 ] [mv] [%] [%] [%] GZ no GZ 20 s CZ no CZ 20 s LZ no LZ 20 s Table 3.3: Impact of BSG thinning by 20 s glass etching, IV data averaged over 3-5 cells per group, uncertainties of IV measurement apply Conclusion In this chapter we have characterized a new and simple method to passivate highly boron doped surfaces. In many solar cell applications the depleted BSG layer in combination with a PECVD SiN x can passivate the surface as good as thermal oxide based passivation. The comparison of BSG passivation with other, state-of-the-art, passivation techniques on lifetime samples and cell structures has demonstrated that passivation quality on the same level can be achieved. This is remarkable since the BSG is usually treated as an unwanted by-product of the emitter diffusion. The simple deposition of PECVD SiN x, a ubiquitous processing step in solar cell fabrication, turns it into a valuable passivation layer. Thereby, not only the additional passivation step, but also one wet chemical processing step, usually required to remove this layer, can be saved compared to the standard solar cell process. It has been reported that also the PSG layers formed during phosphorous diffusion can be exploited in the same manner. Here even enhanced contact formation was observed [85].

55 4 Cell processes 4.1 Cell processes with homogeneous diffusions The processing steps, as outlined in Chapter 2, are applied to fabricate cells with a structure shown in the graphic below. For this cell architecture the term n-pert has been coined, a cell with a Passivated Emitter and Rear Totally diffused. This means that both surfaces are homogeneously diffused, in this case a boron emitter is diffused at the front side and a phosphorous BSF is found at the rear surface. All surfaces are passivated by either thermal oxide and PECVD SiN x stacks, or the BSG passivation method presented in Chapter 3.4. The contacts are formed by screen-printing and a cofiring step of commercial silver pastes. On the front side an aluminium containing silver paste has to be used to enable the contact to the p + surface. The homogeneously diffused back surface allows for the application of an open rear side grid, which makes the cell bifacial and thus enables a variety of new applications. The potential of this cell structure can be evaluated at different stages during the process. We evaluate the V OC potential of the cell precursor. These cell precursors are fired without printing of metal contacts. The measured implied V OC value captures all recombination effects in the bulk, diffused surface layers and at the interfaces. The next section shows the potential of this structure by virtue of PC1D device modelling and corresponding experimental results. The resulting data allows to estimate the benefits that can be expected when certain properties of the cell are optimized. However the necessity to form co-fired metal contacts imposes additional requirements on the cell surfaces. Especially the surface doping concentration needs to be high enough to facilitate contact formation. The surface geometry plays a big role in contact formation as well, as the sharp edges of random pyramid textured surfaces are preferred sites to initiate Ag crystallite growth. Polished surfaces promote significantly less crystallite growth and therefore show higher contact resistances for similar doping conditions. Fortunately the continuing development of contacting pastes allows for ever decreasing surface concentrations to be contacted. This enabled us also to implement a BSF diffusion with less rear side doping. The simulation approach is meaningful in this respect because further paste Figure 4.1: Cell structure of n-pert type cell 47

56 48 Chapter 4: Cell processes improvements and therefore the usability for higher sheet resistances can be expected for the future [5]. In the real cell process the sequential tube diffusions put constraints on the doping profiles that can be achieved. In case of the n-pert cell, this is especially valid for the phosphorous BSF. The investigations for the diffusion profiles are therefore not always referenced to the same baseline process as gradual improvements have been implemented over time. The respective baseline process at the time of experiment is in this work always specified Potential of cell precursors in PC1D device simulations The semiconductor device simulator PC1D [86] was used to perform device simulations of the bifacial cell concept. Here the motivation was to: Understand the bulk-lifetime requirements for the cell concept Demonstrate the influence of the emitter diffusion profile and of emitter passivation Demonstrate the influence of the BSF diffusion profile and of BSF passivation This model can be used to point out the direction of highest efficiency gain for independent process variations. The default model was built using the following device parameters, unless otherwise specified. All silicon properties such as band gap, optical coefficients or mobilities were kept at the default values [86]. The simulation thereby comprises all recombination effects in the bulk (Auger, SRH, radiative), in the diffused layers (Auger recombination and free-carrier-absorption) and at the interfaces (SRV). The circuit evaluation lacks accuracy because it cannot account for inhomogeneities in the sheet resistance or contact resistance which can have a detrimental influence on the cell voltage and fill factor. Anyhow series and parallel resistance are implemented in the simulation as a lumped number extracted from the best cell results. They are therefore small enough not to have any detrimental effect on the V OC or J SC. Parameter Value (Range) Source Device area 241 cm 2 Measurement Front surface pyramid height 11 µm Measurement Rear surface polished Internal reflection (both) 95 % Estimate Series resistance 3 mω Measurement Parallel resistance 10 kω Measurement Cell thickness 150 µm Measurement Bulk resistivity 3 Ωcm Measurement Bulk lifetime (n & p) 1.5 ms Estimate Surface recombination velocity (both) 5000 cm/s fit of IQE curves Diode ideality 1 Measurement Table 4.1: PC1D model input parameters

57 4.1. Cell processes with homogeneous diffusions V O C [ m V ] J S C [ m A / c m ² ] V O C J S C S R V = c m /s S R V = 0 c m /s B u lk life tim e [ s ] V O C [ m V ] J S C [ m A / c m ² ] V O C J S C fro n t S R V v a ria tio n re a r S R V v a ria tio n S R V [c m /s ] Figure 4.2: PC1D simulation for cell performance dependence of bulk lifetime and front or rear SRV. For the front SRV variation the rear SRV was fixed at 5000 cm/s and vice versa. Model was built using ECV doping profiles for 70 Ω/sq. boron emitter and 50 Ω/sq. phosphorous BSF. Values calculated for 6 % metal shading, dashed red lines indicate values for best fit to measured IQE data and stars indicate ideal case without surface recombination. Additionally we assumed 6 % metal coverage on the front side to account reasonably for shading losses. It was confirmed that the process sequence is very robust for resistivity values ranging from 2 10 Ωcm. Also, as can be seen in the upper graph of Fig. 4.2, a bulk lifetime of one millisecond can be considered a threshold beyond which the cell structure is no longer limited by the bulk diffusion length. The impact of the passivation quality was modelled through the surface recombination velocities. The offset between the two current and two voltage curves in the diagram represents the combined surface recombination contributions, star symbols are used to represent ideal surface passivation. Improvements of the interface quality directly result in higher V OC and J SC levels as can be concluded from Fig The surface passivation quality of the front emitter surface is much more critical compared to the influence of the rear surface passivation.

58 50 Chapter 4: Cell processes V O C [ m V ] V O C J S C S R V = c m /s S R V = 0 c m /s J S C [ m A / c m ² ] E m itte r s h e e t re s is ta n c e [ /s q ] V O C [ m V ] V O C J S C S R V = c m /s S R V = 0 c m /s B S F s h e e t re s is ta n c e [ /s q ] J S C [ m A / c m ² ] Figure 4.3: PC1D simulation for cell performance depending on doping profiles of emitter and BSF, with one profile kept constant. Simulations performed at moderate SRV values of 5000 cm/s (best fit to experimental IQE) and for an ideal scenario without surface recombination at either surface. Values calculated for 6 % metal shading, dashed red lines indicate values for baseline cell. For a cell illuminated from the front surface this is the expected behaviour. The majority of carriers is being photogenerated close to the emitter surface, and accordingly the recombination properties at that illuminated surface have a stronger impact on the cell performance. Also the strong BSF diffusion reduces the requirement for rear passivation quality. From the best fit to experimental IQE data we extracted SRV f ront = 5000 cm/s, as indicated by the dashed red line. Judging from that, it is shown that the front passivation from the BSG/SiN x stack is not ideal and additional 3-4 mv in V OC can be expected for near ideal passivation ( 1000 cm/s). The passivation of the heavily doped BSF does not limit the cell performance and a reduction in SRV rear alone would not lead to enhanced V OC or I SC. The set of simulations shown in Fig. 4.3 shows the influence of the variation

59 4.1. Cell processes with homogeneous diffusions 51 in the emitter and BSF profile respectively with the other side kept constant at 70 Ω/sq. for the boron emitter and 55 Ω/sq. for the phosphorous BSF. Since the simulation is based on approximated doping profiles the extracted absolute values may differ from actual experimental data. However the direction and magnitude of the effects can lead to insights for further cell improvements. From Fig. 4.3 it can be concluded that reducing the emitter doping immediately improves the current, due to less Auger recombination, and also gradually increases the voltage. If the front surface passivation (lower SRV) is improved at the same time, a drastic gain in V OC can also be expected. On the rear surface a large gain in voltage is expected for reduced doping concentrations, while the benefit in current is only marginal. The reduction of rear doping at the same time increases the requirements for the rear passivation. The gap between state-of-the-art and ideal passivation widens. In this case a high quality rear-side passivation can unlock the full V OC potential. The comparison of these two graphs indicates that the cell voltage is mainly limited by the rear side doping, while the cell current is determined by the front emitter doping. It has to be kept in mind, that actual decoupling of front and rear side doping profiles is impossible in the experiment. However, the simulation allows us to treat the front and rear side doping profiles independently. From this precursor performance data the efficiency potential of fully metallized cells can be estimated. We assumed a FF of 79% for all cells, which is found for a series resistance contribution of 3 mω and parallel resistance contribution of 10 kω. These numbers were extracted from best cell results. The shading losses are set to 6% which is also achievable using fine line printing of 80 µm finger width. The results for the current doping profile configuration, the case for a 110 Ω/sq. BSF and 110 Ω/sq. diffusion on either side are tabulated below. The doping profiles are approximated by error functions, which gives the best fit to experimental data. This does not yield the same shape of the doping profiles found by ECV analysis. Especially the peak surface concentrations for the simulated profiles are overestimated, which reduces the cell potential. Every configuration is calculated for average recombination values of 5000 cm/s at each surface and a lifetime of 1.5 ms but also for a near ideal scenario with bulk lifetime of 5 ms and without surface recombination. The data can be considered an upper limit for large area cell efficiencies without losses due to nonidealities or recombination at the metal contacts. Emitter R sheet BSF R sheet SRV f ront/rear τ bulk imp. V OC J SC η [Ω/sq.] [Ω/sq.] [cm/s] [ms] [mv] [ma/cm 2 ] [%] Table 4.2: Performance estimates from PC1D simulations.

60 52 Chapter 4: Cell processes Cell processing of full ingot range The lower segregation coefficient of phosphorous as compared to boron in silicon results in a larger resistivity spread along the length of a typical Cz-monocrystalline ingot (compare 2.1). The efficiency distribution over an ingot is a crucial parameter for cell manufacturers, who need to maximize the ingot yield, e.g. the fraction of the ingot that can be processed into solar cells within a specific efficiency range. It was therefore interesting to evaluate the robustness of the process towards the utilization of wafers from different regions of an entire production ingot. A large experiment has been performed using wafers of 240 µm thickness from 5 different positions throughout the ingot. The distribution of IV results of that experiment can be seen in Figure 4.5. The error bars show the standard deviation within the groups comprising of cells. The wafer resistivity was determined with 4-probe measurements after thermal donor annealing. In general the FF decreases with increasing resistivity due to higher series resistance contribution from the bulk material, at the same time the cell current is increased because of higher minority carrier diffusion length. This is a direct consequence of a more effective BSF screening due to a lower doping level in the bulk. The IQE comparison in Fig. 4.4 of cells from 2 and 5 Ωcm resistivity wafers reveals this mechanism. The IQE was measured from either side and the rear side performance of high resistivity cells is clearly superior. Over the range of an ingot, both effects compensate each other. This means that the cells from the high resistivity end of the ingot ( 10 Ωcm) become series resistance limited while cells from the low resistivity bottom ( 2 Ωcm) are ultimately limited by the J SC of the cell. This balancing leads to a narrow efficiency distribution, as desired from the cell manufacturers point of view. It is a topic of debate whether the supposedly larger spread in the current distribution simply shifts the mismatch problems towards the module manufacturers. While it is difficult to see these trends in the experiment due to large error margins, it is noteworthy that cells with stable efficiencies in a narrow range have been demonstrated over the range of an entire production ingot IQ E [% ] fro n t re a r C z, 5 c m C z, 2 c m W a v e le n g th [n m ] Figure 4.4: IQE measurements for two levels of bulk resistivity, measured from either front or rear side.

61 4.1. Cell processes with homogeneous diffusions 53 J S C [ m A / c m ² ] V O C [ m V ] B u lk re s is tiv ity [ c m ] B u lk re s is tiv ity [ c m ] F F [ % ] [ % ] B u lk re s is tiv ity [ c m ] B u lk re s is tiv ity [ c m ] Figure 4.5: IV data from solar cells with different bulk resistivity, as found along production ingot positions, each point represents cells Cell process with lowly doped emitters A variation of the emitter diffusion is carried out to fabricate cells with emitter sheet resistances in the range of Ω/sq. This can easily be adjusted, e.g. by changing the peak diffusion temperature, as indicated in Fig All cells were prepared according to the baseline process with the intentional variation of the BBr 3 diffusion. The POCl 3 diffusion resulted in a 25 Ω/sq. BSF. For each group the BSG method and additionally Al 2 O 3 based passivation was employed. The aluminium oxide passivation ensures the same passivation quality, since the BSG composition and therefore its passivation properties depend on the diffusion parameters. The IV results are reported in Table 4.3. It is obvious that the lowly doped emitter results only marginally in improved implied V OC, which indicates a high influence of the recombination on the rear side. This is confirmed by QSSPC measurements of symmetric n + n n + BSF test structures, which show J 0e values of 180 f A/cm 2 per side. The rear side doping clearly determines the level of the V OC, since the potential of the corresponding p + n p + structures has already been demonstrated earlier. The reduced front side doping improves the blue response of the cells as can be seen in Fig This directly translates to increasing current generation for lowly doped emitters as demonstrated by the IV data. These findings are well in line with the simulation results. In contrast to the cell precursors, the cell V OC is even lower for reduced emitter doping. This can be attributed to less effective shielding of the

62 54 Chapter 4: Cell processes IQ E [% ] /s q. e m itte r 8 8 /s q. e m itte r 6 7 /s q. e m itte r 5 0 /s q. e m itte r W a v e le n g th [n m ] Figure 4.6: IQE close-up of cells with different emitter doping passivated by the BSG method. metal contacts, which enhances recombination as will be demonstrated in the following chapter 5. Emitter R sheet Passivation imp. V OC J SC V OC FF η [Ω/sq.] [mv] [ma/cm 2 ] [mv] [%] [%] 50 BSG BSG BSG BSG Al 2 O Al 2 O Al 2 O Al 2 O Table 4.3: IV results of cells and implied V OC of cell precursors. Measurement uncertainties of IV measurements apply Cell process with optimized BSF diffusions It was shown in section that the V OC potential of this cell is mainly limited due to a high amount of dopant in the silicon. This is largely an inevitable result of the successive diffusions. Especially the heavily doped rear side is limiting the implied V OC. A sufficiently high surface concentration ( cm 3 ) is necessary to achieve low ohmic rear side contacts. In the present order of diffusions the rear side is protected during the BBr 3 step by a PECVD capping layer. The high thermal budget leads to significant migration of phosphorous atoms as ECV measurements before and after BBr 3 diffusion have shown. Due to the limited supply of phosphorous during boron diffusion the dead layer can be fully dissolved. Depending on the POCl 3 deposition this can even

63 4.1. Cell processes with homogeneous diffusions 55 lead to surface depletion. Continuing research in the field of paste development has put forth new generations of contacting silver pastes. These pastes are able to provide low contact resistance even on polished and lowly phosphorous doped surfaces. For many years it was stated that a phosphorous dead layer enables low contact resistances, but also this requirement has been rendered obsolete by todays pastes. It was further found that recombination at the emitter metal contacts leads to significant reduction in the cell V OC. This is confirmed by the earlier result that shallow emitter diffusions lead to lower cell voltage, even though the blue response increases in that case. In this experiment we therefore follow the other direction and also employ deep homogeneous emitters to improve the shielding of contacts. A scan of the BSF diffusion levels thus needs to be done, to maintain the lowest possible rear side doping with low contact resistance. The goals of this experiment were: Determine the BSF sheet resistance with the best trade-off between low recombination/free carrier absorption and low contact resistance, for a standard, 70 Ω/sq. emitter diffusion. Demonstrate the benefit of a deep emitter diffusion on contact recombination. Demonstrate the achievable doping contrast between front- and rear side doping for deep emitter diffusions. The experiment presented is a screening comprising six POCl 3 diffusions, and three different emitter diffusions, resulting in thirteen experimental groups. The resulting sheet resistance values are indicated in Fig. 4.7 and were measured at the end of the process on p-type wafers. For the BSF diffusions the parameters of POCl 3 flow, deposition time and temperature during the process have been varied. It can be seen that extended deposition time of the POCl 3 diffusion leads to lower sheet resistance. The variations in the emitter diffusions involve the diffusion time at peak temperature. This is done B S F S h e e t re s is ta n c e [ /s q ] s t d. B B r 3 l o n g B B r m i n d e p o s i t i o n 1 5 m i n d e p o s i t i o n 2 0 m i n d e p o s i t i o n D i f f u s i o n t e m p e r a t u r e [ C ] Figure 4.7: BSF sheet resistance data over POCl 3 diffusion temperature measured after boron diffusion for std. emitter and extended drive-in emitter.

64 56 Chapter 4: Cell processes Emitter R sheet BSF R sheet imp. V OC J SC V OC FF η rear side ρ c [Ω/sq.] [Ω/sq.] [mv] [ma/cm 2 ] [mv] [%] [%] [mωcm 2 ] Table 4.4: IV data, implied V OC and resistivity values for BSF and emitter variation (averages 3-5 cells) by drive in of the default 70 Ω/sq. emitter for additional 20 min or 60 min at peak temperature. Notice that in all groups the PSG formed during POCl 3 diffusion remains on the wafer as a phosphorous source during the following emitter diffusion. Even so, a high sheet resistance BSF ( 75 Ω/sq.) can be achieved when the boron diffusion with additional drive-in of 60 min at elevated temperature is employed. Table 4.4 shows the IV data of the respective groups averaged over 3-5 cells. The main results are also depicted in graph 4.9. The trade-off between recombination and absorption in the BSF on the one hand, and the series resistance on the other hand can immediately be seen. Independent of the emitter diffusion employed, an increase in BSF sheet resistance leads to an increase in J SC due to less free-carrier absorption and also increases V OC due to less recombination. At the same time the lateral resistance and especially contact resistance contribution reduces the FF. All cells have been printed using the same two metal layouts on the front and rear side respectively. The POCl 3 diffusion used in the baseline process, which is part of this screening, yielded 25 Ω/sq. A new optimum was found in the settings for a sheet resistance of 55 Ω/sq. We have therefore successfully adapted our process to improvements made in metallization pastes. New pastes allow to contact polished phosphorous doped surfaces of higher sheet resistance. The ECV profiles of emitters as well as the old and new BSF diffusions are depicted in Fig Another result from this investigation is that a stable offset in V OC of around 5 mv is seen between the groups with the standard boron diffusion and extended diffusion as visualized in Fig Although the implied V OC samples with the standard diffusion have shown higher values than for the extended diffusion the relation on cell level is reversed. In fact the highest cell V OC is measured on cells with 34 Ω/sq. emitter. This seems counterintuitive as for a deep diffusion the Auger recombination, which is the

65 4.1. Cell processes with homogeneous diffusions 57 ] -3 D o p a n t c o n c e n tra tio n [c m /s q E m. 3 4 /s q E m D e p th [ m ] / s q B S F, n e w / s q B S F, o l d D e p t h [ µ m ] D o p a n t c o n c e n t r a t i o n [ c m - 3 ] Figure 4.8: Doping profiles measured by ECV from default and driven-in boron emitters (left) and default and shallow phosphorous BSF (right), blue squares represent best known configuration. dominant recombination mechanism in the emitter bulk, is certainly higher. Another detrimental effect is the recombination at the front metal contacts and it was shown (refer Chapter 5, [72]) that deep emitter diffusions help to reduce the contact recombination through better shielding of the metal interface. The increased V OC is therefore the result from two competing influences, the significantly reduced contact recombination on the one hand and increased emitter bulk recombination on the other hand. This knowledge leads to the idea of the selective emitter investigation for n-type cells, which is the content of the following section iv O C, V O C [m V ] V O C iv O C 3 4 /s q E m itte r 6 6 /s q E m itte r B S F s h e e t re s is ta n c e [ /s q ] Figure 4.9: Measurements of implied V OC and cell V OC extracted from QSSPC and IV data for combinations of different emitter and BSF diffusions

66 58 Chapter 4: Cell processes 4.2 Concepts for selective cell doping Motivation Many methods have been developed in recent years to manufacture locally selective doping structures for silicon solar cells. The goal of each of these methods is a high doping contrast between areas that are covered by metal contacts and passivated surfaces in between. A high doping concentration underneath the metal contacts is required to facilitate contact formation, whereas low doping concentration in the remaining, open areas reduces emitter recombination and allows for higher passivation quality. Laser doping is by far the most popular method for selective emitter formation in conventional p-type solar cells. A review of state-of-the-art methods was been given by Hahn [87]. We have applied here three methods to achieve local doping of n-type cells without the need for masking steps. Laser doping from highly doped layers Screen-printable doping pastes Selective etching of highly doped silicon Applications of selective cell doping in n-type cells are motivated by the high overall amount of dopants in case of PERT cell architectures and in order to reduce the detrimental influence of contact recombination. The latter goal is therefore different compared to conventional p-type cells, where a reduction of contact resistance along with improved front side passivation quality is at the center of interest Laser doping for selective emitter and BSF formation Introduction Local highly doped regions under screen-printed metal contacts of bifacial n-type solar cells are suggested as a cure for the problem of severe V OC limitation of such cell types [88, 89]. Until today selective boron emitters have not been successfully implemented in high-efficiency devices since the approaches known for n + emitter doping are not easily transferable [90 92]. Laser doping, which is the state-of-the-art technique for n + emitter formation [93], was experimentally investigated for p + doping using different process sequences and precursor layers. Screening experiment Two emitter doping sequences have been suggested: The first is a sequence of a Boron tube Diffusion followed by Laser Doping, hereafter BD/LD. Here an initial BBr 3 diffusion is applied, resulting in a homogeneous emitter followed by removal of the BSG layer. A thin source layer of boron is then sputtered onto the emitter surface and driven-in by laser assisted diffusion. This sequence requires an additional cleaning step prior to the passivation of the emitter by a wet thermal oxide/sin x stack.

