1024 x 1280 pixel dual shutter APS for industrial vision

Transcription

1 Paper presented at SPIE Electronic Imaging, Santa Clara, 22 Jan 2003 p.1/ x 1280 pixel dual shutter APS for industrial vision Herman Witters, Tom Walschap, uy Vanstraelen, enis Chapinal, uy Meynants, Bart Dierickx FillFactory NV, Schaliënhoevedreef 20 b, 2800 Mechelen, Belgium ABSTACT We present a 1.3 megapixel CMOS active pixel sensor dedicated to industrial vision. It features both rolling and synchronous shutter. Full frame readout time is 33 ms, and readout speed can be boosted by windowed region of interest (OI) readout. High dynamic range scenes can be captured using the double and multiples slope functionality. All operation modes and settings can be programmed over a serial or a parallel interface. Keywords: image sensor, APS, industrial vision, dual electronic shutter, high dynamic range 1. INTODUCTION CMOS image sensors do not outperform CCD s in terms of image quality or sensitivity. The power of CMOS is mainly in the ability to add systems functionality to the imager. In industrial vision CMOS image sensors show up in small and versatile smart cameras. The IBIS5 (fig. 5) is a 1.3 Megapixel (1280x1024 pixels, 6.7 µm pitch) CMOS active pixel sensor that is dedicated to industrial vision. It features a dual electronic shutter choice, and multiple slope sensitivity for high dynamic range imaging. Pixel rate is 40 MHz of a single 10 bit output channel, resulting in a full frame readout time of 36.4 ms, or a maximum full frame rate of 27.5 frames/second in rolling shutter. The random windowing or region of interest (OI) readout and several sub-sampling schemes, enable proportionally faster frame rates. The device has a full digital interface. All operation modes and settings can be programmed over a 3-wire serial, 2-wire serial or plain parallel interface. 1.1 IBIS5 specifications The next table summarizes the features of the IBIS5. Most of them are common and expected for a device with these dimensions. Two special features are discussed in subsequent chapters. Criterium Specification emarks Pixels 1280 (x) * 1024 (y) Pixel pitch 6.7 um square pixels Fill factor 60% Is only limited by the metal fill 1,2 Pixel rate 40 MHz Over a single output channel Full frame readout time is 36.3 ms Windowing andom window positions and sizes can be programmed by upload of start/stop coordinates Sub sampling Apart from normal operation, four 4 programmable sub sampling different 2:1 sub sampling modes are modes, both in X and Y, see below available that may be used to sub sample in a monochrome array or in a Bayer pattern color filter array. Color Both color and monochrome versions Color sensors have the Bayer pattern ADC 10 bit

2 Power consumption Two Electronic shutter modes High dynamic range Interfaces 250 mw peak Synchronous shutter olling shutter High dynamic range obtained by multiple slope operation Serial (2-wire, 3-wire) and 16-bit parallel See next chapter See next chapter Is used for upload of setting as windowing, sub sampling, gain/offset, clock granularity, etc. 1.2 Pixel cross section The following figure is a schematic cross section of the core of the pixel. Apart from the metal coverage, the complete pixel is light sensitive. The structure also minimizes the light sensitivity of the (parasitic) diode of the analog memory of the synchronous shutter. M3 e- C n++ Parasitic photo diode M2 p-well photo diode n++ n-well p- epitaxial layer = photo sensitive volume e- e- M1 e- P++ substrate Figure 1. Schematic cross section of the IBIS5 pixel, showing the photodiode, the transistors M1, M2, M3 (see fig. 3) and the storage node C. 1.3 Smart sub sampling modes Simply sub sampling by skipping on out of two pixels in a Bayer pattern color filter array (CFA) will yield a single color. If it is desired to maintain color information, one should sub sample in such way that all available color components remain available. Next figures show a few examples. B B B B B B B B

3 B B B B Figure 2 examples of possible sub sampling modes by combinations of all variants of 2:1 sub sampling in X and Y-direction. (1: no sub sampling; 2: alternating xoxoxo and oxoxox ; 3: xoxoxo both in X and Y; 4: xxooxxoo both in X and Y) 2. DUAL ELECTONIC SHUTTE The dual shutter option chooses between rolling shutter and synchronous (snapshot, triggered) shutter modes. The user may select a method of electronic shuttering that is optimal for his application, taking into account criteria as: synchronous integration, continuous imaging, triggered snapshot, integration-while-read, parasitic light sensitivity, noise and S/N behavior. Ideally one would like to have always a synchronous shutter. There are only very exceptional cases where a rolling shutter provides a functional advantage over a synchronous shutter; the only example that pops to my mind is the use in special effect photography. Then, why should one implement a rolling shutter as full option in a sensor for industrial vision? The reason lies in the circuit complexity of the pixel. The most simple and most popular active pixel is the threetransistor (3T) pixel. Yet the 3T pixel can only be used in conjunction with a rolling shutter. A true synchronous shutter needs at least a memory element inside the pixel. The simplest implementation of a synchronous shutter, used in the actual device, is the four-transistor pixel 4 shown in the next figure. reset M1 reset mux column output C M2 sample mux M3 M4 column output Figure 3. Schematics of the rolling shutter 3T (left) and triggered/snapshot shutter 4T (right) pixels. Synchronous shutter operation consists of following consecutive operations: (1) reset all pixels at the same time by pulsing the reset MOSFETs. (2) Integrate the photocurrent of all pixels synchronously, until the switch MOSFET sample is opened; the resulting voltage level is memorized on the capacitor C. (3) ead all pixel information in a sequential order. While doing that, pixels are read, reset, and read again in order to allow fixed pattern noise (FPN) cancellation in the differential column amplifiers 3. This 4T pixel can also emulate the 3T behavior, by shunting the sample MOSFET. There are two kinds of synchronous shutter operation. The most basic, which is implemented in the actual device, is triggered synchronous shutter (also called snapshot shutter). In such image sensor, integration starts immediately after an external command (trigger), and is followed by readout. eadout and integration phases are clearly separated in time. Continuous imaging is awkward. This mode is preferred in many industrial inspection tasks. The more sophisticated synchronous shutter is called global, or pipelined, as it allows integration of an image while the

