Agilent EEsof EDA This document is owned by Agilent Technologies, but is no longer kept current and may contain obsolete or inaccurate references. We regret any inconvenience this may cause. For the latest information on Agilent s line of EEsof electronic design automation (EDA) products and services, please go to: www.agilent.com/find/eesof
40 Bond wire modeling Characterization and modeling of bond wires for high-frequency applications By: Dean Nicholson and HeeSoo Lee Bond wires are used as the standard method of connecting microwave ICs and interconnect circuitry. The high impedance of the bond wires causes inductive discontinuities, which result in impedance mismatches and unwanted reflections. Various techniques have been used to provide lower mismatch interconnects between circuits, such as minimizing the gap between the circuits to shorten the bond wires, using wider ribbon bonds for lower impedance to connect the circuits, and even going to flip chip ICs to achieve lower inductance interconnects. This article uses lumped-element models and 3D modeling (using EDA software from Agilent Technologies) to accurately model the frequency and time domain performance of bond wire interconnects. Modeled results for a bond wire interconnect between two thin-film circuits are compared against measured results. Bond wires are the traditional way of connecting two circuits in the microwave industry due to their low cost, flexibility to adapt to different layouts, and relatively good return loss performance. There have been numerous papers presenting modeling and experimental results for bond wires to interconnect ICs and circuits together to frequencies up to 100 GHz and beyond [1-5], and for high-speed timedomain applications [6]. This article looks at a subset of that work, examining an efficient technique to extract lumped element models for bond wires used to interconnect 50-ohm microstrip lines on 127 µm-thick alumina circuits by using a combination of lumpedelement and 3D modeling tools, and then using these lumped-element models to optimize the frequency-domain performance of these bond wire transitions. A bond wire connecting two microstrips represents a short, high-impedance transmission line embedded in a 50-ohm system. Ball bonds with their relatively high arch have enough of this high-impedance length to give significant degradation in return loss even at frequencies below 20 GHz. There are various ways to minimize this reflective discontinuity either by going to wider interconnects such as ribbon bonds for a lower impedance, or by placing the two circuits Figure 1: A single 26 µm diameter wedgebond with bond center 25 µm high above circuit: circuit edge separation is 0.1 mm, with 127 µm (5 mil) thick alumina and Er of 9.9. Figure 2: Performance of a single 26 µm diameter wedgebond bond center 25 µm high above circuit: circuit edge separation is 0.1 mm, with 127 µm (5 mil) thick alumina and Er of 9.9. extremely close together to minimize the length of the high-impedance section. However, both of the techniques result in significant cost increase or producibility problems. Even the simplest and most producible way of improving the performance of a bond wire interconnect, by going to a wedge bond with the flattest feasible profile to minimize bond wire length and height as shown in figure 1 still gives Figure 3: A compensated through-line transition and two wedgebonds. Bond center is 25 µm high (maximum) and circuit edge separation is 0.1 mm, with 127 µm (5 mil) thick alumina and Er of 9.9. worse than 20 db return loss at frequencies above 20 GHz for producible circuit to circuit separations and trace setbacks, as shown in the performance plot in figure 2. Instead of using a transmission line model to represent the different impedance sections, it is more useful to use the semilumped element modeling technique described in [7], where the lumped element
42 Bond wire modeling values are derived from the transmission line model for the bond wires. Upon inspection, it can be seen that the simplest way to reduce the reflections caused by a high impedance discontinuity is to incorporate it into an LPF structure. The most compact LPF structure incorporating these bondwires is a shuntseries-shunt C-L-C structure whose physical implementation is shown in figure 3, where the shunt C sections are implemented as widened regions at the ends of the microstrip where the bond wires attach. The different height profiles modeled for the bond wires are shown in Figure 4. The schematic representation of this semilumped element model is shown in Figure 5, where good agreement is shown between the semi-lumped model in Advanced Design System (ADS) RF and Microwave design and simulation software and the 3D model from Electromagnetic Design System (EMDS), the 3D simulation software from Agilent Technologies, Inc. After this model is obtained, it can easily be shown that for a given length of highimpedance bondwires, a shunt-series-shunt C-L-C filter implementation is not only more compact than creating a series-shuntseries L-C-L filter by adding an extra high impedance section of micro-strip, the C-L- C LPF gives better S 11 and S 21 to higher frequency for a given length of bond wires. If Figure 4: Side views of various bond heights. Bond height measured from top of substrate to center of bondwire. Figure 5: EMDS simulation versus ADS simulation, 25 µm bond wire height. the capacitance of the bond wires is also taken into account explicitly with a C-L-C model used to approximate the bond wires, this will give more accurate results from the lumpedelement model if the capacitive compensation traces at the end of the microstrip are left the same, but the length of the bond wires is varied, which is the scenario that would be most likely seen in production. If the bond wires are modeled as simply an inductive element, there would be no reason to expect the value of the compensation capacitive elements at the end of the microstrip line to change as the