67 4.2. Concepts for selective cell doping 59 In the second sequence the diffusion steps occur in reversed order. This is hereafter called Laser Doping/Boron tube Diffusion, or LD/BD. Initially the boron is sputtered on the bare wafer surface and is locally diffused by laser assistance. After removing the residuals of the precursor by chemical cleaning a tube diffusion is applied to yield the desired sheet resistance between the laser doped lines. The benefit is that the BSG layer can be utilized as part of the passivation stack, as is routinely done for homogeneous emitters. Also we have evaluated a two-step diffusion approach, where the BBr 3 flow in the tube is intentionally delayed to provide an additional drive-in period, only for the already doped areas. Thereby the difference in the sheet resistance between laser processed and unirradiated areas, the doping contrast, can be enhanced. The sequence BD/LD theoretically offers the opportunity to use the BSG layer directly as a doping source. We have tried to apply the as-grown BSG, as well as a stack of BSG capped with sputtered boron as a doping source. The sputtering deposition of boron and the following laser processes were carried out at IPV Stuttgart. A vacuum sputtering tool and a 532 nm ns-pulsed Nd:YAG laser with line-shaped beam focus were employed. A detailed description of the laser system and deposition parameters can be found elsewhere [94]. For each sequence, a matrix of 2 2 cm 2 laser fields was irradiated on large area 5 Ωcm Cz n-type wafers, whereas the pulse energy density H, the pulse repetition rate f and the spot-overlap O x were varied. It is assumed that these laser processed fields show the same recombination, passivation and contacting properties as the laser processed lines of about 200 µm width that are to be applied on the cell. The samples were then characterized with 4-probe resistance measurements, ECV and SIMS profiling and laser scanning microscopy (LSM). The implementation of laser doping of p + structures in n-type solar cells might require doping of textured surfaces. Accordingly test structures have been made on random pyramid textured precursors. Based on the diffusion outcome parameters were chosen to be implemented in the n-pert solar cell process. Aside from the laser doping step the cells were processed according to the reference process outlined in Section 2.1. The figures 4.10 and 4.11 show the sheet resistance data measured on polished wafers with sputtered boron layers from an initial screening experiment. The goal was to evaluate the influence of the deposition and laser parameters for future cell runs. It is obvious that for both sequences a significant reduction in sheet resistance can be achieved. As expected, increasing the pulse energy density H leads to lower sheet resistances. The same effect is observed for increasing pulse overlap, here from 27% to 84%. The sputtering deposition time corresponds linearly to the thickness of the boron source layer on the wafer. The measured sheet resistance decreases globally for each lasers setting, for longer deposition times. The difference from 1 min to 2 min is especially pronounced. For 1 min deposition time the lowest achievable sheet resistance is 30 Ω/sq., while values even below 15 Ω/sq. can be achieved by doping from thicker boron layers. This indicates a limited supply of boron in the first case. For sputtering times longer than 2 min the reduction in the sheet resistance is only marginal, which might indicate a sufficient supply of dopant atoms for the given laser parameters. It is found that slightly lower sheet resistances can be achieved in the sequence BD/LD. This is surprising since

68 60 Chapter 4: Cell processes S h e e t re s is ta n c e [ /s q.] m i n d e p o s i t i o n m i n d e p o s i t i o n P u l s e E n e r g y D e n s i t y [ J / c m ² ] O v e r l a p 2 7 % O v e r l a p 6 6 % O v e r l a p 8 4 % S h e e t re s is ta n c e [ /s q.] S h e e t re s is ta n c e [ /s q.] Figure 4.10: Sheet resistance after BBr 3 diffusion and laser doping (BD/LD) for boron deposition time of 1 min (left) and 3 min (right) 1 m i n d e p o s i t i o n m i n d e p o s i t i o n P u l s e E n e r g y D e n s i t y [ J / c m ² ] O v e r l a p 2 7 % O v e r l a p 6 6 % O v e r l a p 8 4 % S h e e t re s is ta n c e [ /s q.] Figure 4.11: Sheet resistance after laser doping and BBr 3 diffusion (LD/BD) for boron source deposition time of 1 min (left) and 3 min (right) the subsequent boron diffusion implies an additional thermal budget, which should in principal lower the sheet resistance. To get a better understanding of the influence of the doping sequence SIMS measurements were performed. These doping profiles are shown in Fig and it can be concluded that the laser step results in high surface doping concentrations, while the subsequent tube diffusion effectively redistributes the dopant atoms wherein the peak concentration is reduced and also deeper junctions are achieved. Again, higher overlap (Fig. 4.12) leads to an deeper melting of silicon and accordingly deeper doping profiles. It has been demonstrated (refer Chapter 5, 4.1.4), that especially deep diffusions are suitable to provide electrical shielding of metal contacts. This is why the sequence LD/BD appears favourable. Laser doping from as-grown BSG layers, as well as BSG layers capped with sputtered boron, did result in very inconsistent doping results. On many wafers with BSG a reduction in sheet resistance could not be observed, while blistering of the source layers was detected. The BSG layer is much thicker compared to the sputtered boron source. It can be assumed that the build-up of thermal stress in the BSG layer due to heating of the underlying silicon can be the cause for this. Also it is known that the BSG grown during a 70 Ω/sq. diffusion contains only little

69 4.2. Concepts for selective cell doping 61 Figure 4.12: SIMS profiles for selected groups with 27% and 84% overlap. boron [95]. This is desirable to avoid the formation of a BRL, but at the same time makes it unsuitable as a doping source for homogeneously doped lines. For the following lifetime analysis the set of parameters was therefore limited to a deposition time of 2 min, pulse energy densities of 1.8 and 2.2 J/cm 2 and 66 % overlap unless otherwise indicated. Still both sequences have been applied. Selective emitter cells in the sequence BD/LD Lifetime values have been obtained by QSSPC measurements on 4 4 cm 2 homogeneously irradiated laser fields of wafers processed along with the cells. The array is irradiated on the emitter side with different laser parameters and a reference value is measured in unexposed areas of the same wafer. The data is reported in Table 4.5. The measurements on cell precursors after sequence LD/BD indicate a high junction quality, devoid of laser-damage. The obvious reduction in implied V OC can be explained by the strong reduction in sheet resistance and the associated increase in emitter recombination. Especially the samples irradiated with high intensities show a drop of up to 16 mv in implied V OC. Given the low area fraction, of around 8% of the irradiated fingers this is still tolerable and would reduce the potential implied V OC by only 2 mv. For the sequence BD/LD we find a greatly reduced implied V OC compared to the LD/BD reference. Since the reference samples did not receive laser treatment similar values should be achieved. It is assumed that this degradation is caused by contamination or insufficient passivation, which would then also have affected the other measurement points of the BD/LD samples. While the overall implied V OC level is therefore low for the BD/LD group, the same trends can be observed as in the LD/BD group. The IV data for the cells processed according to sequence BD/LD is shown in Table 4.6, while each line represents the average of 3 to 5 cells. The overall level of V OC, along with a reference value of 643mV, indicates that the lifetimes of these cells must be higher than what was previously measured on BD/LD cell precursors. Anyhow the J SC for the reference cell is relatively low, compared to the best results of the reference process (compare 4.1.4). Furthermore we see an increased V OC for all selective emitter cells, which can be attributed to reduced contact recombination. This proves that better shielding of the metal contacts can be achieved by selective emitter doping. Here, the reduced metallization losses overcompensate the increased emitter recombination. This

70 62 Chapter 4: Cell processes Sequence H Overlap O x τ e f f at 1sun V OC at 1sun J 0e [J/cm 2 ] [%] [µs] [mv] [fa/cm 2 ] BD/LD ref BD/LD BD/LD BD/LD LD/BD ref LD/BD LD/BD LD/BD LD/BD LD/BD Table 4.5: QSSPC lifetime data measured on 4 4 cm 2 homogeneously processed laser fields for both sequences. results in a V OC improvement of up to 5 mv. The constantly high level of the pff shows that the local melting and recrystallization of the emitter during the laser step does not induce defects near the junction. Unfortunately a significant reduction in the FF is found for cells contacted after the laser process. The application on a front surface emitter requires doping of textured surfaces and it is found that the surface topology is altered with increasing pulse energy density to a large extent. This can be seen from the LSM images in Fig The heavy doping is achieved through local melting of the silicon, this goes hand in hand with a rounding of the pyramid structure in the irradiated areas. It is known from many studies [96, 97] that the growth of silver crystallites, which is desired for low contact resistance, is enhanced at sharp surface features, such as pyramid edges and tips. It may be well expected and is indeed confirmed by 4 probe measurements that the contact resistance to the boron emitter is increased. We have measured a large variation of contact resistance values between mωcm 2 for the 1.8 J/cm 2 group and mωcm 2 for the 2.2 J/cm 2 group. For the reference we measured 8.2 mωcm 2, while the rear side contributes by only 0.9 mωcm 2 to the overall series resistance. The firing optimization showed, that firing at higher temperatures can reduce the contact resistance only slightly, while at the same time having a detrimental effect on the V OC. These values coincide with the contact resistivity measured on polished surfaces after tube diffusion, of around 32 mωcm 2. This is obtained, although the surface concentrations for the sequence BD/LD can exceed surface concentrations after tube diffusions by a factor of 2-3. It is concluded, that sharp surface features like pyramid edges, are required to enable contact formation using aluminium containing silver paste on p + surfaces. Even the highly doped, but flat or rounded surfaces, always exhibit high contact resistance values. Additionally the planarized surface area has a higher reflectivity. Since not the entire laser area will be covered by metal, due to alignment tolerances, a reduction in J SC is also observed. It can therefore be concluded that the laser pulse energy densities chosen in this experiment were too high to be able to contact the boron emitter with sufficiently low contact resistivities.

71 Concepts for selective cell doping H [J/cm2 ] Tpeak [ C] JSC [ma/cm2 ] VOC [mv] FF [%] η [%] pff [%] Ref Table 4.6: IV data of reference and selective emitter cells processed in sequence BD/LD, with 66% overlap and 3 min sputtering time. Selective emitter cells in the sequence LD/BD The BD/LD results have revealed the trade-off between depth of the emitter profile and the contact resistance. From the results of the first cell experiment it is clear that selective emitters can not be created by laser treatment, powerful enough to planarize the surface texture. The optimization has to focus on tailoring the emitter profile while keeping the front contact resistivity at values of, or below the reference emitter. For this study textured wafers were processed in the sequence LD/BD and the source deposition time was varied between 1, 3 and 5 min. The sequence LD/BD has two advantages. On the one hand the BSG formed during boron diffusion can be maintained as a passivating layer and on the other hand the resulting emitter profiles have a greater junction depth due to the thermal budget of the diffusion step (compare 4.12). The subsequent high temperature step even offers the possibility to tailor the dopant distribution under the contacts independent of the passivated areas. This approach is limited by the impact of the thermal budget on the BSF doping. Additional to the standard boron diffusion (70 Ω/sq.) we created an extended temperature profile for a second group. This group was submitted to a two step diffusion profile, where a drive-in plateau at 980 C for 30 min is performed prior to switching of the BBr3 gas flow. All samples were then passivated, printed and fired together, according to the baseline process. Figure 4.14 shows the resistivity results of this experiment. The starting point (0 pulse energy density) represents the reference values for contact and sheet resistance measured on areas without laser processing. It is confirmed that the contact resistance increases steeply with increasing Figure 4.13: LSM images of molten pyramid texture in laser irradiated areas for defocused laser beam (left), focused 1.8 J/cm2 (center) and 2.2 J/cm2 (right).

72 64 Chapter 4: Cell processes laser power above 1.4 J/cm 2, almost independent of the type of BBr 3 diffusion or boron deposition parameters. This pulse energy density thus marks the onset of significant pyramid rounding, as was also confirmed by LSM imaging. The sheet resistance follows the opposite behaviour, it is reduced while the incident laser power is increased. Once more the effect of longer deposition time (3 min over 1 min) is reproduced. The reference sheet resistance of the extended boron diffusion was found to be unintentionally low (40 Ω/sq.) due to early switching of the BBr 3 flow. The values for laser doped fields are therefore also shifted to lower sheet resistances. Still the trade-off between doping depth and contact resistance could be demonstrated clearly and a suitable process window around 1.1 J/cm 2 was identified. In the following the experiment was repeated with a focus on the pulse energy density range of J/cm 2. The extended diffusion was adjusted for a delay time of 38 min at high temperature and reduced deposition of BBr 3. In order to guarantee sufficient boron supply the deposition time of 5 min was added. From Fig we can see that the extended diffusion is now in a suitable range for homogeneous emitter doping (60 Ω/sq.). Also a considerable reduction of the sheet resistance in the laser-irradiated areas could be achieved ( 40 Ω/sq.). A batch of cells with the adjusted laser settings (1.1 J/cm 2, 66% overlap and 3 min deposition) was processed. S h e e t r e s i s t a n c e [ /s q.] c [ m c m ² ] m i n + s t d. B B r 3 3 m i n + s t d. B B r 3 1 m i n + e x t. B B r 3 3 m i n + e x t. B B r P u l s e e n e r g y d e n s i t y [ J / c m ² ] Figure 4.14: Contact resistivity and sheet resistance after laser doping and BBr 3 diffusions (LD/BD) for 1 min (open symbols) and 3 min (full symbols) deposition time.

73 4.2. Concepts for selective cell doping 65 S h e e t r e s i s t a n c e [ / s q. ] m i n + s t d. B B r 3 5 m i n + s t d. B B r 3 3 m i n + e x t. B B r 3 5 m i n + e x t. B B r c [ m c m ² ] P u l s e e n e r g y d e n s i t y [ J / c m ² ] Figure 4.15: Contact resistivity and sheet resistance after laser doping and BBr 3 diffusions (LD/BD) for 3 min (open symbols) and 5 min (full symbols) deposition time. Unfortunately all samples which received a boron sputtering deposition exhibited low lifetime values after processing, indicating a contamination or sputtering damage problem. This lead to significantly reduced V OC and pff on cell level, which makes the evaluation in terms of metal contact recombination impossible. It is interesting however to note the surface topology for these settings as is shown in Fig The image shows that just precisely the tips of all pyramids are molten. The contact resistivities on such surfaces were measured and are stated in Table 4.7. Although the pyramid tips are obviously strongly affected, the contact resistance stays in the same range as for the reference sample. If we assume that contact formation still occurs mainly via imprinting of crystallites to the tips, it can be assumed that the shielding is also improved for these doping conditions. This can, however, only be demonstrated in another cell run. Laser doping for selective BSF formation The impliedv OC limitation imposed by the high doping concentration in the wafer (refer Sec ) can be overcome by applying selective doping for the rear side BSF. Also the rear side contact resistivities can be improved by locally high doped regions underneath the contacts. The given process sequence allows for the application of laser

74 66 Chapter 4: Cell processes Contact Group RSheet [Ω/sq.] ρc [mωcm2 ] front front front rear rear reference selective selective ext. reference selective BSF ± 2.1 ± 2.1 ± 0.6 ± 0.6 ± 0.5 Table 4.7: Summary of resistance measurements from cells fabricated by the LD/BD sequence, with optimized laser parameters. doping from PSG as is widely used for selective emitter formation in p-type solar cells [98 100]. Moreover the power restrictions due to pyramid rounding do not apply on the polished rear side and the requirements on alignment accuracy are lower since there is no risk of accidental cell shunting on the BSF side. We have evaluated the application of a selective rear BSF for bifacial n-type cells in the following experiment. For ease of process integration it is useful to apply the laser step directly after the phosphorous diffusion. The laser processes are again done at IPV Stuttgart. Therefore the PSG will need to be removed after the laser step in this experiment. This imposes a general restriction on the achievable sheet resistance range. The POCl3 diffusion thus has to be adapted. One has to keep in mind, that the doping profile in the contact regions will also be redistributed during further high temperature processing. We evaluated the process for a homogeneous POCl3 diffusion resulting in Rsheet = 160 Ω/sq. as can be seen from the reference values in Figure The following process steps were performed according to the baseline process, i.e. including a simulated boron drive-in step. The sheet resistance shows the expected behaviour for values exceeding the threshold energy density of 0.75 J/cm2. For the PSG layer of this test diffusion a saturation is reached for values around 1.40 J/cm2, at which apparently all available phosphorous from the PSG diffuses into the wafer. Parallel to the reduction in sheet resistance the contact resistance is greatly reduced. Contact resistivities around 1.5 mωcm2 could be demonstrated, which is well below the values on the reference BSF. The removal of the PSG after laser doping requires also the application of different passivation stack. We employ a PECVD deposited SiO/SiNx stack. This passivation method has not yet been fully characterized. Figure 4.16: Collage of LSM images of bare pyramid texture (right) and area irradiated with focused 1.1 J/cm2 line beam.

75 4.2. Concepts for selective cell doping 67 S h e e t r e s i s t a n c e [ / s q. ] B S F + l a s e r c [ m c m ² ] P u l s e e n e r g y d e n s i t y [ J / c m ² ] Figure 4.17: Sheet resistance and contact resistivity for homogeneously laser doped samples with 160 Ω/sq. test BSF diffusion. It was shown that the passivation quality on n + surfaces is higher than for pure SiN x. It is assumed that the PECVD SiO under the SiN x capping layer is converted into a thermal oxide during the high temperature step. This could explain the better passivation quality than pure SiN x. This result is supported by refractive index measurements on passivation stacks after simulated boron diffusion. For the SiO 2 layer a refractive index of n = 1.47, very close to n = 1.45 of thermal oxide has been found. Due to the lower rear side doping of the BSF, 160 Ω/sq. instead of 50 Ω/sq., the requirements on the interface passivation are certainly increased. A thermal oxide is known to provide low D it, while pure SiN x passivation relies on a field effect component from a high positive fixed charge density. A cell batch has been fabricated with selective BSF doping of 250 µm lines. The matching of the metal layout onto the laser doped areas was achieved by edge alignment. A comparable efficiency between reference group and selective BSF cells has been achieved. The selective BSF doping leads to an increase in V OC, while other cell parameters are on the same level as the reference. This proves the suitability of the new rear side passivation and thereby increased implied V OC. Also the FF level above 77.5% proves the low series resistance losses, which implies a low contact resistivity also on the rear side. A successful integration of selective BSF doping has thus been demonstrated. This first demonstration already yielded the same efficiency level as for the reference

76 68 Chapter 4: Cell processes Group R sheet J SC V OC FF η pff [Ω/sq.] [ma/cm 2 ] [mv] [%] [%] [%] Ref. BSF Sel. BSF Table 4.8: IV data of reference and selective BSF cells processed with 66% overlap and 1.4J/cm 2. group. A further optimization of the rear diffusion and passivation layers can certainly result in higher performance. This optimization is anyhow beyond the scope of this work and should be performed on production tools. Even if lower doping contrasts can be achieved in this sequence it might still enable contact formation on Ω/sq. homogeneously doped BSF layers. As demonstrated in the previous section (4.1.4) this can already improve implied V OC and J SC potential of the cell significantly. Additionally this sequence would only require one extra laser step, without the need for further cleaning or passivation steps. Conclusion Laser doping as the state-of-the-art process to form selective emitters on p-type cells was evaluated for selective boron emitter doping of n-type cells. It was demonstrated that the primary goal of reducing contact recombination can be achieved, however currently only at the expense of significantly increased contact resistance. A trade-off between increasing contact resistivity and reducing sheet resistance exist for emitters on textured surfaces using current metallization pastes. This trade-off drastically narrows down the process window in which laser doping for selective emitter formation can be considered. The requirements of having a deep diffusion underneath the contacts, while maintaining the surface topology can only be achieved by laser processing with very low pulse energy densities ( 1.1 J/cm 2 ). With this energy level only the pyramid tips are molten. It can therefore be assumed that also the newly incorporated boron is kept at a minimum. The suggested option to perform a drive-in step, before applying BBr 3 atmosphere can help to increase to doping contrast. In general the process sequence with subsequent boron tube diffusion is favourable since the additional thermal budget leads to greater junction depths. The drive-in option allows for higher doping contrast and the BSG passivation can be maintained. One has to keep in mind, that further drive-in also impacts the rear doping profile, eventually leading to lower sheet resistance. The use of BSG as a doping source has been evaluated and generally suffers from the same limitations. It is found that the relatively low boron content of the glass and high layer thickness additionally discourage this application. A first batch of cells shows that there is no laser damage and no sputtering damage. The application for selective rear side doping has shown promising results. The laser step can easily be implemented in the process sequence and fully optimized will lead to an overall reduction in dopant density for this cell concept. This has beneficial influence on V OC and J SC if improved rear side passivation is applied at the same time.