4 previous is still being read out. Such mode of operation is needed for continuous high speed imaging, in applications of motion analysis. An often-disregarded side effect of the snapshot shutter is the presence of a parasitic light sensitivity element in the analog memory: the access switch to the capacitor is a MOSFET, with a potentially light sensitive parasitic diode (Fig. 1). The effect is, that light impinging on the pixel after the integration time is halted, but before the pixel is read, influences the signal. It is a matter of design and technology optimization to make this parasitic diode orders of magnitude less sensitive than the intended photo diode. 3T, rolling shutter operation does not have a light sensitive parasitic node, and hence does not suffer from its parasitic light sensitivity. A last side effect of this 4T operation is that the integrated photo-charge is divided over two nodes by the sample transistor after the integration time. This has effect on the effective full well charge, and also on the read- and reset noises, and on the effectiveness of the FPN cancellation. The S/N ratio of 4T snapshot operation is roughly one half of the rolling shutter operation. The next table summarizes the main operational differences between with modes of operation. 3T pixel or 4T pixel in 3T pixel 4T pixel operation emulation Electronic shutter type olling shutter Triggered, snapshot synchronous shutter Integration while read Yes No Parasitic light sensitivity None About 0.6 % of the normal sensitivity S/N ratio 2000:1 1000:1 Continuous imaging Possible Awkward Start acquisition after a trigger Not possible Possible High dynamic rage mode Double slope operation (*) Multiple slope (Up to four slopes implemented) (*) The IBIS5 device has a simpler high dynamic range mode in rolling shutter compared to snapshot shutter. This difference is only due to the actual implementation of the rolling shutter. 3. HIH DYNAMIC ANE OPEATION In the present context, dynamic range means the range of light intensities that can be handled by the imager within one frame. As the lowest intensity corresponds to the noise level of the pixels, and the highest intensity corresponds to the signal range, for linear response sensors, the dynamic range is equal to the signal to noise ratio (SN). Dynamic ranges can be sufficiently larger than the actual SN, e.g. if there is a non-linear relation ship between both. One such concept is the multiple slope operation. The light power - to - voltage relation is not linear, but piece-wise linear, as shown in the example of next figure. Vout [V] or ADC range Bright part of the scene Dark part of the scene Light intensity axis

5 Figure 4. Example of a 4-part piece-wise linear response slope In this example, the dark part of the scene is smeared over a considerable portion of the ADC range; the conversion slope is the steepest, i.e. the sensitivity and contrast are the highest. Yet, several orders of over-illumination in the bright parts can be handled too. The IBIS5 can capture high dynamic range scenes by subdividing the signal range in 2, 3 or 4 piece-wise linear slopes. The feature makes it possible to project a scene dynamic of up to 100 db on the voltage output range or the ADC range. As illustration, in fig. 6 a high dynamic range scene (sunlight reflecting on the asphalt) is captured without and with double slope operation. What actually is done is that the same pixel is able to subdivide its total available integration time in smaller parts. At the beginning of the integration time, light may be accumulated as in normal operation up to some predetermined level. If the level is not reached (i.e. that part of the scene is dark), this may continue over the full integration time; in the dark parts, the full integration time is available, in other words, we are in the highest sensitivity part, we are in the steepest slope. If the signal level happens to reach a threshold, further integration is suppressed for a while: for signals above the threshold, the effective integration time is thus shorter, and the pixel response is on a flatter slope. EFEENCES 1. B. Dierickx,. Meynants, D. Scheffer, "Near 100% fill factor CMOS active pixels", IEEE CCD & AIS workshop, Brugge, Belgium, 5-7 June (1997); Proceedings p. P1 2.. Meynants, B. Dierickx, and D. Scheffer, "CMOS active pixel image sensor with CCD performance", AFPAEC Europto/SPIE, Zurich, may 1998; proceedings SPIE, vol. 3410, pp , Meynants, B.Dierickx, D.Uwaerts, J.Bogaerts, Fixed pattern noise suppression by a differential readout chain for a radiation-tolerant image sensor, IEEE Workshop on CCDs and AISs, Patents US 6,225,670; US 6,011,251; EP Figure 5. IBIS5 in LCC package Figure 6. The same scene (with different truck!) in single slope and double slope operation