length of the bond wires is varied, leading to a less accurate lumped element model over varying bond lengths. The value of going to the effort of obtaining an accurate lumped element value is that there is an extensive body of closed-form Figure 6: EMDS simulations of compensated thru line transitions for various max height wedgebonds versus best uncompensated (25 µm height) case, trace 50 µm back from circuit edge. Circuit edge to edge separation is 0.1 mm, with 127 µm thick alumina and Er of 9.9. solutions for lumped-element filter design that has been developed over many years that does not exist for transmission line or 3D structures, so once the lumped-element values are correctly obtained, it becomes more straight forward to design the bond-wire interconnects and capacitive compensation structures with the desired frequency- and time-domain characteristics. For example, when the lumped-element modeling is used, it is fast and simple to make tradeoffs suitable for a particular application between going to a 5-element LPF to extend the best high frequency return loss, versus staying with a 3-element LPF for improved compactness. There are additional benefits to extracting the lumped-element model, as dispersion effects in the time domain can now be quickly calculated and traded off with return loss to give optimized performance. The pattern of bond wires and circuit compensation elements is shown in figures 3 and 4, where the high impedance of the bond wires is first reduced by using two bond wires in parallel, and the resulting high impedance line is then capacitively compensated at both of its ends by widened sections of the microstrip ends to form a LPF with as high a cutoff frequency as possible, while still keeping good in-band return loss, as shown in figure 6. Though these compensated transitions with widened microstrip sections on the end are compared to uncompensated transitions that have no widened sections, it can be more difficult for the narrow 50-ohm microstrips on thinner ceramic substrates to have two bond wires interconnecting adjacent circuits. Note that for the capacitive compensation pattern shown in figure 3, even if the bond wires have significant arch to them, their high frequency return loss performance is still quite good up to 30 GHz. Reasonable dimensions were used that fit into standard contract manufacturer fabrication and assembly tolerances with the circuits separated by a 0.1 mm gap between them, and the traces being relieved 50 µm back from the circuit edges. Wedge bonding was assumed as mentioned previously, as this
44 Bond wire modeling allows for very flat bonds if desired, and is now commonly used by contract manufacturers in the high-frequency industry. To make an accurate measurement of the return loss of an optimized two-bondwire interconnect structure without using a microwave probe station and its associated complex calibration, the simple arrangement shown in figure 7 was used, where this SMT package previously described [8] is mounted on a small evaluation PCB with 2.4 mm end launch connectors suitable for use to 50 GHz. This arrangement allows for the use of a less than one minute electronic coaxial calibration using the electronic calibration module of the network analyzer. The arrangement of figure 7 provides for a long length of reflectionfree 50-ohm transmission line leading up to and away from the optimized two bond wire interconnect structure allowing the time gating function of the network analyzer to be used to accurately isolate only the reflections from the optimized two-bondwire interconnect structure. The measured reflection will be attenuated compared to its actual value by twice the insertion loss of the connector and the transmission lines leading up to the bond wire transition, due to the measured reflected wave having traveled up to the bond wire transition, and then back from there to the connector reference plane. The insertion loss of the connectors and transmission lines leading up to the transition can be accurately accounted for by using a separate calibration PCB with a short circuit at the same position as the bond wire discontinuity that is being measured, and then correcting the measured reflection by this value. This procedure was followed to give the accurately calibrated return loss results shown in figure 8 for the compensated twobond-wire transition shown in figure 7, which has its bond wire center height 25 µm above the substrate surface. The measured results match the modeled results well up to 35 GHz, at which point there begins to be significant deviation between them. Figure 7: Measurement configuration for optimally compensated two-bond-wire transition. The main reason we don t see better agreement between measured and modeled results is that the model parameters are based on the dimensions for the circuit that are in the design, rather than those that were actually measured from the physical implementation of the design. Thus, deviations between the designed and the actually implemented dimensions for things such as substrate thickness, conductor trace width for the 50-ohm and compensating sections of the transmission line, distance between the end of the compensating sections and the end of the substrate, and length of the gap between the two substrates will all cause discrepancies between measured and modeled values. These differences in observed results due to relatively small differences in designed versus implemented dimensions become more pronounced for the two cases where we observe them, specifically for high frequencies, where our measured and modeled results diverge above 35 GHz, and for very small reflections, where our measured and modeled results are offset by approximately 10 db below 35 GHz. However, it is important to note that a 10 db offset between a -30 db and -40 db S 11 measurement represents only the very small Figure 8: Measured versus modeled performance of two-bond-wire interconnect with optimized compensation pads.