77 4.2. Concepts for selective cell doping Screen-printable doping pastes Over the years many attempts have been made to substitute tube based diffusion from gas sources by inline diffusions which could use spray-coated or screen-printed precursors. This was motivated by the high throughput capacity of such inline systems and also by the perspective of improving sheet resistance homogeneity. Moreover tube diffusion systems require sophisticated ventilation and exhaust gas cleaning measures, due to the toxic nature of the process gases. Many studies exist which investigate boron diffusions carried out from paste precursors and find evidence of bulk lifetime degradation during processing [101, 102]. It is commonly believed that contaminants from the diffusion source, which are then incorporated in the wafer, are the reason for this. Screen-printable boron diffusion precursors are investigated as an option to create a selective emitter. In a two-level diffusion the paste can serve as a drive-in source to form a deep diffusion in the areas to be metallized, while a delayed B 2 O 3 deposition is used to form a shallow junction in the areas to be passivated. This approach requires similar processing as the laser diffusion approach but would have the benefit that only one additional processing step is needed. The deposition and patterning of the precursor is here achieved in a single step by using screen printing. Doping paste experiment An initial screening experiment was set up to investigate post-processing lifetimes on cell precursors and the achievable boron diffusion profiles. We employed the latest version of boron containing doping paste from Hitachi Chemicals. The experiment comprised a set of textured wafers with printed boron precursor in areas of 2 2 cm 2. The samples were then diffused in the BBr 3 furnace using a standard recipe and extended recipes by 20, 40 and 60 min at T peak = 980 C. The profiles were measured with ECV as shown in Fig The graph indicates profiles measured in areas covered with paste and areas without paste on the same wafer. An accumulation of boron close to the surface is observed for all groups with boron paste precursor. This is a clear indication that the dopant concentration in the paste is much higher than in the B 2 O 3 precursor. What is interesting is that the junction depth for covered areas is markedly lower than for conventional diffused areas. This finding is in line with studies from Kurachi et al. [54], who reported diffusion retardation for APCVD precursors with excess boron concentration ( 7%). Unfortunately, this property is counterproductive for the selective emitter application. The goal for selective emitter formation is to achieve a greater junction depth in the metallized areas. Moreover lifetime structures have been manufactured in the standard process with an additional doping paste printing step using the standard finger printing layout. The diffused samples were cleaned to remove the BSG and paste residuals before a passivation layer is formed by thermal oxidation. Afterwards the oxide was capped by SiN x of 65 nm thickness. The samples were subjected to a simulated firing step in the belt furnace. Table 4.9 shows IV data for cells fabricated using the doping paste approach. It is immediately clear that the diffusion using the doping paste precursor severely degrades the minority carrier lifetime. It is very likely that contaminants from the paste have diffused into the emitter and are especially recombination active since it is a p + doped layer. This is a common finding for highly doped boron paste

78 70 Chapter 4: Cell processes ] D o p a n t c o n c e n tra tio n [c m /s q D P,5 7 /s q /s q D P,4 0 /s q /s q D P,3 3 /s q + 6 0, 3 4 /s q D P,2 9 /s q D e p th [ m ] Figure 4.18: ECV profiles of set of BBr 3 tube diffusions and fields with printed boron doping paste. precursors [101, 103, 104] and probably not owing to contaminations during the process. The selective emitter cells suffer additionally from a poor FF, which is partly owing to the lower junction quality as expressed in the pff, but also due to increased contact resistivities on the rear side. The front side contact on the contrary is even enhanced using the paste precursor, probably due to high surface concentrations. Conclusion We have analysed the application of boron paste as a precursor for the fabrication of selective boron emitters. Our findings are in line with experiences from other groups, in that printed boron pastes severely degrade lifetime either due to contaminants from the paste or diffusion induced dislocations [2] due to excessively high boron concentrations. Unfortunately in the current stage of development the available products can not be applied for high efficiency solar cell processes Chemical etch-back for selectively doped structures Introduction A different method to form a selective emitter is the selective etching of homogeneous highly doped layers. Here, two general approaches can be distinguished. Both require J SC V OC FF PFF η [ma/cm 2 ] [mv] [%] [%] [%] std. Emitter sel. Emitter Table 4.9: IV data of cells with additional selective boron paste doping

79 4.2. Concepts for selective cell doping 71 Lifetime data Resistances τ e f f at 1 sun imp. V OC Em. R sheet Em. ρ c BSF. ρ c [µs] [mv] [Ω/sq.] [mωcm 2 ] [mωcm 2 ] std. Emitter ± ±1.4 sel. Emitter ± ±5.4 Table 4.10: Additional lifetime and resistance data for boron doping paste process. a heavy diffusion defining the doping level underneath the metal contacts. As shown in Chapter 5, the motivation for a selectively doped boron emitter is to achieve a significantly deeper junction (on the order of 1 µm) underneath the contacts, while maintaining a moderate junction depth ( 0.5 µm) in the passivated areas between the fingers. The first approach comprises a masking step, wherein the wafers are cleaned after diffusion and the areas to be metallized are covered with a screen-printable etch-resist. When the wafer is then dipped into an acidic etching-bath, the unprotected silicon in between the fingers is converted into porous silicon. This process can be controlled very precisely as it is self-limiting. The porosification restricts the supply of etchant to the surface and therefore the reaction is self moderating. The colour impression of the porous layers can be used to evaluate the etching thickness with nano-meter accuracy. In a second step the thin porous silicon layer is removed in diluted alkaline solutions, thereby controllably removing thin layers of the heavily diffused silicon. The alkaline solution also removes the etch-resist in the same step. The wafers are submitted to a cleaning step before a passivation layer is applied. This process has been developed by the University of Konstanz and is commercially applied for the formation of selective phosphorous emitters. Details about the process have been reported by Dastheib-Shirazi et al. [105]. Lately etch-back of boron diffused layers has been developed by the same group and is applied here to yield a selective boron emitter structure. Etching of boron emitters using this process is more complex, since highly boron doped surfaces serve as an etching barrier and therefore different etching solutions had to be developed. Here the goal is to fabricate a high doping contrast, which implies that homogeneous removal of thick emitter layers is required. For comparison, the application on phosphorous doped layers usually requires significantly less silicon removal to get rid of a potentially deteriorating phosphorous dead layer. For boron emitters the requirement is to remove nm from the surface. This depth of silicon removal can even impact the optical surface properties due to pyramid rounding. As an alternative screen-printable etching pastes are available for the formation of selective phosphorous emitters [106]. In this scenario the etching paste can be applied directly by means of screen-printing or inkjetting. The etching reaction is thermally activated in an inline furnace at temperatures ranging from C. The reactants are formed during the temperature treatment and dissolve the silicon surface locally. The pattern which defines the finger grid, is therefore the inverted image of the intended metallization pattern. The achievable etching depth in this approach is rather limited, as the deposited etching paste gets depleted during the process. A reasonable application of this approach for selective boron emitters is questionable, due to the high

80 72 Chapter 4: Cell processes Sample surface Etching time R sheet (initial) R sheet (after etchback) [min] [Ω/sq] [Ω/sq] Cz-polished t Cz-polished t Cz-polished t Cz-polished t Cz-polished t Cz-textured Cz-textured Cz-textured Cz-textured Cz-textured Cz-textured Cz-textured Cz-textured not measurable Cz-textured not measurable Table 4.11: Sheet resistance measured on polished and textured samples in the emitter etch back process, t 1 t 2 t 3 t 4 t 5. requirements on the etching depth. Masked emitter etch-back experiment In a first screening experiment a set of textured and polished wafers was diffused with an extended BBr 3 diffusion (45 Ω/sq.) to yield symmetric test structures for the etching process. The samples were laser diced to 5 5 cm 2 pieces and the sheet resistance was measured individually on each sample before and after the etching step. The polished samples were then characterized by ECV measurements to quantify the silicon removal depth. The aim of the experiment was to demonstrate the homogeneous etchback of the emitter in large areas and to determine process conditions for selective emitter fabrication in cells. The increase in sheet resistance for increasing etching times is reported in Table Removal of significant fractions of diffused emitter surface could be demonstrated on polished and textured samples with great homogeneity. Starting from a reference profile as shown in Fig three levels of etch-back were chosen for ECV characterization. The etching depth was evaluated from this sequence of ECV measurements in Fig Also the impact of the optical degradation was to be assessed, as the increasing reflection from the etched surfaces becomes visible to the naked eye. Since the SiN x deposition was found to be inhomogeneous over the wafer pieces this evaluation had to be done on cell level. The reflection of the cell surface was later measured on the fully processed selectively etched cells and compared to the reference cell. A batch of solar cells was processed with an emitter diffusion of 45 Ω/sq. and was etched back after printing the etch-resist in the areas to be metallized. Due to the separation of finger and busbar printing layout only the finger areas were protected. We

81 4.2. Concepts for selective cell doping 73 printed the SD-2052-Al series etch-resist paste to cover lines of µm width. The paste was applied immediately after the diffusion and dried in the furnace at 100 C. Four levels of etch-back were targeted to result in 60, 90, 120 and 150 Ω/sq. emitter sheet resistances. Aside from the cells protected by the etch-resist, cells and lifetime samples were processed which were homogeneously etched for the same times. After the etching step the samples received a full IMEC cleaning before passivation with thermal oxide and PECVD SiN x. The IV data (Table 4.12) and IQE measurements are reported from this experiments. The data reveals a contamination issue for all cells and lifetime samples which were covered with the etch resist. This reduces the implied V OC of nonetched samples by more than 10 mv, which is why improved contact shielding can not be detected. Unexpectedly, a significant increase in the front contact resistivity is found with increasing etching time, which explains the FF reduction for the cells with masking layout. Both may be processing related problems, that could be due to shipping and transport of the samples, but what is more important to notice is the J SC behaviour. Initially the J SC is low due to a strong 45 Ω/sq. emitter diffusion. The current level should recover by etching of the highly doped layer to reach levels comparable to the 70 Ω/sq. emitter. It is found that even for a measured sheet resistance of 90 Ω/sq. the J SC never exceeds 38 ma/cm 2. Although the blue response is improved for the etch back emitter (Fig. 4.20), this effect is overcompensated by the increase in reflection, resulting in constantly low J SC. Conclusion The etch-back approach for selective boron emitters was investigated as an option to reduce contact recombination. It was shown that this process is inherently incompatible with the goals of a selective boron emitter. The main reason for this is that the required etching depth is always accompanied by an increase in reflectivity due to rounding of the pyramid structures. Any potential improvement in V OC due to improved contact ] D o p a n t c o n c e n tra tio n [c m n m e tc h e d n m e tc h e d n m e tc h e d R e fe re n c e p ro file D e p th [ m ] Figure 4.19: ECV profiles of deep boron diffusion (reference) and emitter profiles after etch-back.

82 74 Chapter 4: Cell processes R sheet Surface implied V OC J SC V OC FF η pff [Ω/sq] [mv] [ma/cm 2 ] [mv] [%] [%] [%] 45 Reference masked (170) homogeneous masked homogeneous masked (x) homogeneous (x) masked homogeneous masked Table 4.12: IV data of cells with masked and homogeneous emitter etch-back process. shielding will ultimately be compensated by increasing reflection losses. Additionally to the etching step itself the process will gain in complexity since masking, cleaning and aligned printing processes will be required Conclusion Selective doping processes are employed in the cell process for two reasons. The recombination at the front metal contacts is contributing by more than one third to the total dark saturation current density. This figure can be reduced by achieving very deep local diffusions underneath the contacts while maintaining moderate doping in the passivated areas. Currently only laser diffusion can be employed to implement this structure. The IQ E [% ] IQ E R e fl. 4 5 /s q. n o n -e tc h e d 9 0 /s q. e tc h e d /s q. e tc h e d W a v e le n g th [n m ] Figure 4.20: IQE and reflection data for selected cell groups for selective emitter etch-back.

83 4.3. Rear emitter cell concept 75 etch-back approach is limited due to increasing reflectivity in the etched areas between the metallized areas and the doping paste approach does not yield deeper p-n junctions than the conventional tube diffusion. At the same time both processes suffer from lifetime degradation after boron diffusion, which might be process related but can also indicate insufficient purity of the precursors. The laser diffusion yields a 5 mv enhancement in V OC, which corresponds to a reduction in contact recombination by 50%, taking into account the area weighed J 01 contributions (compare Chapter 5). In the best case, when contact recombination on the front side is negligible, a 10 mv enhancement in V OC can be expected, when all other recombination factors remain unchanged. The second motivation for applying selective doping is the reduction of the BSF doping concentration without compromising the rear contact resistivity. The importance of this reduction has been demonstrated in experiment and simulation. Here a promising approach is found by laser doping from PSG layers, which can be further improved. 4.3 Rear emitter cell concept Motivation The investigations have clearly indicated the limitations of the bifacial cell concept due to metal induced recombination (ref. Section 5). Available methods to reduce contact recombination have shown only marginal improvements while at the same time being complex and costly and therefore not suitable for the industrial environment. Moreover, any such method can only reduce the impact of contact recombination caused by strongly surface damaging firing through pastes. It seems quite evident, that as long as aluminium containing firing through pastes are employed for the p + emitter contact the metal/silicon interface will pose a major source of recombination. An entirely different metallization scheme is therefore suggested here. It is known from p-perc devices that aluminium paste can be used to form local BSF structures by alloying. The line or point contacts used in those architectures show low recombination at the metal-silicon interface due to the depth of the doped region underneath the alloy [107]. For such local BSF regions an alloying depth of 2 4 µm is routinely achieved [108]. Therefore we are interested in the recombination at local aluminium alloyed contacts. A process sequence is suggested that merges the process for bifacial n-pert cells with process steps adapted from p-perc cell processing at ISC Konstanz. Specifically the ablation of the rear passivating dielectrics by picosecond laser pulses has been employed here additionally. The resulting structure would have a boron diffused emitter on the rear Figure 4.21: Cell structure of rear emitter n-pert type cell with local aluminium contacts.

84 76 Chapter 4: Cell processes surface, interrupted by local aluminium alloyed contacts and there additionally a local aluminium doped emitter. This emitter layer underneath the aluminium alloy has to be deep enough to provide sufficient shielding of the contacts. In this scenario the presence of a p + emitter must not interfere with the formation of local aluminium alloyed contacts through openings in the rear dielectric. The front surface would be phosphorous diffused to provide a shallow n + FSF layer. Both surfaces are passivated with oxide/sin x stacks. In order to understand these questions we have started investigating the process for the structure as shown in Fig Some benefits of this cell concept over the front emitter n-pert architecture would be: Reduced metal contact recombination in the emitter Reduced silver consumption BSG passivation applicable, without the need for homogeneous appearance Upgradeable from existing p-type PERC lines On the other side the bifaciality of the front emitter n-pert cell is compromised in this new architecture. The high resistivity of bulk aluminium requires the coverage of the full rear side area with paste. In comparison to a p-perc process the number of steps and the process complexity would be even increased and furthermore, different requirements on the material quality do apply since it is a rear emitter concept [18] Experiment It is desirable from a process integration point of view to maintain most of the well known experimental sequence. We evaluated a process sequence starting with the POCl 3 diffusion followed by BBr 3 diffusion. Keeping this order of diffusion processes brings about the same challenges that have already been solved for the n-type PERT cell with a front boron emitter. In detail this meant: Finding a POCl 3 diffusion for a high sheet resistance FSF that exhibits sufficiently low contact resistance. Finding a suitable passivation layer for the front FSF. Potential need for easily removable diffusion barriers. It is also known from earlier analyses that the BSG passivation is not stable towards extended high temperature processing. During a phosphorous diffusion the passivation quality of BSG/SiN x stacks on symmetric, polished samples is severely degraded. While 685 mv implied V OC are measured on passivated samples after contact firing the implied V OC of test structures after simulated POCl 3 diffusion degrades to only 652 mv. It is assumed that hydrogen depletion of SiN x during the diffusion step does play a role. Therefore a fresh layer of hydrogenated SiN x was deposited on one side of the sample and an additional firing step was carried out. In analogy to the processes with homogeneous diffusions (4.1) a screening of different POCl 3 FSF diffusions has been prepared. The final FSF sheet resistances in the range of Ω/sq. are achieved. On the rear

85 4.3. Rear emitter cell concept 77 side a boron diffusion with 70 Ω/sq. was diffused and passivated with BSG and 150 nm of SiN x. This stack was shown to be stable against penetration of aluminium paste during firing, which could otherwise lead to cell shunting. This passivation layer is ablated in a line pattern of 50 µm width and 1500 µm spacing between the lines. Since the width of the alloyed lines is measured to be 67 µm on average this corresponds to an effective metallized area of around 4.5% (refer Fig. 4.23). The LSM image shows a cross section of a typical aluminium contact line after sample preparation. A defect etch is used to delineate the doped layer underneath the alloy and an aluminium emitter thickness of 1.5 ± 0.2 µm is measured in the LSM image. The resulting IV data is tabulated together with implied V OC values, as measured on cell precursors, in Tab The implied V OC is also depicted in Fig and obviously increases for higher FSF sheet resistance. The FSF passivation with bare SiN x yields only poor passivation quality and also the oxide/sin x passivated cells exhibit limited V OC performance compared to the front emitter PERT cell. This also leads to a poor blue response for the cells and accordingly a strong dependence of cell current on the FSF doping. For high FSF sheet resistance the contact resistivity is enlarged and starts to affect the FF as can be deducted for sheet resistances in excess of 100 Ω/sq. What is remarkable however, is the gap between implied V OC and cell V OC for non resistance limited devices. This can be as low as 13 mv, which indicates a much lower metal contact recombination compared to Ag/Al paste emitter contacts. This sequence resulted in a best cell efficiency of 19.47% with V OC of 658 mv, J SC of 38.2 ma/cm 2, 77.42% FF and 82.8% pff. Given the known process limitations in front passivation and contacting, this is a remarkable result and certainly proves the potential of this concept. In order to quantify these losses a metal contact recombination analysis according to the method explained in Sec was carried out. The cells for this experiment were prepared identically, except for a variation in the rear contact pitch ( µm) as defined by laser ablation. This enabled a variation of the effective metal contact area between 4.5 and 13.4% (ref. Fig. 4.24). Two linear fits to the J 01 data are shown, whereas the blue curve exempts the 13.4% data point from the linear fit. The resulting dark saturation current density under the metal contact J 0p + (met) is extracted to be either 660 f A/cm 2 or 1220 f A/cm 2 depending on the fitting range. During the alloy formation silicon diffuses laterally within the aluminium matrix [109]. This spread can be seen by a dark greyish staining of the aluminium layer after firing. It can be assumed that for a contact pitch of 500 µm the alloying behaviour at a single contact line is no longer independent of the neighbouring contacts. If the contact recombination is thereby enhanced, it can be justified to exempt the last data FSF R sheet imp. V OC J SC V OC FF η front ρ c [Ω/sq.] [mv] [ma/cm 2 ] [mv] [%] [%] [mωcm 2 ] ± ± ± 8.2 Table 4.13: IV data, averaged from 5 cells, impliedv OC and resistivity values for a range of FSF diffusions.

86 78 Chapter 4: Cell processes im p lie d V c e ll V O C O C iv O C, V O C [m V ] F S F s h e e t r e s is ta n c e [Ω /s q ] Figure 4.22: Implied VOC data of cell precursors and IV data of cells manufactured to yield different sheet resistance FSF. Figure 4.23: LSM micrograph of optimally fired aluminium conctact line. Cross section of sample diced by laser after bulk and alloyed metal was removed in HCl, aluminium emitter delineated by defect etch. point from the linear fit. In either case it could be demonstrated that recombination at local aluminium contacts is less severe than recombination at conventional Ag/Al paste contacts (J0p+ (met) = 4270 f A/cm2 ). In the latter case the metal induced recombination of aluminium alloyed metallization exhibits lower J0p+ (met) by a factor of six Conclusion A process is suggested to circumvent the problem of severe contact recombination by applying an all aluminium based rear metallization on top of homogeneously boron diffused and passivated rear emitter. The cell is thus a rear emitter concept which involves local opening of a passivating dielectric followed by alloying of aluminium contacts through those openings. In this brief analysis the cell concept is still limited by insufficient front side passivation quality and poor contact formation to high ohmic front FSF layers. Both represent manageable challenges, here largely owing to the fact that processes were adopted from the front emitter cell concept. A remarkable result is the facile formation of alloyed aluminium doped emitters next to previously diffused boron emitters. The high pff values indicate an intact junction even though the boron emitter is removed on a large area and replaced by an aluminium emitter with much lower doping concentrations. The contact recombination at such p+ alloyed contacts has

87 4.3. Rear emitter cell concept J 0 1 [fa /c m 2 J 0 p + ( m e t ) = f A / c m ² ] J 0 p + ( m e t ) = f A / c m ² e x t r a c t e d J 0 1 d a t a l i n e a r f i t o f J 0 1 ( F M 4-9 % ) l i n e a r f i t o f J 0 1 ( F M % ) M e t a l l i z a t i o n F r a c t i o n o n p +, F M [ % ] Figure 4.24: J 01 extracted using 1-diode equation from the experimental IV data as a function of effective aluminium metal fraction. J 0p + (met) extracted from the linear fit to the experimental data using equations 5.1 and 5.2. been found to be 6 times less detrimental than firing through metallization using aluminium containing silver pastes. These findings show that the rear emitter cell has great potential to follow the p-perc concept in production which is currently being adopted by many p-type manufacturers.

88

89 5 Metallization of bifacial solar cells 5.1 Introduction Recent studies have shown that the metallization using screen printing and firing through technique still limits the open-circuit voltage (V OC ) of such cell types [20, 89]. The already industrially proven metallization for the n + emitters by using screen printed Ag pastes does not successfully work also for the p + diffused emitters, due to very high contact resistance. Instead, an addition of aluminum to Ag pastes has been found to significantly reduce the contact resistance and enable fabrication of n-type cells with high fill factors. Several research groups have investigated the contact formation of aluminum containing Ag paste on p + emitters and found evidence of Si removal and formation of local pits producing metal spikes, that can be deep enough to cross the emitter junction and contact the base [ ]. Although the role of aluminum in the Ag paste is not yet fully understood it seems to play a critical role for minimizing contact resistance losses on p + diffused layers but, at the same time, in combination with a more aggressive glass frit it degrades the metal-si interface to a greater extent than state-of-the-art Ag pastes used for contacting n + layers. In this chapter we have investigated the metallization induced recombination losses on the emitter and back-surface-field (BSF) side of bifacial p- and n-type silicon solar cells. We applied recently proven methods to investigate the influence of the metal contact on the cell performance by using a printing technique with varying metallization fraction [113, 114]. These allow us to separate the influence of contact recombination on either side of a bifacial cell. Furthermore, we employed a two-dimensional simulation model to get further insights into the recombination mechanism at the contacts under normal operating condition of the solar cells. Figure 5.1: Cross-section view of the fabricated n- and p-type bifacial solar cells. 81

90 82 Chapter 5: Metallization of bifacial solar cells ] -3 C a rrie r c o n c e n tra tio n [c m n + ( 2 5 /s q ) p +, d e e p ( 6 8 /s q ) p +, s h a l l o w ( /s q ) D e p t h [ µ m ] Figure 5.2: Experimental carrier concentration profiles of the n + and p + diffused regions used on the solar cells throughout this study. 5.2 Experimental Details Small area (3 3 cm 2 ) silicon solar cells were fabricated on mm 2 n- and p-type Cz wafers with base doping of 3 Ωcm and 1 Ωcm, respectively. Prior to the diffusion processes all wafers were alkaline polished and cleaned in HCl and HF solutions. The p + (boron) and n + (phosphorous) diffused regions were formed, one after the other, in a quartz tube furnace containing either a BBr 3 or a POCl 3 gas. The sheet resistance of the p + diffused region was adjusted to yield either 68 Ω/sq. (labeled p + deep) or 110 Ω/sq. (labeled p + shallow), whereas the n + diffusion was adjusted to 25 Ω/sq. for all fabricated solar cells. The p + and n + regions were subsequently passivated by a thermal SiO 2 and PECVD SiN x stack. The metallization was applied by screen printing and firing of a metal paste using an H-pattern grid design on both sides of the wafers. For contacting the n + region a commercial silver paste was used, whereas the p + region was contacted by an aluminum containing silver paste. Both, Ag and AgAl, pastes used in this study are present state-of-the-art products for this type of cells. Fig. 5.1 shows a cross-section view of our fabricated n- and p-type solar cells. Since an open metallization grid was printed on both sides, the resulting solar cells are bifacial and can be illuminated from either side or both simultaneously. For the n-type cells the emitter is the p + diffused layer whereas the n + layer serves as a back surface field (BSF), and vice versa for the p-type solar cells. In the present study, two main groups of n- and p-type cells were fabricated in the same experiment. The n-type cells have either a p + deep or a p + shallow emitter and a relatively deep n + BSF whereas the p-type cells have the same n + deep emitter and p + deep BSF. The resulting p + and n + carrier concentration profiles, measured by the electrochemical capacitance-voltage (ECV) technique, are shown in Fig As can be seen in Fig. 5.2, the depths of the boron diffused regions, defined here as the depth where carrier concentration equals the bulk doping, are 0.35 µm (shallow) and 0.55 µm (deep), respectively. Similarly, the phosphorous diffusion yielded a profile depth of 0.53 µm.

91 5.3. Results and Discussion im p. V O C V O C, im p. V O C [m V ] V O C n -ty p e (d e e p ) n -ty p e (s h a llo w ) p -ty p e (d e e p ) c e ll ty p e a n d p + d iffu s io n p ro file Figure 5.3: Averaged implied V OC (iv OC ) and V OC values of n- and p-type bifacial cells fabricated with shallow and deep p + diffused regions. All cells have the same n + diffusion profile as presented in Fig The finished cells have a metallization fraction of 5% on both sides. Following the fabrication of the solar cells we proceeded with their electrical characterization. In the first instance we measured the implied open-circuit voltage (iv OC ) of the precursor cells before the metallization step, but after a firing step, using the quasi-steady-state photoconductance (QSSPC) technique [115]. The iv OC measurements allowed us to estimate the maximum V OC that the solar cells would achieve if no metallization induced recombination losses were present in the device. To investigate the influence of recombination losses at metal contacts we used printing layouts with increasing finger width on either side, resulting in a variation in metal coverage from 5% to 25%. A total of 4 to 10 solar cells were fabricated for each metal coverage and group variation. The IV measurements (illuminated and dark curves) of the finished cells were recorded under a steady state solar simulator setup on a brass chuck. The solar cell parameters reported throughout this study correspond to illumination from the side where metallization fraction (F M ) was kept constant (typically to 5%). This ensures that the effects observed in cell performance by varying F M on the opposite side are not superimposed by the effects caused from different photo-generated currents inside the cell. Thus, a solar cell with F M variation on the emitter side was illuminated from the BSF side and vice versa (unless otherwise stated). 5.3 Results and Discussion Identification of main losses The iv OC values of each fabricated cell type are compared in Fig. 5.3 with their averaged V OC after metallization, with both values measured under 1 sun illumination. A drop in V OC of typically 30 mv is observed on n- and p-type cells with deep p + diffused region, while for the shallow p + diffusion the drop exceeds even 40 mv.

92 84 Chapter 5: Metallization of bifacial solar cells IQ E [% ] n -t y p e c e l l s p +, d e e p ( 5 % ), n + ( 5 % ) p +, d e e p ( 2 0 % ), n + ( 5 % ) p -t y p e c e l l s n + ( 5 % ), p + ( 5 % ) n + ( 2 0 % ), p + ( 5 % ) p +, d e e p ( 5 % ), n + ( 5 % ) p +, d e e p ( 2 0 % ), n + ( 5 % ) W a v e l e n g t h [ µ m ] Figure 5.4: Internal quantum efficiency (IQE) measured on p- and n-type cells with varying metallization fraction on p + or n + diffused regions. All solar cells were illuminated on the metal variation side. On the one hand the significant drop in V OC observed after the metallization step can be attributed to a general reduction in surface passivation and/or bulk lifetime or, on the other hand, to a locally enhanced recombination directly beneath the metal contact. The former effect would result also in a quantum efficiency degradation of the metallized cells. To verify this hypothesis we compared the internal quantum efficiency (IQE) of each cell type and for different F M on either p + or n + diffused region. The results are plotted in Fig As can be seen from experimental data, the IQE of both cell types is not affected when neither their emitter nor their BSF F M is increased. This clearly indicates that the increase in recombination, which leads to the degradation in V OC of metallized cells, is caused by a local effect directly beneath the metal contacts. A constant IQE as a function of F M also indicate that the short-circuit current (J SC ) of the cells, except for the obvious shading effect when illuminated from the metal variation side, is not affected by the recombination losses induced by metallization. This is demonstrated in Fig. 5.5 where the J SC values, corrected for metallization shading, of n-type bifacial cells are plotted as a function of F M on the emitter side. For clarity, only the results for n-type cells with deep emitter profile are exemplified here. Fig. 5.5 also shows the series (R S ) and shunt (R P ) resistances of the same solar cells. The R S values range between 0.2 to 0.5 Ωcm 2 decrease with increasing F M due to the reduction in line and contact resistance components. On the other hand the R P values measured on these cells are higher than 12 kωcm 2, and are relatively unaffected by the metallization fraction variation. Thus, the R P is high enough and R S low enough not to affect the V OC of the cells significantly. In fact, the combined effect of R S and R P will only change V OC by at most 0.1 mv within the metal fractions investigated here. Concomitantly, the pseudo fill factor (pff) extracted from the suns-v OC curve [33], shows only values around 79-81% despite the large measured R P values. We attribute this reduction in pff to the influence of edge recombination, as demonstrated in literature [32]. The 3 3 cm 2 solar cells used in this study were laser cut from the substrate

93 5.3. Results and Discussion 85 J S C [ m A / c m ² ] J S C /( F M (p ) ) R S R P M e ta lliz a tio n fra c tio n o n p +, F M (p ) (% ) R S, R P [ c m ² ] Figure 5.5: Experimental J SC, series (R S ) and shunt (R P ) resistances of bifacial n-type cells fabricated using the deep p + emitter profile. The parameters are plotted against metal fraction variation on p + emitter while the other side has a constant F M of 5% for all cells. Since the cells were illuminated on the emitter side the J SC values were corrected for metal shading (as seen in the legend). after the metallization step, resulting in an unpassivated edge. A control batch of large cells (6-inch), having edge passivation, fabricated and metallized under identical conditions show indeed pff values in the expected range of 82-83%. These observations clearly show that the foremost affected parameter by the metallization step of our bifacial cells is the V OC. Therefore understanding these V OC losses can lead to a significant boost in conversion efficiency of screen printed solar cells Losses in open-circuit voltage In order to investigate the losses induced by the metallization steps on V OC of our cells we used an experimental approach similar to that presented by Fellmeth et al. [114], but adapted to our bifacial cell configuration. That is, maintaining a constant F M on the emitter (BSF) side of the cell we observe the effect on V OC by changing F M on the BSF (emitter) side. Fig. 5.6 shows the net loss in V OC (i.e., V OC -iv OC ) as a function of F M for all fabricated n- and p-type bifacial cells. As seen in Fig. 5.6a, the influence of metallization on a p + (boron) emitter is much more detrimental than it is on a n + (phosphorous) BSF, which shows negligible influence towards increasing F M. This trend is even more dramatic for the shallow p + emitter, where a degradation as high as 70 mv is observed. Since the surface concentration of the shallow and deep p + emitters are comparable (see Fig. 5.2) we assume a similar contacting behaviour, which was also confirmed by measuring the contact resistances. This indicates that the depth of the doping profile plays a critical role in minimizing the V OC losses by more effective shielding of minority charge carriers from reaching the metal contacts. It should be noted that, while V OC is decreasing with increasing F M, the R P stays above 12 kωcm 2, indicating that the cells do not suffer from

94 86 Chapter 5: Metallization of bifacial solar cells V O C - im p. V O C [m V ] a ) n +, B S F p +, d e e p e m itte r p +, s h a llo w e m itte r n -ty p e c e lls -2 5 b ) p -ty p e c e lls V O C - im p. V O C [m V ] p +, d e e p B S F n +, e m itte r M e ta lliz a tio n fra c tio n F M [% ] Figure 5.6: Net loss in V OC, with respect to iv OC, as a function of metallization fraction F M for n-type (a) and p-type (b) solar cells. The F M was varied only on one side of the bifacial cells (see legend) while the other side that is illuminated has a constant F M of 5% for all cells. The dashed lines represent a guide for the eye. low shunt resistance. For the p-type cells, Fig. 5.6b, the results show a comparable loss for the n + emitter and the deep p + BSF, whereas the overall reduction in V OC with increasing F M is lower as compared to the n-type cells. The relatively deep n + emitter with low sheet resistance (25 Ω/sq.) used in this study is expected to better shield the minority carriers from reaching the metal contacts as the more shallow n + emitters ( Ω/sq.) commonly used in industry. Furthermore, the results in Fig. 5.6 also show that a significantly lower degradation occurs when F M is varied on the p + BSF rather than the p + emitter, despite of the fact that the same fabrication conditions were used (i.e., doping profile, metal paste and firing temperature and dielectric thickness). Hence, the recombination of minority charge carriers at the metal-emitter interface, probably in the SCR of the p-n junction, is more detrimental for the cell V OC than the recombination losses at the metal-base interface.

95 5.3. Results and Discussion Quantifying the metallization losses Few methods to study the recombination losses caused by screen printed metal contacts have been proposed recently [107, 113, 114]. Hoenig et al. quantified SCR-recombination from the difference between ideal fill factor and pff (FF 0 -pff) by varying F M on the solar cells, while maintaining a constant series resistance (pff-ff) and a high shunt resistance [113]. This FF loss analysis was then attributed to an increase in the dark saturation current density J 02, of the two diode model, that in turn reduces cell efficiency. Fellmeth et al. used the one diode model to extract J 01 values from V OC, thermal voltage (V T ) and J SC ; V OC = V T ln(j SC /J 01 ), providing that the local ideality factor m of the cells at V=V OC was close to one [114]. The contribution to the J 01 from the recombination at the metal-silicon interface (J 0(met) ) was then determined from the slope of J 01 versus F M. Using a similar analogy, we used suns-v OC measurements [33] to evaluate the local ideality factor of our bifacial n- and p-type cells at V=V OC and found m values sufficiently close to unity, ranging from 1.05 to Thus, we directly extracted the J 01 values at V OC conditions using the ideal diode equation presented above, since the losses due to J 02 can then be neglected. The J 01 sums the losses in the bulk, p + and n + diffusion regions as follows, J 01 = J 0(bulk) + J 0p + + J 0n + (5.1) For our n-type bifacial cells the J 0p + represent the emitter contribution whereas J 0(bulk) +J 0n + quantifies the losses in the bulk and BSF, and vice versa for the p-type cells. In order to separate J 01 into contributions from the passivated and metallized surfaces, equation 5.1 is modified to include the area weighted J 0(met) : J 0p + = J 0p + (pas) (1 F M(p) ) + J 0p + (met) F M(p) (5.2) and J 0n + = J 0n + (pas) (1 F M(n) ) + J 0n + (met) F M(n). (5.3) The contributions from the passivated diffused regions were extracted from QSSPC measurements of symmetrical p + (or n + ) diffused samples under high injection [43]. We determined J 0p + (pas) (for deep diffusion profile) to be 142 fa/cm 2 and J 0n + (pas) to be 146 fa/cm 2. Since in the experiment we vary only the metal fraction on one side at a time it is possible to extract the local recombination current density J 0p + (met) (or J 0n + (met)) under the metal contacts using the equations 5.2 (or 5.3) above. This leads to an estimation of the contact recombination current densities as shown in Fig On contacted p + diffused regions (Fig. 5.7a) an extremely high J 0p + (met) is observed at the metal-emitter interface. The recombination current strongly depends on the diffusion profile (sheet resistance and depth) and can be up to two orders of magnitude higher than the J 0p + (pas). Concomitantly, the J 0p + (met) at the metal-bsf interface is significantly lower although the same processing conditions were applied. These experimental results highlight the necessity to further reduce the metallization fraction F M(p) on the emitter region of our n-type cells and to increase the diffusion depth under the metal contact by creating selective emitter structures. The selective emitter approach, however,

96 2 88 Chapter 5: Metallization of bifacial solar cells ] J 0 1 [fa /c m M e ta lliz a tio n fra c tio n o n p +, F M (p ) [% ] p + p + p + s h a llo w, e m itte r a ) d e e p, e m itte r d e e p, B S F J 0 p + (m e t) = fa /c m J 0 p + (m e t) = fa /c m J 0 p + (m e t) = fa /c m n + n + e m itte r B S F b ) ] J 0 1 [fa /c m J 0 n + (m e t) = fa /c m J 0 n + (m e t) = fa /c m M e ta lliz a tio n fra c tio n o n n +, F M (n ) [% ] Figure 5.7: J 01 extracted using 1-diode equation from the experimental IV parameters under illumination (symbols) as a function of metal fraction variation on either p + (F M(p) ) or n + (F M(n) ) diffused regions of the bifacial cells. The J 0(met) is then extracted from the linear fit (solid lines) to the experimental data using equations 5.2 or 5.3. may be difficult to realize since a p + emitter with a depth of >0.55µm still yields a significantly higher J 0(met) value as compared with a similar doping profile but on n + emitters (Fig. 5.7b). The contact recombination losses measured on a 68 Ω/sq. p + emitter of our cells (Fig. 5.7a) is also about a factor of four higher than the literature reported value for an n + emitter of similar (72 Ω/sq.) sheet resistance [114]. A much higher J 0p + (met) as compared with J 0n + (met) demonstrate that, presently, the recombination losses induced by screen printing metallization of n-type cells is limiting the efficiency more than it does for p-type cells Modelling the metallization losses The microscopic details of the metal-si interface formed by screen printing and firing through of Ag paste have been extensively studied [ ]. The Ag-Si contact interface is typically characterized by the formation of Ag crystallites grown into Si, etching of

97 5.3. Results and Discussion 89 Figure 5.8: SEM micrographs of contact area of optimally fired front contacts after sequential removal of silver finger, glass layer and silver imprints. Images of typical n+ contact (ρc 5 mωcm2 ) (left), down typical p+ contact (ρc 10 mωcm2 ) (right). Si by the glass frit during the firing process, and a possible migration of metal impurities into the diffused emitter region [120]. In a parallel investigation on contact formation of both n+ and p+ contacts a SEM analysis was performed. Figure 5.8 shows an example for the contacting behaviour for typical, optimally fired contacts on n+ and p+ surfaces, respectively. It can clearly be seen that, while the n+ contact is characterized by few silver crystallites grown at the pyramid tips, the p+ contact has only few crystallites but significant fractions of the pyramid tips are removed. The relation between these microscopic observations and the observed VOC losses on the finished cells is not yet clearly understood. To gain a better understanding of the mechanism causing the VOC degradation after the metallization step, we used ATLAS, a two-dimensional device simulator from SILVACO, to reproduce the solar cell performance operating under VOC conditions. Herein we computed the performance of our bifacial n-type cells with deep boron diffusion using experimental data for doping profiles, device geometry, and material properties as input parameters. A set of simulations was performed in which the metal contacts were assumed to penetrate into the p+ emitter region at various depths, as an attempt to qualitatively reproduce the experimental observations. In the model we attributed different surface recombination velocities to the metal-si interface (Smet ) and to the passivated area (S pas ) between the metal contacts. The Smet was assumed cm/s (corresponding to an unpassivated surface) Since the presence of metallization does not affect the IQE curves, as shown in Fig. 5.4, the S pas was determined from the numerical model by fitting the IQE curve of the n-type cell at short wavelengths with the front surface recombination velocity as the fit parameter. An S pas of approximately cm/s was obtained. A cross-section view together with a snapshot of the total electron current flow

98 90 Chapter 5: Metallization of bifacial solar cells Figure 5.9: (a) Cross-section view and snapshot of the electron current flow (minority carriers in the emitter region) beneath the metal contact at V OC condition, as taken from our Silvaco Atlas 2-dimensional simulation model. The snapshot was taken for a metal penetration of 0.4 µm into a p + (boron) emitter of 0.6 µm depth. The S met was assumed cm/s around the metal surface whereas a S pas of cm/s was used for the passivated areas between the metal contacts. (b) Magnitude of the electron diffusion J di f f usion, electron drift J dri f t, and total electron current J total directly beneath the metal contact, respectively. (J total =J dri f t +J di f f usion ) beneath the metal contact at V=V OC is shown in Fig In the n-type Si solar cells the electrons (minority charge carriers) photogenerated in the p + emitter should cross the p-n junction into the bulk. Generally, the doping concentration decreases with increasing depth resulting in an electric field component that should drive minority carriers towards and across the junction. However, as observed in Fig. 5.9, under deep metal penetration into the emitter region the electrons diffuse instead towards the metal-si interface where they recombine. The presence of a highly recombinative interface virtually depletes the volume beneath the metal contact from electrons which causes J di f f usion to overcompensate J dri f t and thus to create a leakage current across the junction. With increasing metal penetration into the emitter region this leakage current increases and so does the J 01 of the cell. This is further demonstrated in Fig where

99 5.4. Conclusion 91 several simulation scenarios are presented. In Fig. 5.10a the net loss in V OC is calculated as a function of F M for several metal penetrations and interface conditions. First, without metal penetration and with a passivated contact area (i.e., S met =S pas = cm/s) the simulation correctly predicts no V OC losses and a J 0p + (met) value equal to the J 0p + (pas) (as shown in Fig. 5.10b). Yet when emitter contact areas are unpassivated (i.e., S met = cm/s) and although no metal penetration was assumed the V OC losses become important. The J 0p + (met) of 475 fa/cm 2 for this case is still an order of magnitude lower than the experimental data observed for such a cell (Fig. 5.7). Hence, simply the loss of passivated area beneath the metal contacts cannot explain the magnitude of V OC degradation observed experimentally. However, as the metal contacts penetrate into the emitter region a strong decrease in V OC and a strong increase in J 0p + (met) is observed, culminating with shunting of the cell when the emitter junction is crossed. A qualitative match with the experiment is achieved when contact penetrates about 0.4 µm inside the emitter region, being in close proximity to the p-n junction ( 0.6 µm). A more quantitative modelling of these metallization induced losses is presented elsewhere [121]. Although the geometrical model, as shown in Fig. 5.9, is a crude simplification of the experimental metal-si contact it does capture two important features; a strong local increase in recombination resulting in a leakage current across the junction, and a reduction in the shielding effect of the diffused emitters as the metal penetrates into Si or the Si is etched by the glass frit. With the current metallization methods and available pastes there are two ways to further reduce the detrimental effect on cell efficiency. One is simply to further reduce the metal coverage on the emitter surface, which can be accomplished by e.g., printing floating busbars and thinner fingers which at the same time reduces the use of expensive Ag. The other approach will require the formation of a much deeper junction beneath the contacts using selective emitter technology for p + diffused layers which is at an early stage of development and certainly increases process complexity. 5.4 Conclusion The metallization induced recombination losses on high efficiency bifacial n- and p-type crystalline Si solar cells were investigated in this study. From the experimental data we observed that open-circuit voltage (V OC ) is the foremost efficiency limiting parameter affected by the screen printed metallization. We showed that these recombination losses which limit the V OC can readily be explained by a local enhancement in the dark saturation current density of the cell beneath the metal contacts (J 0(met) ). We found that, under optimum fabrication conditions, the J 0(met) at metal-p + emitter interfaces is significantly higher compared to the values obtained for metal-n + emitters. Using a two-dimensional simulation model, and considering the experimental observations at metal-si interfaces, we were able to get further insight into the recombination mechanism leading to these V OC losses. The model assumes that metal contacts penetrate (or etch) into the diffused region following the firing process and degrade the metal-si interface. The presence of a highly recombinative (unpassivated) interface deep inside the emitter region reduces the shielding effect of the diffused layer and attracts the minority charge carriers in the

100 92 Chapter 5: Metallization of bifacial solar cells V O C - i V O C [ m V ] J 0 1 [ f A / c m ² ] m e t a l p e n e t r a t i o n : 0 µ m, S m e t = S p a s 0 µ m, S m e t > > S p a s µ m, S m e t > > S p a s 0. 4 µ m, S m e t > > S p a s a ) J 0 p + (m e t) = f A / c m 2 = J 0 p + (p a s ) b ) J 0 p + (m e t) = f A / c m 2 J 0 p + (m e t) = f A / c m 2 J 0 p + (m e t) = f A / c m M e t a l l i z a t i o n f r a c t i o n o n p +, F M (p ) [ % ] Figure 5.10: Calculated net loss in V OC (top) and J 01 (bottom) as a function of metallization fraction F M for our n-type solar cells with the deep p + emitter profile. The F M was varied only on p + (emitter) side. On the n + (BSF) side a metal fraction of F M =5% with no penetration and S met =S pas was assumed. All calculated V OC values were corrected using one-diode equation to account for the reduction in J SC as a result of metal shading. The J 0p + (met) values were extracted from the linear fit to the calculated J 01 data using equation 5.2. emitter which recombine upon reaching the unpassivated interface. Under open-circuit conditions, this produces a leakage current across the junction and thus an increase in J 0(met). Applying the model to our n-type solar cells with a boron p + emitter metallized by an aluminum containing Ag paste we demonstrated that the simple loss of passivated area beneath the metal contact cannot explain the degradation observed in the V OC of the cell without considering a significant surface damage, e.g., by metal penetration or surface etching of the emitter region. This study convincingly demonstrates that p + contacting pastes, potentially without aluminium, that do not induce surface damage should therefore be an important development goal, which will enable fabrication of solar cells with much higher V OC and efficiency values.