46 Bond wire modeling difference between 0.1 percent of the power being reflected for the 30 db case and 0.01 percent of the power being reflected for the -40 db case. Thus a 10 db difference between very low values of reflection is not nearly as significant as a 10 db difference between a -10 db and -20 db S 11 measurement for example, and the fact that the shape of the S 11 plots matches well for measured and modeled results below 35 GHz, and they both represent a very low value of reflection leads to our description of them as being in good agreement. In conclusion, a technique is shown to optimally compensate the bond-wire interconnect between two microstrip circuits to give excellent return loss and insertion loss performance up to 50 GHz using circuit spacings and layouts suitable for a manufacturing environment. Good agreement was obtained between measured and modeled results. Lumped-element representations of the 3D structures were obtained that gave good agreement with the 3D simulated results, and these lumped-element models can be used to accurately include the effects of the bondwire transitions in more complex circuit simulations inlvolving multiple components and active devices. References U. Goebel, DC to 100 GHz Chip-to- Chip Interconnects with Reduced Tolerance Sensitivity by Adaptive Wirebonding, pp182-185, IEEE 3rd Topical Meeting on Electrical Performance of Electronic Packaging, November 1994. Hai-Young Lee, Wideband Characterization of a Typical Bonding Wire for Microwave and Millimeter-Wave Integrated Circuits, IEEE Trans. On MTT, Vol. 43, No 1, pp 63-68, January 1995. T. Krems, W. Haydl, H. Massler, J. Rudiger, Millimeter-Wave Performance of Chip Interconnections Using Wire Bonding and Flip Chip, IEEE MTT-S Digest, pp 247-250, 1996. Liam Devlin, How to Design Low-Cost MM-Wave Equipment, Presented at 2nd Annual Wireless Broadband Forum, Nov. 25-26, 2003, Cambridge, England. W. Simon et al, Interconnects and Transitions in Multilayer LTCC Multichip Modules for 24 GHz ISM-Band Applications, IEEE MTT-S Digest, pp 1047-1050, Boston, June 2000. G. Matthei, L. Young and EMT Jones, Microwave Filters and Impedance Matching Networks, and Coupling Structures, Artech House, 1980 reprint of 1964 ed., pp 85-104. Intel Packaging Databook, Chapter 4, 2000. D. Nicholson, Low Return Loss DC to 60 GHz SMT Package With Performance Verification by Precision 50 Ohm Load, 35th European Microwave Conference Digest, pp 157-160, October 4-6, 2005. About the authors Dean Nicholson is a hardware R&D engineer at the Santa Rosa site of Agilent s Test and Measurement Operation, where he is responsible for the development of advanced components for use in nextgeneration frequency- and time-domain test instrumentation. Hee-Soo Lee is a RF SiP/Module design flow specialist working at Agilent EEsof Division, where he is responsible for developing and promoting ADS/RFDE software solutions for RF SiP/Module market. Agilent Technologies www.agilent.com Company Information
For more information about Agilent EEsof EDA, visit: www.agilent.com/find/eesof Agilent Email Updates www.agilent.com/find/emailupdates Get the latest information on the products and applications you select. Agilent Direct www.agilent.com/find/agilentdirect Quickly choose and use your test equipment solutions with confidence. www.agilent.com For more information on Agilent Technologies products, applications or services, please contact your local Agilent office. The complete list is available at: www.agilent.com/find/contactus Americas Canada (877) 894-4414 Latin America 305 269 7500 United States (800) 829-4444 Asia Pacific Australia 1 800 629 485 China 800 810 0189 Hong Kong 800 938 693 India 1 800 112 929 Japan 0120 (421) 345 Korea 080 769 0800 Malaysia 1 800 888 848 Singapore 1 800 375 8100 Taiwan 0800 047 866 Thailand 1 800 226 008 Europe & Middle East Austria 0820 87 44 11 Belgium 32 (0) 2 404 93 40 Denmark 45 70 13 15 15 Finland 358 (0) 10 855 2100 France 0825 010 700* *0.125 /minute Germany 01805 24 6333** **0.14 /minute Ireland 1890 924 204 Israel 972-3-9288-504/544 Italy 39 02 92 60 8484 Netherlands 31 (0) 20 547 2111 Spain 34 (91) 631 3300 Sweden 0200-88 22 55 Switzerland 0800 80 53 53 United Kingdom 44 (0) 118 9276201 Other European Countries: www.agilent.com/find/contactus Revised: March 27, 2008 Product specifications and descriptions in this document subject to change without notice. Agilent Technologies, Inc. 2008