101 6 Measurement of high efficiency solar cells 6.1 Motivation Solar cells with an open rear-side grid can be assembled in dedicated bifacial modules. The light incident on the transparent back side can additionally couple into the cell and thus increase the energy yield. The additional energy harvest depends on various factors such as the module fabrication method, installation angles and ground reflection. For this purpose the reflectivity of natural surfaces can be evaluated by its albedo factor. Albedo is defined as the ratio between the photon flux that is reflected from an object and the photon flux reflected from a white, lambertian reflector of the same geometrical size. Surfaces with a high fraction of direct reflection, such as water, can accordingly have a low albedo value. Fig. 6.1 gives an overview of albedo values of common outdoor grounds. High hopes are associated with the large scale installation of modules employing bifacial cells in combination with a transparent rear side. Several studies illustrate the potential of this approach [ ]. However, a comprehensive prediction method for the additional energy yield is still missing. The influence of different seasons, latitudes, installation angles and background surfaces make establishing meaningful and versatile measurement conditions a complicated task. Even the measurement of bifacial modules under STC conditions is influenced by the specific make and the measurement background. Whereas a white back-sheet scatters the light incident in between the cells so that it has an additional chance of hitting onto the cell surface due to internal reflection, a transparent back-sheet simply transmits this light. This reduces the generated Figure 6.1: Ground reflectance of different natural surfaces [122]. 93

102 94 Chapter 6: Measurement of high efficiency solar cells current of the module and therefore the rated power output, giving the bifacial module an apparent disadvantage in an STC measurement. The intriguing point is that the same module can outperform the one with the white back-sheet in terms of energy yield, when installed in a dedicated bifacial application. This includes practically any installation in the field or on a flat roof, given sufficient spacing on the rear side and a bright background. From the albedo chart one can see that even sand or soil can reflect around 30% of the light, a value that is easily boosted by employing white gravel or white paint on a flat roof. Aside from traditional field or roof installation a bifacial module could also be installed standing upright with the light harvesting sides facing directly east and west. This approach could even have a beneficial impact on the energy grid stability. The east/west oriented modules would have peak energy production in the morning and afternoon hours, thus broadening the electricity peak that occurs at midday in electricity grids with high photovoltaic energy penetration. Due to the aforementioned embedding losses which are higher for transparent, compared to white back-sheets the net energy gain has to be made quantifiable before an installer can be convinced of the principles of bifaciality. In order to build such modules in the first place accurate IV characterization of bifacial solar cells is required. Here the same complications as outlined for the module measurements do apply. The present chapter deals with the measurement of the IV characteristics of high efficiency bifacial cells using flash IV testers. It is divided into three parts. At first the uncertainties in the measurement of bifacial cells in conventional setups are described. The second section deals with problems arising from fast IV measurements of modern high efficiency cells. It was noticed that capacitive effects can be a significant source of error in fast IV measurements of n-type cells. Special measurement procedures, as presented, can be required to record the IV curve of such cells without measurement artefacts. The remainder of this chapter deals with the measurement of bifacial solar cells under both sided illumination. We have built a mounting system for a flash tester that enables the controlled illumination of bifacial cells from both sides. The cell performance of bifacial cells under both sided illumination can therefore be tested. 6.2 Measurement uncertainties of bifacial solar cells The open rear side grid of bifacial solar cells allows for a partial transmission of light in the long wavelength range. Accordingly the reflectivity of the measurement background can lead to overestimation of the current collection of such a bifacial device. IV data reported from manufacturers and research institutes is therefore difficult to compare. For research purposes most often steady state IV testers are used, which exhibit a more homogeneous light intensity distribution both over the measurement area and over time and higher spectral stability over time. In these setups the samples are often mounted on temperature controlled, conductive chucks. Hohl-Ebinger et al. [126] have pointed out the influence of the reflectivity of the measurement chuck on the measured current. They showed that commonly used materials, such as brass, bare copper or gold coated copper can easily lead to an overestimation in the range of 0.5 1% in the measured short circuit current. Naturally the degree of the error depends greatly on the metallization fraction of the cell rear side. A conductive chuck can also influence the fill factor evaluation since the contribution from the line resistance to the total resistance is eliminated for a fully

103 6.2. Measurement uncertainties of bifacial solar cells 95 Measurement I SC J SC V OC FF η P MPP [A] [ma/cm 2 ] [mv] [%] [%] [W] front, black background front, white background rear, black background Table 6.1: Influence of back-sheet glued to the back of the bifacial cell during IV measurements. contacted rear side. We have evaluated these uncertainties when measuring bifacial cells in our setups. The IV characterization of fabricated solar cells was performed on various IV setups at ISC-Konstanz and BOSCH Solar Energy AG. The majority of measurements was taken by the flash lamp IV tester PSS 10 II from BERGER Lichttechnik. The system uses a halogen flash lamp, with AM1.5G filter and no additional collimating optics. This system can emit flashes of a plateau length of up to 5 ms with constant intensity of 1000 W/m 2. An additional intensity plateau of 500 W/m 2 can be triggered for series resistance evaluation in compliance with DIN EN The spectrum is well within the specification for class A according to DIN EN on a measurement spot of 30 cm diameter at 1.15 m away from the light source. The IV curve is recorded within this period by switching of a passive electronic load to set the bias voltage at the cell. A nonlinear voltage curve ensures an equal distribution of data points along the curve. The external current is measured by the load and corrected for intensity fluctuations as measured by a small silicon reference cell. The current and voltage are also corrected for temperature deviations from STC conditions (25 C). The temperature measurement is normally done by IR sensing of the cell temperature on the rear side, while the IR sensor is calibrated for aluminium BSF cells. The error in the temperature reading of cells with an open rear side grid is however marginal at room temperature. In the contacting frame the sample is suspended freely between rows of spring loaded contact pins. The contact pins of a contact bar are alternating current and voltage pins connected in a 4-probe measurement circuit to eliminate the feed cable contribution to the series resistance. The sample holder can be adapted for different conventional busbar geometries. In any case the sample is suspended in mid-air and the rear surface is not covered. The configuration underneath the cell can have a great impact as is demonstrated in Table 6.1. A more advanced tester was established at ISC-Konstanz during the course of this work. The IV tester from h.a.l.m. was of comparable composition with several key changes, the most important one being the option for extended measurement times. The plateau length of this IV tester can be extended to 100 ms at 1000 W/m 2 intensity. This is strongly required to reduce the hysteresis error in the IV curve of n-type cells. The active load control further allows for direct selection of the measured voltage interval, in any desired step sequence. All cell testers get calibrated with cells of the same type according to a certified measurement from Fraunhofer ISE Callab, prior to each measurement. German version of IEC

104 96 Chapter 6: Measurement of high efficiency solar cells 6.3 Flasher measurements of high efficiency n-type cells Introduction In a production environment the end of line characterization of solar cells plays an important role for manufacturers in order to sort the cells according to expected P MPP power output. As the cycle times are in the range of seconds this is commonly achieved by flasher IV testers, which scan the entire IV curve of a cell within few milliseconds. This means applying fast voltage changes to the cell, which change the equilibrium charge distribution in the cell drastically. Since modern cell concepts tend to have higher open circuit voltages, it requires longer times to reach equilibrium condition. This is especially important for voltages exceeding V MPP. It is this inert behaviour that leads to the known effects of IV curve distortion and the corresponding fill factor under- or overestimation. We have investigated the influence of the charging or discharging currents respectively, in order to design a measurement procedure, which can circumvent this problem. The continuing development in cell technology has allowed for technical advances such as selective emitters, p-perc or n-type cell concepts to take the leap from R&D into production. Many of these cell types include the use of rear side dielectrics which ultimately increases the operating voltage of the cell. Since these advances aggravate the difficulties in correctly measuring these cell types we are convinced that new measurement procedures have to be implemented for the inline characterization of such cells Observation The FF deviation of several common (Cz and mc p-type, Alu-BSF) and also advanced cell types (p-type PERC, n-type Rear-alu-emitter or n-pert type) have been compared. It is known and confirmed here, that the measured value of P MPP and therefore the FF can be influenced by the direction of the IV sweep. A scan of the IV curve beginning at I SC (hereafter called SC-OC) can result in an underestimation of P MPP, whereas the sweep in the opposite direction (hereafter OC-SC) yields an overestimation of P MPP [ ]. The influence on V OC is a lot smaller in size. It is also demonstrated that the capacitance effect depends on the intensity of illumination which means that the determination of the series resistance, which is evaluated from the slopes around P MPP for a 1000 W/m 2 and 500 W/m 2 intensity curve can also be erroneous. Fig. 6.2 shows the influence of the measurement direction on IV curve of an n-pert type solar cell. The evaluated FF for these two measurements varies between 75.3% and 80.1% depending on the measurement direction. The IV-curve distortion further shows an asymmetric shape whereas the distortion of the OC-SC curve is always more pronounced. Therefore a correction by averaging of two measurements of opposing directions does not automatically yield the correct IV curve. Fig. 6.3 shows a series of measurements with a 6 Ω cm n-type cell using a h.a.l.m. flasher system with adjustable flash pulse length and linear voltage ramp. Extending the pulse length obviously reduces the induced errors, but even at 100 ms measurement time the spread of the FF evaluated for the two measurement directions is still 0.5% abs. In this time regime the incident flash power starts to increase the cell temperature markedly, which puts an effective limit on

105 6.3. Flasher measurements of high efficiency n-type cells C u rre n t [A ] IV s w e e p d ire c tio n O C -S C IV s w e e p d ire c tio n S C -O C V o lta g e [V ] Figure 6.2: IV curves of n-pert cell (Cz, 6 Ωcm) scanned from short circuit current to open circuit (SC-OC) voltage and vice versa. the measurement duration for systems without active temperature control. Also it can be assumed that extending the flash duration comes at the cost of increased variation in the lamp spectrum over time and potentially reduced flash lamp lifetime. 8 0 O C -S C s w e e p S C -O C s w e e p m e a n 7 9 S te a d y S ta te F F [% ] S w e e p tim e [m s ] Figure 6.3: FF evaluation for varying measurement time and direction, steady state value (solid) is framed with 0.5 % error margin (dashed) Explanation The origin of these effects lies in the nature of a diode operating under forward bias. The equilibrium level of minority charge carriers, in the bulk of the cell, increases exponentially with increasing applied bias voltage. The IV sweep corresponds to a step by step increase (or decrease for measurements from OC to SC) in bias voltage. After each step charge carriers, which are either injected through the junction or generated near the

106 98 Chapter 6: Measurement of high efficiency solar cells M i n o r i t y c a r r i e r c o n c e n t r a t i o n [ c m ³ ] J u n c tio n v o lta g e 0 V V V V D is ta n c e fro m th e fro n t [ m ] Figure 6.4: Simulated minority charge carrier distribution in the bulk of a solar cell with rear side passivation under illumination for different forward bias voltages illuminated surface, have to redistribute to reach a new equilibrium distribution. This allocation of large amounts of charge carriers can be seen as a charging current of the differential diffusion capacitance which can also be expressed as: C di f f usion = q kt qn 2 i N D L P } {{ } C 0 exp( qu kt ) (6.1) Herein U denotes the applied forward bias voltage, N D the base doping density and L p the minority carrier diffusion length (both are here indexed for an n-type base). As can be deducted from the PC1D simulation result in Fig. 6.4, this capacitance increases rapidly for operating voltages which exceed V MPP values of modern solar cells. The charge carriers which contribute to the charging current cannot contribute to the externally measured current, which explains the observed underestimation of the measured current. This diffusion capacitance is not to be mistaken with the junction or depletion layer capacitance, which is made up by charge carriers located in the space charge region and is negligible in forward bias condition: C depletion = Q U = qɛ s N A N D 1 2 N A + N D U bi U Fig. 6.5 shows the calculated diffusion capacitance for n- and p-type cells for a range of production relevant resistivity values and using long diffusion lengths as required for advanced cell processes. The offset between the n- and p-type chart is mainly caused by the lower mobility of holes compared to electrons. Also the higher resistivity range and longer diffusion lengths make n-type silicon more prone to capacitive effects. However advanced p-type cell concepts can be equally affected. A quick estimation can be given when the cell is pictured as an RC-circuit made up by its diffusion capacitance and its series resistance. A characteristic time constant for this circuit would then be: (6.2)

107 6.3. Flasher measurements of high efficiency n-type cells 99 R e s is tiv ity [ c m ] J u n c t i o n v o l t a g e [ m V ] F / c m ² F / c m ² J u n c t i o n v o l t a g e [ m V ] R e s i s t i v i t y [ c m ] Figure 6.5: Simulated diffusion capacitance in µf/cm 2 for p-type base (left) and n-type base (right) of 150µm thick solar cell, common V MPP and resistivity range indicated. τ = R series C di f f usion C Di f f represents the difference in diffusion capacitance between two voltage steps. The time constant denotes the time required to charge or discharge the capacitor to 2/3 of its capacity. This allows to give an estimate as to the required time per measurement interval. Using the voltage ramp as applied during a flasher measurement we can calculate the measurement interval per data point as 50µs. For a 6 inch n-type cell with V MPP of 540mV and R = 3mΩ the value of τ is calculated to be around 30µs. Since the characteristic time required to reach the equilibrium condition increases sharply with voltage, the given measurement time per interval is clearly not sufficient. Obviously the equilibrium distribution is never reached for voltages exceeding a critical voltage during a fast measurement. As soon as the associated time constant is bigger than the available measurement interval, the charging currents will add up for each following voltage step and accordingly the external current reading will be faulty. Therefore the measurement error at one point of the IV curve is always a result of the measurement history. This can explain why the under- or overestimation of the current can be in the range of amperes, which is the size of the measurement error. This hysteresis effect can explain the strong asymmetry which is found for very short measurement times. A modelling approach to correct for this error mathematically requires the knowledge of cell parameters such as doping, diffusion lengths and resistances and is therefore unsuitable for an inline application Maximum power point scan We have seen that n-type cells, especially with high resistivity base material and high V OC, induce the most severe IV distortion with in FF between SC-OC and OC-SC sweep direction of up to 5% abs. Also advanced p-type cell concepts can show FF of up to 2% abs measured with high sweep rates. It is therefore always recommended to use the maximum flash duration available by the hardware for an IV measurement. We have also

108 100 Chapter 6: Measurement of high efficiency solar cells seen that methods such as averaging of P MPP of two opposing measurements or a reduction of measurement points can only reduce the error to around 0.5% abs. Therefore we suggest a method that traces a narrow interval around P MPP during a single flash. Since the measurement time is thereby artificially extended a charge distribution according to the voltage and illumination dependent equilibrium concentration can be achieved. This minimizes the measurement error and the P MPP is again extracted with high accuracy. As the measurement with reversed sweep direction traces the same power curve, we know that the measurement is not distorted. An additional measurement could give an accurate evaluation of I SC and V OC. For R&D applications this can easily be extended to record the full curve as a sequence of individual IV scans yielding even higher accuracy. For IV testing in a production environment it might be suitable to test solely for P MPP performance, aside from the reverse testing which can take place in the dark and therefore always be at equilibrium. MPP Scan results As shown for a high resistivity n-type sample in Fig. 6.6, the P MPP scan measurement is not affected by capacitance induced distortion, although the full IV scan of this sample showed a pronounced effect. This example demonstrates the potential of the approach since samples with lower diffusion capacitance can easily be corrected as well. An overview of the results for various common cell types measured with the new approach at the BERGER flasher system is shown in Table 6.2. All cells in this overview can be measured with high accuracy. C u rre n t [A ] S C -O C M P P s w e e p O C -S C M P P s w e e p S C -O C fu ll s w e e p O C -S C fu ll s w e e p V o lta g e [m V ] V o lta g e [m V ] P o w e r [W ] Figure 6.6: Detail of IV curves (left) and associated power curves (right), showing a comparison between full IV sweep and P MPP scan Conclusion We have demonstrated the influence of the diffusion capacitance on various cell designs on the IV measurement using standard measurement equipment. It is found that charging or discharging currents can significantly distort the IV-curve for many advanced cell types. In the case of limited flash pulse durations a full IV curve can not always be

109 6.3. Flasher measurements of high efficiency n-type cells 101 full IV sweep P MPP Scan Cell Sweep ρ P MPP FF V OC P MPP FF type direction [Ωcm] [W] [%] [mv] [W] [%] Cz p-type SC-OC Alu BSF OC-SC mc p-type SC-OC Alu BSF OC-SC Cz p-type SC-OC PERC OC-SC Cz n-type SC-OC PERT OC-SC Cz n-type SC-OC PERT OC-SC Cz n-type SC-OC AluRearEm OC-SC Cz n-type SC-OC IBC(ZEBRA) OC-SC Table 6.2: IV data for various n-type and p-type cell architectures measured with full IV sweep and P MPP scan technique. recorded without distortion. Any silicon solar cell or module exhibits the hysteresis effects during fast IV measurements. Therefore when measuring a solar cell it is highly recommended to test for hysteresis effects, which can easily be done by reversing the sweep direction. The deviation of the FF value is a good indicator whether the measurements are distorted and should stay within the error limits. The problem is usually caused by insufficient measurement time, a limitation which is an inherent problem for flash IV testers. Nevertheless distortions can also be observed on constant illumination IV testers if the actual measurement time is kept short (e.g. to avoid heating of the cell). Special attention has to be paid when measuring cells or modules with either one or more of the following characteristics: n-type silicon base High resistivity base Rear emitter architecture High operating voltage, V MPP 520mV If a significant deviation between the two measurements is found, it is advisable to extend the measurement time in the high voltage regime. Therefore it is always recommended to choose the maximum measurement time of the hardware. Also the voltage ramp, i.e. the distribution of data points along the curve, can be adjusted so that more time is available in the high voltage regime. As a next step FF deviations in the order of 1% abs can be reliably eliminated by averaging of the two hysteresis curves. For larger

110 102 Chapter 6: Measurement of high efficiency solar cells deviations, as might occur in modern industrial silicon solar cells, this procedure is not recommended because of the asymmetric behaviour of the distortion. If the measurement time can not be extended for such examples, different measurement procedures to extract the important solar cell parameters will be required. For this reason we have tested an approach to measure only a part of the IV curve in the given time to artificially extend the measurement time per data point. It was demonstrated that this approach allows to measure an interval around P MPP with high accuracy and no observable distortion. Obviously this method can be used to assemble the entire IV curve from a series of separate measurements. 6.4 Bifacial measurements of solar cells Motivation At the end of a solar cell production line the finished cells are subject to extensive IV characterization and are sorted according to their performance. This is an essential part of the quality control since shunted, low performing or optically imperfect cells are sorted out. Sorting is usually done with a flasher IV tester due to the required cycle times. The performance measurement for bifacial cells however might require new sorting algorithms since the cell performance under bifacial illumination can not easily be estimated just from two separate single side measurements. While the current generation should follow the light intensity linearly, injection dependent recombination properties could cause nonlinearities which lead to deviations from the linear curves [131]. Also the series resistance losses do not vary linearly with the incident light intensity, but are proportional to I 2 SC Experimental setup We have assembled a bifacial measurement rack which is inserted into an existing IV measurement environment. The system of choice was the BERGER IV tester described earlier. Besides the obvious advantages of using the IV flasher periphery, it offers easy alignment and synchronicity since only a single light source is used. A measurement of bifacial cells with illumination from both sides simultaneously has to be finished within a tenth of a second since the cell heats up quickly. Flasher IV testers are therefore especially suited, but constant illumination light sources are imaginable if ultra fast mechanical shutters are applied. The contacting frame of the flasher has been adopted without changes to contact the samples. The whole frame can be mounted into a rack with two aluminium coated mirrors as seen in Fig Similar setups have been presented by Ezquer et al. [132] and Ohtsuka et al. [131]. The setup is illuminated from the top so that the mirrors, angled at 45 to the sample plane, reflect the light directly onto the sample. The effective distance between sample (and reference cell, which is in the same plane) and light source is 1.15m, the same distance as for the conventional arrangement. The angular distribution of the incident light is therefore unchanged. During the calibration procedure we fully cover the rear side of the cell and the mirror behind the cell with black paper. In this configuration the same IV results as in a standard monofacial configuration

111 6.4. Bifacial measurements of solar cells 103 Figure 6.7: Top and side view of the bifacial cell tester. Figure 6.8: Schematic of a measurement setup taken from Ref. [132] for both sided illumination of solar cells. are obtained. We wanted to comply as close as possible with the STC requirements. The cell is therefore kept at room temperature which is close to 25 C and the cell temperature is additionally measured by a PT-100 sensor on one busbar to correct for deviations. The lamp spectrum and the reflected spectrum were characterized with a spectrophotometer in the plane of measurement. The deviations from the AM1.5G spectrum induced by the mirrors were found to be well within the specifications for a class A tester. The requirements for the homogeneity of illumination were met for a class B device. The latter is the result of the larger effective measurement area as seen in the top view 6.7. The adjustment of the intensity on the rear side is achieved with meshes or perforated plates of different density or combinations thereof as tabulated in Table 6.3. This data was recorded by measuring the response of a reference cell illuminated with 1 sun intensity from either side while the rear side is covered with different configurations of meshes. The electronic load limits the current measurement currently to ± 16A, which is why the maximum illumination intensity for large area cells was 1476 W/cm2. Comparing with different albedo values this is sufficient to cover the range of potential irradiance conditions of a bifacial cell in a module. IEC 60904, german version DIN EN 60981

112 104 Chapter 6: Measurement of high efficiency solar cells Mesh Transmission Calc. Intensity Meas. Intensity [W/cm 2 ] [W/cm 2 ] T+0.5T T+0.8T T T+0.8T T Table 6.3: Intensity steps with meshes of different density or combinations thereof. Cumulative values including 1000 W/cm 2 intensity on the front side Experimental evaluation of bifacial properties A set of solar cells has been measured in the bifacial setup to evaluate their performance under true bifacial operating conditions. All measurements were performed with a front illumination intensity of 1000 W/cm 2, while the rear side was illuminated with varying intensity between 0 and 476 W/cm 2 in levels shown in Table 6.3. For our measurements we used single side textured cells with 2 and 5 Ωcm similar to the ones presented in 4.1, fabricated at ISC-Konstanz. We compared them to double side textured cells with 4 Ωcm of the same structure, fabricated at BOSCH Solar Energy AG. The cell architecture plays a big role for the bifacial performance and it is difficult to evaluate the polished rear side approach compared to a cell with a textured rear side, since the recombination, optical and contact properties are changed. The net differences become immediately obvious however, when comparing the current collection in a bifacial measurement with increasing rear illumination fraction. All data has been recorded using the MPP scan technique as explained in Section 6.3.4, since the hysteresis errors depend significantly on the illumination intensity. We have seen that the efficiency ratio f bi f acial = η rear /η f ront between the front and rear side efficiency of the ISC cells (compare Table 6.1) is around This is primarily due to the polished rear side of the cell and a more dense rear metal grid. It can be seen from IQE measurements that the collection efficiency is almost identical from either side. The cell could therefore be optimized to yield the same single side efficiency no matter which side is illuminated. In order to achieve an efficiency ratio f bi f acial close to one a textured rear surface and similar metallization grids on either side have to be applied. This is currently not implemented in our baseline process and could only be achieved at the cost of front side efficiency. From the data in Fig. 6.9 we see immediately that the cells with textured rear side gain significantly more in current with additional rear illumination compared to the polished rear side cells. This in turn leads to higher FF losses. The metallization pattern of the cells was optimized for 1 sun intensity and needs to be adapted if dedicated bifacial installations are planned. Although the gap in the current widens, the FF losses for textured cells reduce the gap between both side textured and polished cells in the power output. The FF curve also shows the expected linear trend with increasing current. In the tested illumination range (down to 500 W/cm 2 ) we do not find any evidence of nonlinear behaviour for either cell structure. In Fig. 6.9 we have cited the cell efficiency in the lower chart, but also annotated

113 6.4. Bifacial measurements of solar cells 105 the 1 sun equivalent efficiency in the power chart. The former is calculated using the cumulated intensity incident on front and rear side. The equivalent η 1sun is calculated based on STC intensity of 1000 W/cm 2. Therefore this corresponds to the efficiency value of a cell in monofacial operation required to yield the same power output as the bifacial cell in bifacial operation. It is not a common physical parameter, but is often mistaken as one and confused with the true efficiency of the device. We mention it because it is indeed useful to demonstrate the benefit of bifaciality. I S C [A ] re a r te x. re a r te x. re a r p o l., 2 c m re a r p o l., 5 c m lin e a r fits T o ta l in te n s ity [S u n s ] re a r te x. re a r te x. re a r p o l., 2 c m re a r p o l., 5 c m T o ta l in te n s ity [S u n s ] V O C [m V ] F F [% ] re a r te x. re a r te x. re a r p o l., 2 c m re a r p o l., 5 c m T o ta l in te n s ity [S u n s ] P M P P [W ] re a r te x. re a r te x. re a r p o l., 2 c m re a r p o l., 5 c m re a r te x. re a r te x. re a r p o l., 2 c m re a r p o l., 5 c m T o ta l in te n s ity [S u n s ] T o ta l in te n s ity [S u n s ] Figure 6.9: IV data taken with bifacial measurement setup for two types of cell architectures, with polished rear side (green) or textured rear side (blue) and different resistivities e q u iv a le n t 1 S u n [% ] [% ]

114 106 Chapter 6: Measurement of high efficiency solar cells Conclusion Up to now there is no widely accepted measurement standard for the measurement of bifacial solar cells. When cell manufacturers evaluate their cells at the end of a production line, they follow STC as close as possible to minimize the measurement uncertainties. However, since STC does not yet encompass bifacial cells, the uncertainties are enlarged or the cell performance is even underestimated, e.g. when the cell is measured over a dark background or chuck. We believe that on cell level and on module level BSTC, or Bifacial Standard Testing Conditions, need to be specified, to ensure the correct evaluation and comparability for such cells. For the cell level a procedure was proposed by Jan Lossen from BOSCH Solar Energy in his talk at the Bifacial Workshop in Konstanz (April 2012). The suggested procedure involves a calibration with the cell current which is measured on a high reflective chuck. This resembles the optical configuration in a monofacial module with a white back-sheet. The FF evaluation always has to be done on a non-conductive chuck. This might pose an agreeable proposition for bifacial cells which are to be used in conventional modules but it is certainly far from capturing the real bifacial potential. It is a pity to see the wasted potential of assembling bifacial cells in modules with non-transparent, especially black back-sheets especially when thinking about configurations where no changes in the installation of the modules would be required. The presented setup allows to measure bifacial cells under specified illumination from both sides simultaneously which enables us to evaluate the performance of bifacial cells under true designated operating conditions.

115 7 Summary and Outlook With worldwide photovoltaic installations exceeding a cumulative rated power output of 100 GW [5] solar power is becoming a mainstream source of electricity. This was made possible by cost reductions due to a continuous stream of innovations transitioning from research laboratories to production facilities and advances in manufacturing technology. In the effort to reduce the cost of photovoltaic solar power further, the importance of achieving high efficiencies can not be stressed enough. The ingredients for making silicon solar cells highly efficient are long known and accordingly the records in conversion efficiency for PERT or PERL type laboratory devices have been established decades ago. They all share several common features. The base material has very high diffusion lengths, dopant diffusions result in low surface concentration and accordingly superb passivation quality of the emitter surface, as well as surfaces of opposing polarity is achieved. The required metallization features are sharply defined and induce negligible surface damage. Ever since, it has been the goal of the industry to close the gap between the efficiency of these champion cells and the average production efficiency. This implies that the aforementioned characteristics of highly efficient solar cells have to be accomplished with industrially viable methods. In the zoo of silicon solar cells concepts, such as p-perc, p- or n-mwt, n-ibc or n-pert cells, are competing for the succession of the industry workhorse, the p-type aluminium-bsf cell. In this work we have looked at the n-pert concept and investigated to what extent the different high efficiency requirements can be achieved. At first the bulk lifetime stability of industrial Cz wafers with respect to high temperature treatments, and especially boron tube diffusions was confirmed. The available n-type Cz wafers exhibited diffusion lengths of over 1000µm, after high temperature processing which is significantly higher than in comparable p-type material and therefore suitable for high efficiency processes. The boron tube diffusion was employed to form homogeneous p + doped emitter layers. Using a balanced supply of BBr 3 and O 2 the boron surface concentration can be kept low enough to prevent the formation of a boron rich layer and high enough to prevent emitter surface depletion. Both are prerequisites to achieving high carrier lifetime in the emitter. The requirements of achieving homogeneous sheet resistance and even boron glass thickness can be met by tube diffusion, although the limitations of the single inlet/ single exhaust reactor design become apparent. The passivation of p + layers is usually realized with SiO 2 or today Al 2 O 3 based stacks. A new passivation method based on the boron silicate glass has been suggested and is evaluated here. Using this approach emitter saturation current densities as low as 29 ± 3 fa/cm 2 on polished and 53 ± 7 fa/cm 2 on textured surfaces with 70 Ω/sq. emitters have been measured, which is in the same range as state-of-the-art techniques. The achievable implied V OC for such a structure is up to 690 mv or 685 mv, respectively. 107

116 108 Chapter 7: Summary and Outlook In the chapter on cell processes the scope is widened to the full device. This means conciliating the superior wafer and emitter quality with the requirements of forming a full area BSF diffusion and metal contacts. It was demonstrated that minimizing the recombination in the BSF by reducing the doping concentration is paramount to maintaining the high V OC potential, while a high surface concentration is required to facilitate contact formation. This balance has to be continually adjusted with the development of better contacting pastes. A significant gap of over 30 mv between implied V OC and the V OC of the metallized cell has been found. In order to understand this, we have applied a method using varying metal fractions on either side to distinguish and quantify the metal induced recombination at the emitter and BSF contacts. The main reason for this loss is found to be the surface damage of firing through metallization, especially from aluminium containing silver pastes at p + emitter contacts. The local saturation current density J 0e(met) at the p + emitter contact was found to be around 4000 fa/cm 2, which is about 4 times higher than the recombination at a silver contact to a n + emitter of comparable doping. A simplified model of the contact cross section was implemented in the Silvaco 2D simulation suite to explain the losses qualitatively. Returning to the comparison with the early champion cells the present situation can therefore be summarized as follows. A cell precursor enabling very high V OC is achieved on 6 inch Cz wafers, however, what determines the efficiency today, is the detrimental impact of the screenprinting metallization. Reducing the metal coverage, e.g. by printing narrow lines with high aspect ratio or floating busbars, becomes all the more important. Since screen-printing of aluminium containing silver pastes is expected to remain the dominant contacting method for the industry, the shielding of metal contacts by deep emitter profiles was investigated. This can be realized by the concept of selective emitter doping. Different methods of selective p + emitter formation are suggested and evaluated here. It is found that approaches, which are well-matched for selective phosphorous emitter doping, such as emitter etch-back or doping paste, are inherently unsuitable for preparing selective boron emitters. Currently laser diffusion from highly doped layers is the only approach to yield viable results. An entirely different approach to circumvent the limitations due to metal contact recombination would be the replacement of conventional aluminium containing silver paste. This is evaluated in a new metallization scheme where local all-aluminium alloyed contacts are formed through a diffused boron emitter layer. For this cell structure the emitter is placed on the rear side and the passivating dielectric layer is opened by laser ablation. Promising first results have been achieved with this structure, especially since the local p + contact recombination is reduced to less than 1000 fa/cm 2 using this approach. The n-pert structure with front side emitter is a bifacial cell. This property, common to many n-type cell concepts, is rarely exploited because of the difficulty to quantify the additional power output. A measurement system, that fully captures the bifacial performance of such cells by illuminating both sides simultaneously, is presented. This can show the effective power output of bifacial cells to estimate the additional energy yield in dedicated bifacial modules. The bifaciality induces measurement uncertainties, due to influence of the reflectivity of the measurement background, which are not covered by standard testing conditions. Another source of error is the distortion of the IV curve due to capacitive effects occurring during fast IV measurements. Methods to verify this influence, and to suppress it

117 wherever the measurement duration is not sufficient, are discussed. As a result of the cell development an efficiency of 20.14% was achieved on front emitter n-pert cells measured with negligible rear side contribution, while the best rear emitter cell showed already 19.53%. This is a notable result since neither the FSF doping nor the rear contact pitch are optimized for this structure yet. Printing of finer metal structures is a crucial task for further optimization, which will impact on both J SC and V OC, as well as reduce the silver consumption. It has to be shown whether the required homogeneity of boron emitter sheet resistance can be maintained in volume production. Alternatives, such as the predeposition of doping precursors outside the tube by APCVD or solution printing, can be a anticipated. In the long run modifications to a tube design towards multiple gas inlets could help to improve the homogeneity of sheet resistance and glass layer. For either cell concept it is an interesting approach to employ ion implantation for the phosphorous surface doping. Since ion implantation is a single side process, this removes the necessity to etch or protect the opposite surface from cross doping and thus decouples the two doping processes. Moreover the selective BSF/(FSF) approach is promising to reduce the overall dopant amount in the cell. This is desired, as shown, to enhance the V OC potential while the selectivity ensures a low ohmic contact. This can readily be achieved by adding a single laser step. Last but not least the concept of bifaciality has to be promoted to live up to its expectations. This requires adjustments to standard testing conditions and studies that show the additional energy yield for bifacial modules for a wide range of installations. 109

118

119 Zusammenfassung Mittlerweile hat die weltweit installierte Photovoltaikleistung die Marke von 100 GW überschritten und es ist unbestreitbar, dass Solarstrom eine tragende Säule der Energieversorgung der Zukunft sein wird. Die kontinuierliche Kostensenkung in der Photovoltaik, die dies ermöglicht hat, ist zurückzuführen auf die industrielle Adaption von Forschungsergebnissen aus den Labors sowie auf Lerneffekte in der großindustriellen Fertigung. Um die Kosten in Zukunft weiter zu senken, ist Effizienzsteigerung nach wie vor von oberster Priorität. Die grundsätzlichen Solarzelleigenschaften, die dafür nötig sind, sind seit langem bekannt. Hocheffizienzkonzepte, wie PERL oder PERT Zellen, nutzen zum einen eine Basis aus hochwertigem Silizium mit sehr hohen Diffusionslängen, außerdem zeichnen sich die Emitterdiffusionen durch niedrige Oberflächenkonzentrationen aus. Der Emitter und Oberflächen entgegengesetzter Polarität sind daher exzellent passivierbar. Die notwendigen Metallkontakte sind klar strukturiert und zeigen nur einen geringen Oberflächenangriff. Das Ziel industrieller Forschung ist es daher seit langem diese Eigenschaften mit kostengünstigen, also industriell umsetzbaren Methoden zu erzeugen. Um das zu erreichen, stehen im Moment verschiedene Zellkonzepte, wie PERC, IBC, MWT oder PERT, in den Startlöchern, um die industrielle Standard-p- Typ Solarzelle abzulösen. In dieser Arbeit haben wir uns eingehend mit der Herstellung von n-basierten PERT Zellen mit industriellen Fertigungsmethoden beschäftigt. Zuerst wurden auf den verwendeten n-typ Wafern stabile Minoritätslebensdauern nachgewiesen, nachdem diese mit dem Zellprozess vergleichbaren Hochtemperaturschritten ausgesetzt wurden. Insbesondere wurde nachgewiesen, dass die Bordiffusion nicht notwendigerweise mit Lebensdauerverlusten einhergehen muss, wenn die Bildung einer borreichen Schicht (BRL) vermieden werden kann. Mit Diffusionslängen über 1000 µm eignen sich somit auch kostengünstige Cz n-typ Wafer für die Fertigung von Hocheffizienzzellen. Zur Erzeugung von homogenen Boremittern wurde weitgehend die Rohrdiffusion aus einer BBr 3 Quelle genutzt. Dabei ist durch ein ausgeglichenes Verhältnis der BBr 3 und O 2 Gaszufuhr darauf zu achten, dass die Bildung einer schädlichen borreichen Schicht vermieden wird. Gleichzeitig muss die Borkonzentration im Borsilikatglas aber hoch genug bleiben, um das Absinken der Oberflächenkonzentration zu verhindern. Beides sind Eigenschaften, die einen industriellen Hocheffizienzboremitter auszeichnen. Es konnte außerdem im Kleinversuch gezeigt werden, dass die erforderliche Homogenität sowohl im Schichtwiderstand, als auch in der Schichtdicke des Borglases mit konventionellen Rohrdiffusionsöfen gewährleistet werden kann. Allerdings wurde deutlich, dass bei der gleichzeitigen Optimierung von Schichtwiderstandshomogenität und Glasdicke Zielkonflikte entstehen. Dies ist vermutlich eine Limitierung aufgrund der Konstruktion des Rohrofens mit je nur einem Gasein- und Auslass. Zur 111

120 112 Chapter 7: Summary and Outlook Passivierung von hochdotierten p + -Schichten kommen heute meist Schichtstapel auf Basis von thermischem SiO 2 oder abgeschiedenem Al 2 O 3 zum Einsatz. Es wurde gezeigt, dass auch das Borsilikatglas als Passivierung genutzt werden kann und diese Schicht wurde diesbezüglich umfassend charakterisiert. Mit diesem Ansatz wurden Emittersättigungsstromdichten J 0e von 29 ± 3 fa/cm 2 auf polierten und 53 ± 7 fa/cm 2 auf texturierten Oberflächen mit 70 Ω/sq. Emitter gemessen. Diese Ergebnisse zeigen, dass mit BSG/SiN x Schichten eine gute Passivierung erzielt werden kann und sie somit eine kostengünstige Alternative zu aktuellen anderen Passivierungsmethoden darstellt. Mit dieser Passivierung wurden implizierte Spannungen von 690 mv auf polierten, respektive 685 mv auf texturierten, beidseitig diffundierten Teststrukturen erreicht. Im Zellprozess kommen dann allerdings noch Rekombinationsbeiträge der vollflächigen BSF Diffusion sowie der Metallkontakte zum Tragen. Die hohe rückseitige Dotierung limitiert dabei das V OC Potential, ist aber gleichzeitig für die Ausbildung des Rückseitenkontakts notwendig. Dadurch ergibt sich ein Optimum, das sich mit fortschreitender Pastenentwicklung immer weiter hin zu geringer Dotierung verschieben wird. Ein über 30 mv geringeres Zell V OC verglichen mit dem gemessenen V OC Potential auf unmetallisierten Zellvorläufern deutet auf starke Rekombination an den Metallkontakten hin. In einem umfassenden Experiment wurden daher Zellen mit unterschiedlichem Metallisierungsgrad auf jeder Seite gefertigt, um zwischen Vorder- und Rückseiteneinfluss unterscheiden zu können und diese Verluste zu quantifizieren. Die Hauptursache wurde in der Oberflächenschädigung der p + Emitterschicht durch durchfeuernde, aluminiumhaltige Silberpasten gefunden. Die lokale Sättigungsstromdichte J 0p + (met) an diesen Kontakten wurde auf 4000 fa/cm 2 bestimmt, was etwa dem Vierfachen des bekannten Werts für n + Kontakte mit vergleichbarer Dotierung entspricht. Der Verlustmechanismus konnte in einem vereinfachten Modell in einer 2D Simulation nachgestellt werden. Im Vergleich mit den frühen Champion-Zellen kann man zusammenfassen, dass man bereits heute n-pert Zellvorläufer mit sehr hohem V OC Potential mit industriellen Methoden herstellen kann, aber die Effizienz letztendlich drastisch durch die Siebdruckmetallisierung limitiert wird. Die Effizienz der n-pert Zelle mit Frontseitenemitter konnte bisher auf 20.14% erhöht werden. Da zu erwarten ist, dass Siebdruckmetallisierung mittelfristig die wichtigste Methode zur Kontaktbildung bleiben wird, haben wir untersucht, inwieweit die Emitterkontakte durch das Emitterprofil abgeschirmt werden können. Homogene Emitterschichten mit tiefem p-n Übergang zeigen eine effektive Abschirmwirkung der Kontakte, aber gleichzeitig auch stark erhöhte Augerrekombination, weshalb sie im Zellprozess ungeeignet sind. Daher wurden selektive Dotierprozesse untersucht, die beide Eigenschaften kombinieren sollen. Geeignete Prozesse müssen also eine sehr tiefe Dotierung ( 1 µm) im Bereich der Metallisierung bei moderater Dotierung der Bereiche zwischen der Metallisierung ermöglichen. Leider sind viele der erprobten Ansätze für selektive n + Dotierung gänzlich ungeeignet für die Erzeugung tiefer, selektiver p + Strukturen. Erfolge wurden jedoch mit der Laserdiffusion aus hoch borhaltigen Schichten erzielt. An Zellen mit derartigem selektiven Boremitter konnte dadurch ein um 5 mv erhöhtes V OC gemessen werden. Eine deutliche Reduzierung der Kontaktrekombination ist nur zu erwarten, wenn die aluminiumhaltigen Silberpasten ganz ersetzt werden können. Dazu wurde ein gänzlich neues Metallisierungskonzept erprobt, bei dem Aluminiumpaste

121 durch einen Boremitter legiert wird, welcher durch eine lokal geöffnete Passivierungsschicht geschützt ist. Dieser Prozess wurde an einer n-pert Zelle mit Rückseitenemitter evaluiert. Insbesondere wurde dabei gezeigt, dass der lokale Rekombinationsstrom J 0p + (met) an diesen Kontakten nur 1000 fa/cm 2 beträgt. Die Effizienz der Rückseitenemitterzelle wurde bisher auf 19.53% erhöht, was ein bemerkenswertes Resultat ist, wenn man berücksichtigt dass weder Dotierung noch die Kontaktierung der Frontseite optimiert sind. Die ursprüngliche n-pert Zelle mit Frontseitenemitter ist bifazial dank des offenen Kontaktlayouts auf der Rückseite. Diese Eigenschaft findet man bei vielen n- Typ Konzepten, wobei sie selten aktiv genutzt wird. Das liegt nicht zuletzt an der Schwierigkeit den Mehrertrag in bifazaler Anwendung sicher zu bestimmen. Selbst bei Messungen bifazialer Zellen unter einseitiger Beleuchtung ergeben sich Messunsicherheiten durch den Einfluss der Reflektion des Messhintergrunds. Wir haben daher eine Messvorrichtung konstruiert, mit der Zellen gezielt unter beidseitiger Beleuchtung gemessen werden können, um die Zelleffizienz unter realen bifazialen Bedingungen zu errechnen. Dies kann einen Beitrag leisten, das Standardtestverfahren auch auf bifaziale Solarzellen auszudehnen. Sowohl Standardisierung der Messverfahren, als auch Studien, die den bifazialen Mehrertrag in verschiedenen Szenarien quantifizierbar machen, werden für die Akzeptanz und Verbreitung von bifazialen Zellen und Modulen von entscheidender Bedeutung sein. Eine weitere große Fehlerquelle beim Messen der IV Kennlinien von Hocheffizienzzellen ist die Hysterese beim schnellen Durchfahren der angelegten Spannung. Dies ist auf die Diffusionskapazität zurückzuführen und Methoden zum Erkennen und Vermeiden dieses Fehlers wurden aufgezeigt. Für die weitere Entwicklung des n-pert Zellkonzepts liegt die Priorität weiterhin auf der Metallisierung. Die Reduzierung des Silbermetallisierungsgrades, z.b. durch Feinliniendruck oder nicht-durchfeuernde Busbars, führt unmittelbar zur Verbesserung des V OC sowie des J SC aufgrund geringerer Abschattung. Eine graduelle Verbesserung der aluminiumhaltigen Pasten ist auch zu erwarten. Bei der Bordiffusion kann letztendlich nur in industriellem Maßstab gezeigt werden, ob die Homogenität der Emitterdotierung mit aktuellen Rohrdiffusionsöfen sichergestellt werden kann. Alternativ gibt es sonst interessante Ansätze eine BSG Dotierquelle ex-situ mit APCVD Methoden oder gar druckbaren Pasten aufzubringen. Eine Modifikation der konventionellen Rohröfen hin zu mehreren Gaszuführungen kann ebenfalls helfen die Homogenität, insbesondere auch die des Borsilikatglases, zu verbessern. Für alle vorgestellten Zelltypen besteht eine interessante Option in dem Einbringen des Phosphordotanden durch Ionenimplantation. Da dies ein ausschließlich einseitiger Prozess ist, kann auf die Rückätzung, respektive Dotierbarrieren, im Prozess verzichtet werden. Der Einsatz selektiver Strukturen für das Phosphor BSF (wie auch das FSF) könnte weiter verfolgt werden, um die Gesamtdotierung zu reduzieren. Dies ist, wie gezeigt wurde, im Hinblick auf das V OC Potential entscheidend, die lokal hohe Dotierung gleichzeitig aber bislang notwendig, um einen niedrigen Kontaktwiderstand zu erreichen. 113

122

123 List of Acronyms, Symbols and Constants Acronym a-si Ag/Al Al 2 O 3 ALD AM1.5G APCVD atm BD/LD BRL B 2 O 3 BBr 3 BCl 3 BN BSF BSG c-si Cz DI DP FGA ECN ECV EQE FSF FZ IV IR IMEC HIT IBC INES Description amorphous silicon Aluminium containing Silver Paste Aluminium Oxide Atomic Layer Deposition Air Mass 1.5 Global (irradiation spectrum) Atmospheric Pressure Chemical Vapour Deposition atmosphere, pressure unit (1013mbar) Boron (tube) Diffusion (followed by) Laser Diffusion Boron Rich Layer Boron Oxide Boron Tribromide Boron Trichloride Boron Nitride Back Surface Field Boron Silicate Glass crystalline silicon Czochralski (method) De-Ionized (water) Doping Paste Forming Gas Anneal Energieonderzoek Centrum Nederland Electrochemical Capacitance Voltage External Quantum Efficiency Front Surface Field Float Zone (method) Current-Voltage (characteristic) Infrared Interuniversity Microelectronics Centre (also name of cleaning procedure) Heterojunction with Intrinsic Thin layer Interdigitated Back Contact Institut National de l Énergie Solaire 115

124 116 Chapter 7: Summary and Outlook Acronym IPA IQE ITO KOH LD/BD LID LSM mcz µw-pcd MWT NaOH NAOS PeCVD PERC PERL PERT POCl 3 PSG QSSPC R&D SC-OC sccm SCR SEM SIMS SiN x slm SRH SRV STC STD TCA TD TMA Description Isopropyl Alcohol Internal Quantum Efficiency Indium Tin Oxide Potassium Hydroxide Laser Diffusion (followed by) Boron (tube) Diffusion Light Induced Degradation Laser Scanning Microscope Magnetic Czochralski (method) Microwave-detected Photoconductance Decay Metal Wrap Through Sodium Hydroxide Nitric Acid Oxidation of Silicon Plasma Enhanced Chemical Vapour Deposition Passivated Emitter and Rear Cell Passivated Emitter and Rear Locally diffused Passivated Emitter and Rear Totally diffused Phosphorous Oxychloride Phosphorous Silicate Glass Quasi Steady State Photoconductance Research & Development Short Circuit to Open Circuit (sweep) Standard Cubic Centimeters per Minute Space Charge Region Scanning Electron Microscopy Secondary Ion Mass Spectroscopy Silicon Nitride Standard Liters per Minute Shockley-Read-Hall Surface Recombination Velocity Standard Testing Conditions Standard Deviation Trichloroethane Thermal Donors Trimethyl-Aluminium

125 Symbol Description Unit B coefficient for radiative recombination [cm 3 /s] C 0 doping concentration in melt [cm 3 ] C di f f usion diffusion capacitance [µf/cm 2 ] C depletion depletion capacitance [nf/cm 2 ] C n,p Auger recombination coefficients (electrons, holes) [cm 3 /s] n, p excess charge carrier density (electrons, holes) [cm 3 ] D it interface state density [cm 2 ] FF fill factor [%] FF 0 ideal fill factor [%] f bi f acial bifaciality factor F met metal fraction [%] H pulse energy density [J/cm 2 ] η conversion efficiency [%] iv OC implied open circuit voltage [mv] I L incident light intensity [W/cm 2 ] I ph photo generated current [A] I SC short circuit current [A] J SC short circuit current density [ma/cm 2 ] J 01 saturation current density [ f A/cm 2 ] J 0e emitter saturation current density [ f A/cm 2 ] J 0b base saturation current density [ f A/cm 2 ] J di f f usion diffusion current density [ f A/cm 2 ] J dri f t drift current density [ f A/cm 2 ] J rec recombination current density [ f A/cm 2 ] k segregation coefficient L n, L p diffusion length (electrons, holes) [µm] λ wavelength [nm] m 1, m 2 diode ideality factors µ n, µ p mobility (electrons, holes) [cm 2 /Vs] n,p charge carrier concentration (electrons, holes) [cm 3 ] n i intrinsic carrier concentration density [cm 3 ] N D, N A doping density (donors, acceptors) [cm 3 ] N T density of defect (trap) states [cm 2 ] P MPP power output at maximum power point [W] pff pseudo fill factor [%] R(λ) reflectivity [%] R S series resistance [Ωcm 2 ] R bulk bulk resistance [Ωcm 2 ] R Sheet sheet resistance [Ω/sq.] R Contact contact resistance [Ω] R Line line resistance [Ω/cm] ρ c contact resistivity [mωcm 2 ] S e f f effective surface recombination velocity [cm/s] 117

126 118 Chapter 7: Summary and Outlook Symbol Description Unit S met surface recombination velocity under metal [cm/s] S pas passivated surface recombination velocity [cm/s] SR spectral response [A/W] σ n, σ p defect capture cross section (electrons, holes) [cm 2 ] σ L excess conductivity [S/m] τ e f f effective minority carrier lifetime [µs] τ bulk bulk minority carrier lifetime [µs] τ emitter emitter minority carrier lifetime [µs] O x laser pulse overlap [%] T peak peak firing temperature [ C] U rad rate of radiative recombination [cm 3 s 1 ] U Auger rate of Auger recombination [cm 3 s 1 ] U SRH rate of Shockley-Read-Hall recombination [cm 3 s 1 ] U S surface recombination rate [cm 2 s 1 ] V MPP voltage at maximum power point [mv] V OC open circuit voltage [mv] V T thermal voltage [mv] ν th thermal velocity [cm/s] x etch etching depth [µm] Constant Value Description Unit c speed of light in vacuum [m/s] e elementary charge [C] h Planck constant [J s] k B Boltzmann constant [J/K]

127 Bibliography [1] D. M. Chapin, C. S. Fuller, and G. L. Pearson. A New Silicon p-n Junction Photocell for Converting Solar Radiation into Electrical Power. Journal of Applied Physics, 25 (5):676, May ISSN doi: / [2] J. E. Cotter, J. H. Guo, P. J. Cousins, M. D. Abbott, F. W. Chen, and K. C. Fisher. P-Type Versus n-type Silicon Wafers: Prospects for High-Efficiency Commercial Silicon Solar Cells. IEEE Transactions on Electron Devices, 53(8): , August ISSN doi: /TED [3] S. W. Glunz, S. Rein, J. Y. Lee, and W. Warta. Minority carrier lifetime degradation in boron-doped Czochralski silicon. Journal of Applied Physics, 90(5):2397, September ISSN doi: / [4] D. Macdonald and L. J Geerligs. Recombination activity of interstitial iron and other transition metal point defects in p- and n-type crystalline silicon. Applied Physics Letters, 85(18): , [5] A. Froitzheim, F. Suarez, M. Fischer, T. Spieß, K. Petter, A. Mohr, A. Gerlach, J. Müller, M. Träger, A. Guo, Y. Gao, K. Xie, G. Xing, L. Yong, Y. Wan, S. Zeng, A. C. Wu, G. Catin, S. de Magnienville, T. Vlasenko, I. Buchovska, O. Anspach, S. Julsrud, W. Osborne, A. Flores, D. Williams, A. Metz, N. Lenck, S. Raithel, P. Palinginis, F. Schitthelm, H. Neuhaus, O. Frank, S. Pingel, A. Boueke, G. Schubert, and A. Ristow. International Technology Roadmap for Photovoltaic. Technical report, SEMI, URL [6] G. Masson, M. Latour, M. Rekinger, I.-T. Theologitis, and M. Papoutsi. Global Market Outlook for Photovoltaics Technical report, European Photovoltaic Industry Association, [7] S. W. Glunz, J. Benick, D. Biro, M. Bivour, M. Hermle, D. Psysch, M. Rauer, C. Reichel, A. Richter, M. Rüdiger, C. Schmiga, D. Suwito, A. Wolf, and R. Preu. n-type silicon- enabling efficiencies >20% in industrial production. In Proc. 35th IEEE Photovoltaic Specialists Conf. (PVSC), [8] J. Schmidt, J. Bothe, R. Bock, C. Schmiga, R. Krain, and R. Brendel. N-type silicon - The better material choice for industrial high efficiency solar cells. In Proceedings of the 22th European Photovoltaic Solar Energy Conference and Exhibition, [9] S. P. Phang and D. Macdonald. Direct comparison of boron, phosphorus, and aluminum gettering of iron in crystalline silicon. Journal of Applied Physics, 109(7): , ISSN doi: /

128 120 BIBLIOGRAPHY [10] A. Cuevas, M. J. Kerr, C. Samundsett, F. Ferrazza, and G. Coletti. Millisecond minority carrier lifetimes in n-type multicrystalline silicon. Applied Physics Letters, 81(26):4952, December ISSN doi: / [11] M. Rüdiger, C. Schmiga, M. Rauer, M. Hermle, and S. W. Glunz. Efficiency Potential of n-type Silicon Solar Cells With Aluminum-Doped Rear p+ Emitter. IEEE Transactions on Electron Devices, 59(5): , May ISSN doi: /TED [12] F. Book, T. Wiedenmann, G. Schubert, H. Plagwitz, and G. Hahn. Influence of the Front Surface Passivation Quality on Large Area n-type Silicon Solar Cells with Al-Alloyed Rear Emitter. Energy Procedia, 8: , [13] J. Zhao, A. Wang, and M. A. Green. 24.5% Efficiency silicon PERT cells on MCZ substrates and 24.7% efficiency PERL cells on FZ substrates. Progress in Photovoltaics: Research and Applications, 7(6): , November ISSN doi: /(SICI) X(199911/12)7:6<471::AID-PIP298>3.0.CO;2-7. [14] J. Benick, B. Hoex, M. C. M. van de Sanden, W. M. M. Kessels, O. Schultz, and S. W. Glunz. High efficiency n-type Si solar cells on Al2O3-passivated boron emitters. Applied Physics Letters, 92(25):253504, June ISSN doi: / [15] J. Benick, S. W. Glunz, M. Hermle, A. Richter, W. M. M. Kessels, G. Dingemans, and B. Hoex. High-Efficiency n-type Silicon Solar Cells with Front Side Boron Emitter. In Proceedings of the 24th European Photovoltaic Solar Energy Conference, pages WIP-Munich, November ISBN doi: / 24thEUPVSEC2009-2BP.1.3. [16] J. Zhao, J. Schmidt, A. Wang, Guangchun Z., B. S. Richards, and M. A. Green. Performance instability in n-type PERT silicon solar cells. In Proc. 31st IEEE Photovoltaic Specialists Conf. (PVSC), volume 1, pages Vol.1, [17] A. Richter, J. Benick, A. Kalio, J. Seiffe, M. Hörteis, M. Hermle, and S. W. Glunz. Towards industrial n-type PERT silicon solar cells: rear passivation and metallization scheme. Energy Procedia, 8: , January ISSN doi: /j.egypro [18] S. Bordihn, V. Mertens, P. Engelhart, T. Florian, J. Cieslak, F. Stenzel, P. Kappe, T. Ballmann, J. Y. Lee, T. Lindner, M. Junghänel, J. W. Müller, W. M. M. Kessels, and P. Wawer. Large area n-type Cz double side contacted back junction boron emitter solar cell. In Proceedings of the 26th European Photovoltaic Solar Energy Conference and Exhibition, [19] V. Mertens, T. Ballmann, J. Cieslak, M. Kauert, A. Mohr, F. Stenzel, M. Junghänel, K. Suva, C. Klenke, G. Zimmermann, and J.W. Müller. Large Area n-type Cz Double Side Contact Back Junction Solar Cell with 21.3% Conversion Efficiency. In Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, [20] I. G. Romijn, B. B. Aken, J. Anker, and A. R. Burgers. Industrial implementation of efficiency improvements in n-type solar cells and modules. In Proceedings of

129 BIBLIOGRAPHY 121 the 27th European Photovoltaic Solar Energy Conference and Exhibition, pages , [21] N. Guillevin, B. Heurtault, L.J. Geerligs, B.B. Van Aken, I.J. Bennett, M.J. Jansen, A.W. Weeber, J. Bultman, J. Wang, Z. Wang, J. Zhai, Z. Wan, S. Tian, W. Zhao, Z. Hu, G. Li, B. Yu, and J. Xiong. High Power n-type Metal-Wrap-through Cells and Modules Using Industrial Processes. In Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition. WIP, [22] S. Gall, A. Lanterne, S. Manuel, V. Sanzone, R. Cabal, Y. Veschetti, A. Bettinelli, H. Robin, P. Lefillastre, and C. Gillot. High Efficient Industrial N-Type Technology: From Cell to Module. In Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, [23] D. Kania, T. S. Böscke, M. Braun, P. Sadler, C. Schöllhorn, M. Dupke, D. Stichtenoth, A. Helbig, R. Carl, K. Meyer, A. Grohe, J. Lossen, and H.-J. Krokoszinski. Pilot Line Production of Industrial High-Efficient Bifacial n-type Silicon Solar Cells with Efficiencies Exceeding 20.6 %. In Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, [24] S. Ramanathan, V. Meemongkolkiat, A. Das, A. Rohatgi, and I. Koehler. Understanding and Fabrication of 20% Efficient Cells Using Spin-on-Based Simultaneous Diffusion and Dielectric Passivation. IEEE Journal of Photovoltaics, 2(1):22 26, January ISSN doi: /JPHOTOV [25] F. Granek. High efficiency back-contact back-junction solar cells. PhD thesis, Universität Freiburg, [26] R. Woehl, M. Rüdiger, D. Biro, and J. Wilde. All-screen-printed back-contact backjunction silicon solar cells with aluminum-alloyed emitter and demonstration of interconnection of point-shaped metalized contacts. Progress in Photovoltaics: Research and Applications, pages n/a n/a, September ISSN doi: /pip [27] D. D. Smith, P. J. Cousins, A. Masad, A. Waldhauer, S. Westerberg, M. Johnson, X. Tu, T. Dennis, G. Harley, G. Solomon, S. Rim, M. Shepherd, S. Harrington, M. Defensor, A. Leygo, P. Tomada, J. Wu, T. Pass, L. Ann, L. Smith, N. Bergstrom, C. Nicdao, P. Tipones, and D. Vicente. Generation III high efficiency lower cost technology: Transition to full scale manufacturing. In Proc. 38th IEEE Photovoltaic Specialists Conf. (PVSC), pages IEEE, June ISBN doi: /PVSC [28] J.-B. Heng, J. Fu, B. Kong, Y. Chae, W. Wang, Z. Xie, A. Reddy, C. Bietel, K. Lam, and Z. Xu. Commercialization of Tunnel Oxide Junction Based Cell and Module with Efficiency and Good Reliability. In Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, [29] F. Buchholz, G. Galbiati, A. Edler, E. Wefringhaus, B. Ferstl, and M. Ross. Comparison of Wafer Cleaning Procedures for High Efficiency n-type Solar Cells. In WIP, editor, Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, 2013.

130 122 BIBLIOGRAPHY [30] Hans-Günther Wagemann and Heinz Eschrich. Photovoltaik. Teubner, ISBN [31] M. A. Green. Solar cells: Operating principles, technology, and system applications, volume -1. Prentice-Hall, [32] K. R. McIntosh. Lumps, Humps and Bumps: Three Detrimental Effects in the Currentvoltage Curve of Silicon Solar Cells. PhD thesis, University of New South Wales, [33] R. A. Sinton and A. Cuevas. A Quasi-Steady-State Open-Circuit Voltage Method for Solar Cell Characterization. In Proceedings of the 16th European Photovoltaic Solar Energy Conference and Exhibition, [34] T. Trupke, M. A. Green, P. Würfel, P. P. Altermatt, A. Wang, J. Zhao, and R. Corkish. Temperature dependence of the radiative recombination coefficient of intrinsic crystalline silicon. Journal of Applied Physics, 94(8):4930, October ISSN doi: / [35] J. Dziewior and W. Schmid. Auger coefficients for highly doped and highly excited silicon. Applied Physics Letters, 31(5):346, September ISSN doi: / [36] M. J. Kerr and A. Cuevas. General parameterization of Auger recombination in crystalline silicon. Journal of Applied Physics, 91(4):2473, ISSN doi: / [37] W. Shockley and W. Read. Statistics of the Recombinations of Holes and Electrons. Physical Review, 87(5): , September ISSN X. doi: / PhysRev [38] R. Hall. Electron-Hole Recombination in Germanium. Physical Review, 87(2): , July ISSN X. doi: /PhysRev [39] D. K. Schroder. Semiconductor Material and Device Characterization. Wiley-IEEE Press, ISBN [40] B. Fischer. Loss analysis of crystalline silicon solar cells using photoconductance and quantum efficiency measurements. PhD thesis, Universität Konstanz, [41] R. A. Sinton and D. Macdonald. WCT-120 Photoconductance Lifetime Tester and optional Suns-VOC Stage User Manual. Sinton Instruments, [42] G. Masetti, M. Severi, and S. Solmi. Modeling of carrier mobility against carrier concentration in arsenic-, phosphorus-, and boron-doped silicon. IEEE Transactions on Electron Devices, 30(7): , July ISSN doi: /T-ED [43] D. E. Kane and R. M. Swanson. Measurement of the emitter saturation current by a contactless photoconductivity decay method. In Proc. 18th IEEE Photovoltaic Specialists Conf. (PVSC), [44] M. Wilson, Al. Savtchouk, J. Lagowski, K. Kis-Szabo, F. Korsos, A. Toth, R. Kopecek, and V. D. Mihailetchi. QSS-µPCD measurement of lifetime in silicon wafers: advantages and new applications. Energy Procedia, 8: , 2011.

131 BIBLIOGRAPHY 123 [45] M. Wilson, P. Edelman, J. Lagowski, S. Olibet, and V. Mihailetchi. Improved QSSuPCD measurement with quality of decay control: Correlation with steady-state carrier lifetime. Solar Energy Materials and Solar Cells, 106:66 70, [46] D. K. Schroder and D. L. Meier. Solar cell contact resistance: A review. IEEE Transactions on Electron Devices, 31(5): , May ISSN doi: /T-ED [47] PN4300PC Electrochemical C-V Profiler with Photovoltage Spectroscopy Operating Manual. Technical report, Accent Semiconductor Technologies Inc., [48] E. Arai, H. Nakamura, and Y. Terunuma. Interface Reactions of B2O3-Si System and Boron Diffusion into Silicon. Journal of The Electrochemical Society, 120(7):980, ISSN doi: / [49] P. Negrini. Boron Predeposition in Silicon Using BBr3. Journal of The Electrochemical Society, 125(4):609, ISSN doi: / [50] G. U. Pignatel and G. Queirolo. Further insight on boron diffusion in silicon obtained with Auger electron spectroscopy. Thin Solid Films, 67(2): , April ISSN doi: / (80)90454-X. [51] T. L. Aselage. The coexistence of silicon borides with boron-saturated silicon: Metastability of SiB3. Journal of Materials Research, 13(7): , [52] M. A. Kessler, T. Ohrdes, B. Wolpensinger, and N.-P. Harder. Charge carrier lifetime degradation in Cz silicon through the formation of a boron-rich layer during BBr 3 diffusion processes. Semiconductor Science and Technology, 25(5):055001, May ISSN doi: / /25/5/ [53] D. M. Brown and P. R. Kennicott. Glass source B diffusion in Si and SiO2. Journal of The Electrochemical Society., 118(2): , [54] I. Kurachi and K. Yoshioka. Investigation of boron solid-phase diffusion from BSG film deposited by APCVD for solar cell application. In Proceedings of the 27th European Photovoltaic Solar Energy Conference and Exhibition, pages 2 5, ISBN [55] P. J. Cousins and J. E. Cotter. Minimizing lifetime degradation associated with thermal oxidation of upright randomly textured silicon surfaces. Solar Energy Materials and Solar Cells, 90(2): , January ISSN doi: /j.solmat [56] D. Macdonald, H. Mäckel, and A. Cuevas. Effect of gettered iron on recombination in diffused regions of crystalline silicon wafers. Applied Physics Letters, 88(9):092105, ISSN doi: / [57] V. Vähänissi, A. Haarahiltunen, H. Talvitie, M. Yli-Koski, J. Lindroos, and H. Savin. Physical mechanisms of boron diffusion gettering of iron in silicon. physica status solidi (RRL) - Rapid Research Letters, 4(5-6): , June ISSN doi: /pssr [58] S. P. Phang, W. Liang, B. Wolpensinger, M. A. Kessler, and D. Macdonald. Tradeoffs Between Impurity Gettering, Bulk Degradation, and Surface Passivation of

132 124 BIBLIOGRAPHY Boron-Rich Layers on Silicon Solar Cells. IEEE Journal of Photovoltaics, 3(1): , January ISSN doi: /JPHOTOV [59] K. Ryu, A. Upadhyaya, H.-J. Song, C.-J. Choi, A. Rohatgi, and Y.-W. Ok. Chemical etching of boron-rich layer and its impact on high efficiency n-type silicon solar cells. Applied Physics Letters, 101(7):073902, August ISSN doi: / [60] R. Falster and V. Voronkov. The engineering of intrinsic point defects in silicon wafers and crystals. Materials Science and Engineering B, 73(1-3):87 94, April ISSN doi: /S (99) [61] G. Müller, J.-J. Metois, and P. Rudolph. Crystal Growth - From Fundamentals to Technology. Elsevier, [62] H. Föll and B. O. Kolbesen. Agglomerate von Zwischengitteratomen (Swirl- Defekte) in Silizium - Ihre Bedeutung für Grundlagenforschung und Technologie. In Jahrbuch der Akademie der Wissenschaften in Göttingen, page p. 27, [63] A. Usami, Y. Fujii, and K. Monoka. The effect of swirl defects on the minority carrier lifetime in heat-treated silicon crystals. Journal of Physics D: Applied Physics, 10: , doi: / /10/6/013. [64] V.V. Voronkov. The mechanism of swirl defects formation in silicon. Journal of Crystal Growth, 59(3): , [65] M. Miyoshi, N. Shimizu, Y. Imanishi, O. Oda, and J. Nishizawa. Investigation of Boron Diffusion into Silicon Using a Liquid Boron Tribromide Source and Its Application to Buried-Gate-Type Static-Induction Thyristors. Journal of The Electrochemical Society, 152(8):G601, ISSN doi: / [66] J. J. Biernacki, R. Subramanian, M. R. Islam, C. Schewe, and M. C. Rogers. Gas- Phase Mixing and Dispersion in a Diffusion Furnace. Journal of The Electrochemical Society, 151(1):G24, ISSN doi: / [67] J. Libal, R. Petres, T. Buck, R. Kopecek, G. Hahn, R. Ferre, M. Vetter, I. Martín, K. Wambach, I. Roever, and P. Fath. N-Type multicrystalline silicon solar cells: BBr3- diffusion and passivation of p+ diffused silicon surfaces. In Proceedings of the 20th European Photovoltaic Solar Energy Conference and Exhibition, number June, [68] J. Schmidt, A. Merkle, R. Bock, P. P. Altermatt, A. Cuevas, N.-P. Harder, B. Hoex, R. de Sanden, E. Kessels, and R. Brendel. Progress in the Surface Passivation of Silicon Solar Cells. In Proceedings of the 23th European Photovoltaic Solar Energy Conference, [69] P. P. Altermatt, H. Plagwitz, R. Bock, J. Schmidt, R. Brendel, M. J. Kerr, and A. Cuevas. The surface recombination velocity at boron doped emitters: comparison between various passivation techniques. In Proceedings of the 21st European Photovoltaic Solar Energy Conference and Exhibition, [70] F. W. Chen, T. A. Li, and J. E. Cotter. Passivation of boron emitters on n-type silicon by plasma-enhanced chemical vapor deposited silicon nitride. Applied Physics

133 BIBLIOGRAPHY 125 Letters, 88(26):263514, June ISSN doi: / [71] R. R. King, R. A. Sinton, and R. M. Swanson. Studies of diffused phosphorus emitters: saturation current, surface recombination velocity, and quantum efficiency. IEEE Transactions on Electron Devices, 37(2): , [72] R. R. King and R. M. Swanson. Studies of diffused boron emitters: saturation current, bandgap narrowing, and surface recombination velocity. IEEE Transactions on Electron Devices, 3(6), [73] Y. Larionova, V. Mertens, N.-P. Harder, and R. Brendel. Surface passivation of n-type Czochralski silicon substrates by thermal-sio2/plasma-enhanced chemical vapor deposition SiN stacks. Applied Physics Letters, 96(3):032105, ISSN doi: / [74] A. G. Aberle, S. W. Glunz, and W. Warta. Impact of illumination level and oxide parameters on Shockley Read Hall recombination at the Si SiO2 interface. Journal of Applied Physics, 71(9):4422, ISSN doi: / [75] A. F. Thomson and K. R. McIntosh. Degradation of oxide-passivated borondiffused silicon. Applied Physics Letters, 95(5):052101, August ISSN doi: / [76] V. D. Mihailetchi, Y. Komatsu, and L. J. Geerligs. Nitric acid pretreatment for the passivation of boron emitters for n-type base silicon solar cells. Applied Physics Letters, 92(6):063510, ISSN doi: / [77] R. C. G. Naber, N. Guillevin, A. R. Burgers, L. J. Geerligs, and A.W. Weeber. ECN N-Type silicon solar cell technology: an industrial process that yields 18.5%. In Proceedings of the 24th European Photovoltaic Solar Energy Conference, [78] D. Kania, P. Saint-Cast, D. Wagenmann, M. Hoffmann, J. Rentsch, and R. Preu. Industrial negatively charged c-si surface passivation by inline PECVD AlOx. In Proceedings of the 24th European Photovoltaic Solar Energy Conference, [79] B. Hoex, J. Schmidt, P. Pohl, M. C. M. van de Sanden, and W. M. M. Kessels. Silicon surface passivation by atomic layer deposited Al2O3. Journal of Applied Physics, 104 (4):044903, August ISSN doi: / [80] F. Werner, W. Stals, R. Görtzen, B. Veith, R. Brendel, and J. Schmidt. High-rate atomic layer deposition of Al2O3 for the surface passivation of Si solar cells. Energy Procedia, 8(null): , January ISSN doi: /j.egypro [81] P. Saint-Cast, J. Benick, D. Kania, L. Weiss, M. Hofmann, J. Rentsch, R. Preu, and S. W. Glunz. High-Efficiency c-si Solar Cells Passivated With ALD and PECVD Aluminum Oxide. IEEE Transactions on Electron Devices, 31(7): , doi: /LED [82] H.-P. Sperlich, D. Decker, P. Saint-Cast, E. Erben, and L. Peters. High productive Solar Cell Passivation on Roth&Rau MAiA MW-PECVD inline machine- a comparison of Al2O3, SiO2 and SiNx-H process conditions and performance. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition,

134 126 BIBLIOGRAPHY pages , [83] G. Dingemans, P. Engelhart, R. Seguin, F. Einsele, B. Hoex, M. C. M. van de Sanden, and W. M. M. Kessels. Stability of Al2O3 and Al2O3/a-SiNx:H stacks for surface passivation of crystalline silicon. Journal of Applied Physics, 106(11):114907, December ISSN doi: / [84] M. Garin, U. Rau, W. Brendle, I. Martin, and R. Alcubilla. Characterization of a- Si:H c-si interfaces by effective-lifetime measurements. Journal of Applied Physics, 98(9):093711, November ISSN doi: / [85] V. Prajapati, R. Mertens, J. Poortmans, T. Janssens, P. Choulat, and J. Horzel. Oxidation Enhanced Diffusion on p-type PERC Silicon Solar Cells. In Proceedings of the 27th European Photovoltaic Solar Energy Conference and Exhibition, pages WIP, October ISBN doi: /27thEUPVSEC2012-2BO.7.6. [86] P. A. Basore. Numerical modeling of textured silicon solar cells using PC-1D. IEEE Trans. Elec. Dev., 37((2)):337, [87] G. Hahn. Status of Selective Emitter Technology. proceedings of the, pages , October doi: /25thEUPVSEC2010-2DP.2.2. [88] A. Edler, V. Mihailetchi, R. Kopecek, R. Harney, T. Boescke, D. Stichtenoth, J. Lossen, K. Meyer, R. Hellriegel, T. Aichele, and H.-J. Krokoszinski. Improving screen printed metallization for large area industrial solar cells based on n-type material. Energy Procedia, 8(April): , January ISSN doi: /j.egypro [89] Y. Veschetti, R. Cabal, P. Brand, V. Sanzone, G. Raymond, and A. Bettinelli. High Efficiency on Boron Emitter n-type Cz Silicon Solar Cells With Industrial Process. IEEE Journal of Photovoltaics, 1(2): , [90] M. Colina, I. Martín, C. Vos, A. Morales-Vilches, P. Ortega, G. López, A. Orpella, R. Alcubilla, I. Sánchez-Aniorte, and C. Molpeceres. Optimization of laser doping processes for the creation of p+ regions from solid dopant sources. In Proceedings of the 27th European Photovoltaic Solar Energy Conference and Exhibition, pages , [91] K. Hirata, T. Saitoh, A. Ogane, E. Sugimura, and T. Fuyuki. Selective Emitter Formation by Laser Doping for Phosphorous-Doped n-type Silicon Solar Cells. Applied Physics Express, 5(1):016501, January ISSN doi: / APEX [92] M. Lemiti, Y. Pellegrin, B. Semmache, B. Bazer-Bachi, A. Focsa, D. Blanc-Pélissier, and G. Poulain. Laser-Induced Boron Diffusion for Selective Emitter n-type Solar Cells. In Proceedings of the 27th European Photovoltaic Solar Energy Conference and Exhibition, pages WIP, October ISBN doi: / 27thEUPVSEC2012-2CV [93] T. C. Röder, S. J. Eisele, P. Grabitz, C. Wagner, G. Kulushich, J. R. Köhler, and J. H. Werner. Add-on laser tailored selective emitter solar cells. Progress in Photovoltaics: Research and Applications, 18(7): , ISSN X. doi: /pip.

135 BIBLIOGRAPHY [94] M. Dahlinger, S. J. Eisele, J. R. Köhler, and J. H. Werner. Laser doped boron emitters with sputtered precursor. In Proceedings of the 26th European Photovoltaic Solar Energy Conference and Exhibition, [95] S. Fernandez-Robledo, U. Jäger, E. Lohmüller, and J. Nekarda. Laser Doping from Borosilicate Glass for Metallization of Boron Emitters. In Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, [96] S. Kontermann. Characterization and modeling of contacting crystalline silicon solar cells. PhD thesis, Universität Konstanz, [97] E. Cabrera, S. Olibet, J. Glatz-Reichenbach, R. Kopecek, D. Reinke, and G. Schubert. Experimental evidence of direct contact formation for the current transport in silver thick film metallized silicon emitters. Journal of Applied Physics, 110(11):114511, ISSN doi: / [98] L. Venture, A. Slaoui, J. C. Muller, L. Pirozzi, U. Besi-Vetrella, and E. Salza. Realization of selective emitters by rapid thermal and laser assisted techniques. In Proceedings of the 13th European Photovoltaic Solar Energy Conference and Exhibition, pages , [99] T. C. Röder, P. Grabitz, S. Eisele, C. Wagner, J. R. Kohler, and J. H. Werner. 0.4% absolute efficiency gain of industrial solar cells by laser doped selective emitter. In th IEEE Photovoltaic Specialists Conference (PVSC), pages IEEE, June ISBN doi: /PVSC [100] U. Jäger, M. Okanovic, M. Hörteis, A. Grohe, and R. Preu. Selective Emitter by Laser Doping from Phosphosilicate Glass. In Proceedings of the 24th European Photovoltaic Solar Energy Conference, pages WIP-Munich, November ISBN doi: /24thEUPVSEC2009-2CV.5.9. [101] J.Y. Lee. Boron-back surface field with spin-on dopants by rapid thermal processing. In Proceedings of the 19th European Photovoltaic Solar Energy Conference and Exhibition, pages , [102] J. Jourdan, Y. Veschetti, S. Dubois, T. Desrues, and R. Monna. Formation of Borondoped region using spin-on dopant: investigation on the impact of metallic impurities. Progress in Photovoltaics: Research and Applications, 16(5): , August ISSN doi: /pip.818. [103] T. W. Krygowski. A novel simultaneous diffusion technology for low-cost, high-efficiency silicon solar cells. PhD thesis, Georgia Institute of Technology, [104] A. Das. Development of High-Efficiency Boron Diffused Silicon Solar Cells. PhD thesis, Georgia Institute of Technology, [105] A. Dastgheib-Shirazi, H. Haverkamp, B. Raabe, F. Book, and G. Hahn. Selective Emitter for Industrial Solar Cell Production : A Wet Chemical Approach Using a Single Side Diffusion Process. In Proceedings of the 23th European Photovoltaic Solar Energy Conference, [106] D. Rudolph, K. Peter, O. Doll, A. Meijer, and I. Köhler. Etch Back Selective Emitter

136 128 BIBLIOGRAPHY Process with Single POCl3 Diffusion. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, pages WIP, October ISBN doi: /26thEUPVSEC2011-2BV [107] J. Müller, K. Bothe, S. Gatz, F. Haase, C. Mader, and R. Brendel. Recombination at laser-processed local base contacts by dynamic infrared lifetime mapping. Journal of Applied Physics, 108(12):124513, ISSN doi: DOI: / [108] D. Chen, Z. Li, Z. Liang, H. Shen, Z. Feng, Y. Yang, and P. Verlinden. Analysis of Morphologies and Distribution of Al-Doped Local Back Surface Field for Screen Printed i-perc Solar Cell. In Proceedings of the 27th European Photovoltaic Solar Energy Conference and Exhibition, pages , [109] E. Urrejola, K. Peter, H. Plagwitz, and G. Schubert. Silicon diffusion in aluminum for rear passivated solar cells. Applied Physics Letters, 98(15):153508, April ISSN doi: / [110] R. Lago, L. Pérez, H. Kerp, I. Freire, I. Hoces, N. Azkona, F. Recart, and J. C. Jimeno. Screen printing metallization of boron emitters - Lago Progress in Photovoltaics: Research and Applications - Wiley Online Library. Progress in Photovoltaics: Research and Applications, 18(1):20 27, January ISSN doi: /pip.933. [111] S. Riegel, F. Mutter, T. Lauermann, B. Terheiden, and G. Hahn. Review on screen printed metallization on p-type silicon. Energy Procedia, 21(null):14 23, January ISSN doi: /j.egypro [112] H. Kerp, S. Kim, R. Lago, F. Recart, I. Freire, L. Perez, K. Albertsen, J. C. Jimeno, and A. Shaikh. Development of screen printable contacts for p+ emitters in bifacial solar cells. In Proceedings of the 21st European Photovoltaic Solar Energy Conference and Exhibition, page 892. WIP, [113] R. Hoenig, M. Glatthaar, F. Clement, J. Greulich, J. Wilde, and D. Biro. New measurement method for the investigation of space charge region recombination losses induced by the metallization of silicon solar cells. Energy Procedia, 8(0): , ISSN doi: /j.egypro [114] T. Fellmeth, A. Born, A. Kimmerle, F. Clement, D. Biro, and R. Preu. Recombination at Metal-Emitter Interfaces of Front Contact Technologies for Highly Efficient Silicon Solar Cells. In Energy Procedia 8, pages , [115] R. A. Sinton, A. Cuevas, and M. Stuckings. Quasi-steady-state photoconductance, a new method for solar cell material and device characterization. In Proc. 25th IEEE Photovoltaic Specialists Conf. (PVSC), pages IEEE, ISBN doi: /PVSC [116] G. Schubert. Thick Film Metallisation of Crystalline Silicon Solar Cells. PhD thesis, Universität Konstanz, [117] M. M. Hilali. Understanding and Development of manufacturable Screen-Printed Contacts on High Sheet Resistance Emitters for Low-Cost Silicon Solar Cells. PhD thesis, Georgia Institure of Technology, 2005.

137 BIBLIOGRAPHY 129 [118] C. Ballif, D. M. Huljic, G. Willeke, and A. Hessler-Wyser. Silver thick-film contacts on highly doped n-type silicon emitters: Structural and electronic properties of the interface. Applied Physics Letters, 82(12):1878, ISSN doi: / [119] D. Pysch, A. Mette, A. Filipovic, and S. W. Glunz. Comprehensive analysis of advanced solar cell contacts consisting of printed fine-line seed layers thickened by silver plating. Progress in Photovoltaics: Research and Applications. Photovoltaics, 17(2): , doi: /pip.855. [120] G.C. Cheek, R.P. Mertens, R. Van Overstraeten, and L. Frisson. Thick-film metallization for solar cell applications. IEEE Transactions on Electron Devices, 31(5): , May ISSN doi: /T-ED [121] L. Koduvelikulathu. 2D simulation of solar cell losses [122] Albedo article. URL [123] J. Lossen. Increased energy yield for BOSCH n-type solar cells in bifacial application. In bifi Workshop, Konstanz, Germany, [124] Bsolar field results, URL aspx?sel=fieldresults. [125] P. Sanchez-Friera, B. Lalaguna, D. Montiel, J. Gil, L. J. Caballero, J. Alonso, M. Piliougine, and M. de Carmona. Development and characterization of industrial bifacial PV modules with ultrathin screen-printed solar cells. In Proceedings of the 22th European Photovoltaic Solar Energy Conference and Exhibition, [126] J. Hohl-Ebinger and W. Warta. Bifacial solar cells in STC measurement. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, volume 49, [127] R. A. Sinton. A solution to the problem of accurate flash testing of high-efficiency modules. In Proceedings of the 21st European Photovoltaic Solar Energy Conference and Exhibition, [128] C. Monokroussos, R. Gottschalg, A. N. Tiwari, G. Friesen, D. Chianese, and S. Mau. The Effects of Solar Cell Capacitance on Calibration Accuracy When using a Flash Simulator. In Proc. 4th WCPSEC, volume 2, pages , doi: / WCPEC [129] G. Friesen and H. A. Ossenbrink. Capacitance effects in high-efficiency cells. Solar Energy Materials and Solar Cells, 48(1-4):77 83, ISSN doi: DOI: /S (97)00072-X. [130] T. Roth, D. Wichmann, K. Meyer, and M. Orlob. In-depth analysis of transient errors of inline IV measurements. Energy Procedia, 8(1):82 87, January ISSN [131] H. Ohtsuka, M. Sakamoto, M. Koyama, K. Tsutsui, T. Uematsu, and Y. Yazawa. Characteristics of bifacial solar cells under bifacial illumination with various intensity levels. Progress in Photovoltaics: Research and Applications, 9(1):1 13, ISSN X. doi: /pip.336.

138 130 BIBLIOGRAPHY [132] M. Ezquer, I. Petrina, and J. M. Cuadra. Design of a special set-up for the IV characterization of bifacial photovoltaic solar cells. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, 2010.

139 List of Publications [1] V. D. Mihailetchi, J. Jourdan, A. Edler, R. Kopecek, R. Harney, D. Stichtenoth, J. Lossen, T. S. Böscke, H.-J. Krokoszinski. Screen printed n-type silicon solar cells for industrial application. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, [2] A. Edler, J. Jourdan, V. D. Mihailetchi, R. Kopecek, R. Harney, D. Stichtenoth, T. Aichele, A. Grochocki, J. Lossen, T. S. Böscke. High lifetime on n-type silicon wafers obtained after boron diffusion. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, 2010, pp [3] A. Edler, V. D. Mihailetchi, R. Kopecek, R. Harney, J. Lossen, T. S. Böscke, D. Stichtenoth, K. Meyer, R. Hellriegel, T. Aichele, H.-J. Krokoszinski. Improving screen printed metallization for large area industrial solar cells based on n-type material. Energy Procedia, 8: , [4] V. D. Mihailetchi, A. Edler, R. Harney, R. Kopecek, T. S. Böscke, J. Lossen. N-type Bifacial Solar Cells for Industrial Application, Future Photovoltaics,1, [5] A. Edler, M. Schlemmer, J. Ranzmeyer, and R. Harney. Flasher setup for bifacial measurements. In bifi Workshop, Konstanz, [6] A. Edler, M. Schlemmer, J. Ranzmeyer, and R. Harney. Understanding and Overcoming the Influence of Capacitance Effects on the Measurement of High Efficiency Silicon Solar Cells. Energy Procedia, vol. 27, pp , [7] A. Edler, V. D. Mihailetchi, C. Comparotto, L. J. Koduvelikulathu, R. Kopecek, R. Harney, T. Böscke, and J. Lossen. On the Metallization Losses of Bifacial n-type Silicon Solar Cells. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, 2012, pp [8] L. J. Koduvelikulathu, A. Edler, V. D. Mihailetchi, C. Comparotto, A. Halm, R. Kopecek, and K. Peter. Metallization and Firing Process Impact on Voc - A Simulation Study. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, 2012, pp [9] A. Edler, V. D. Mihailetchi, C. Comparotto, L. J. Koduvelikulathu, R. Kopecek, and R. Harney. Metal induced losses in bifacial n-type silicon solar cells: Investigation in Simulation and Experiment. In Proc. 22nd PVSC Asia, Hangzhou, [10] A. Edler, C. Comparotto, V. D. Mihailetchi, R. Harney, and R. Kopecek. BiSon: Bifacial Industrial Solar Cell on N-type, In npv workshop, Chambery, [11] A. Edler, P. Lill, M. Dahlinger, M. M. Eberspächer, V. D. Mihailetchi, C. Comparotto, 131

140 132 LIST OF PUBLICATIONS R. Harney, and R. Kopecek. Bifacial n-type Solar Cells with Selective Emitter, In Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, [12] S. Barth, O. Doll, I. Koehler, K. Neckermann, M. Blech, A. Lawerenz, A. Edler, R. Kopecek, and J. J. Schneider. 19.4% Efficient Bifacial Solar Cell with Spin-on Boron Diffusion, Energy Procedia, vol. 38, pp , [13] A. Edler, V.D. Mihailetchi, L. J. Koduvelikulathu, C. Comparotto, R. Kopecek, R. Harney. Metallization-induced recombination losses of bifacial silicon solar cells, accepted for publication in Progress in Photovoltaics: Research and Applications, 2014,

141 Acknowledgements I would like to thank: Prof. Dr. Ernst Bucher for the opportunity to write this thesis and his patronage, that established the success of ISC-Konstanz. For his vision of a world powered by solar energy and his generous support for promoting this great idea. Prof. Dr. Mikhail Fonin for writing the second expert assessment on this thesis. Valentin Mihailetchi for being a great mentor in all scientific questions and for sharing his broad knowledge on n-type solar cells. I could learn a lot from him and his dedication was often contagious for me. Geschäftsführer of ISC-Konstanz Rudolf Harney, Radovan Kopecek, Kristian Peter and Eckard Wefringhaus for being such a well-matched, committed team. For the ever open door, and for creating a great working atmosphere. Corrado Comparotto for two helping hands in our quest for the 20% efficient cell (measured with black back-sheet, obviously!) and especially for keeping my back free during the months of writing. All chicos from my office, Lejo Koduvelikulathu, Giuseppe Galbiati, Pirmin Preis and Corrado Comparotto for staying late with me every night before a conference deadline and lots of other fun. You re the best. All of you. All colleagues from the Brotzeit/Kaffee group for always keeping me healthy and awake and endless semi-scientific discussions. Michel Schlemmer and Joachim Ranzmeyer for designing the bifacial measurement system and fighting the quirks of the flasher with me. The entire staff of ISC for great improvisation skills and endless patience whenever a process needed to be finished as soon as possible. Our project partners at BOSCH Solar Energy AG, Tim Böscke, Jan Lossen, Daniel Stichtenoth, Hans Krokoszinski, Daniel Kania, Matthias Braun and Lutz Bornschein. Working with such knowledgeable people was always enriching and really introduced me to the trade of making solar cells on an industrial scale. Project partners at IPV Stuttgart, Patrick Lill, Morris Dahlinger and Maria Eberspächer for sharing their experience in laser processing of solar cells and a great collaboration. 133

142 134 LIST OF PUBLICATIONS Caro, my love, my parents and my sister for always supporting me and the encouragement to pursue my goals.