Towards lower Uncooled IR-FPA system integration cost

Transcription

1 Towards lower Uncooled IR-FPA system integration cost Benoit DUPONT 1,2,3, Michel VILAIN 1 1 ULIS, Veurey-Voroise, FRANCE 2 Laboratoire d'electronique de Technologie de l'information, Commissariat à l Energie Atomique, Grenoble, FRANCE 3 Institut d Electronique Fondamentale, Université Paris Sud, Orsay, FRANCE ABSTRACT This paper presents the recent progress at ULIS to reduce IR-FPA integration cost for camera manufacturers. The inherent wide offset and responsivity spread of classical uncooled infrared focal plane arrays, leads to complex compensation electronics, making camera integration far more complex and expensive. ULIS low dispersion a Si:H focal plane arrays (FPAs) address already this issue by offering wide dynamic range, low NETD and low cost with no extra custom components. ULIS continues his effort towards even lower signal dispersion. Within this scope, this paper reviews the latest development at ULIS of low dispersion FPA integrated readout circuits and FPA-integrated tools enabling camera manufacturers to improve the image quality. 1 ABOUT FPN IN DIGITAL IMAGER 1.1 FIXED PATTERN NOISE AND VISUAL IMAGE QUALITY Not only infrared imager but all digital imaging arrays suffer from Fixed Pattern Noise (FPN) that degrades the image quality. However when it comes to uncooled infrared arrays, many research document focus on NETD or Detectivity whereas camera manufacturers spend tremendous effort on FPN reduction techniques, especially where Tec-less operation mode is a concerned. The reason for that is FPN is a key figure of Merit for human perception of actual image quality. Many parameters create dispersion from pixel to pixel, from column to column and from chip to chip. Input quantity Detector (FPA) Detector to detector Electrical Column readout Column to column Electrical Formatting circuits (ADC, offset Chip to chip Sensor output Figure 1: Schematic detection chain

2 Dispersion may have various impacts on image quality depending on its source. FPN may for instance be randomly present across the focal plane array as it may vary from column to column. The following pictures are digitally simulated pictures with respectively negligible FPN (Figure 2.a), 2,5 % random PFN (Figure 2.b) and 2,5 % column to column FPN (Figure 2.c). a b c Figure 2: Effect of FPN on image quality It is clear that column-to-column fluctuations are much more damaging for visual image quality. It is indeed shown that human eye is much more sensitive to geometrically patterned disturbances. Straight lines and/or columns are therefore much more annoying than non patterned random dispersion [1]. Of course this does not imply that random FPN is to be neglected in image quality evaluation. Random FPN creates a blurry render of the original image that impacts the effective sensor sharpness. 1.2 FPN IN INFRARED BOLOMETRIC SENSOR Figure 3 represents the signal conversion path from incident infrared radiation to sensor output and the associated possible dispersion sources. It is a particular case of Figure 1 specifically applied to uncooled infrared bolometric focal plane array. Figure 3: Schematic microbolometer detection chain

3 In infrared sensors, random FPN is caused by various dispersed elements. However, one can distinguish between random pixel-to-pixel and column-to-column dispersion sources. 2.1 PIXEL WISE DISPERSION SOURCES 2 PIXEL TO PIXEL DISPERSION Considering figure 3, various dispersions impact the pixel-to-pixel non-uniformity. As far as amorphous silicon devices are concerned, we demonstrated that, even if the 35 µm pitch technology is highly uniform, a potential progress was still existing. 35µm technology non-uniformity has proven to be very low, even without compensation [2][3] with over 425 K scene dynamic, which corresponds in this configuration to a 70 mv output voltage standard deviation under standard biasing and timing. Based on figure 5 schematic, a 6pF capacitance with 2.5V VFID bias and 120us integration time will resume in a responsivity of 6.1mV/K. This means the overall non-uniformity expressed in scene temperature is 11.5K. Figure 4 shows, in comparison with 35µm, the improvement of response uniformity (σ/average) of the new 25µm 160x120 focal plane arrays. As shown, the average detector response dispersion was 1,4 % in It has dropped now to under 1,11 %. This improvement has been achieved with continuously improving αsi:h process control. 1 1,4 1,8 2,2 2,6 % 1 1,09 1,13 1,21 1,25 % 4a: Response uniformity (σ/m) distribution of a population of x 120 detectors, 35 µm pixel pitch (2004) 4b: Response uniformity (σ/m) distribution of a population of x 120 detectors, 25 µm pixel pitch (2006) Figure 4: Comparison of responsivity distribution for 35µm (4a) and 25µm (4b) Blind bolometer Vbolo Active bolometer Figure 5: Schematic pixel layout

4 Such high response uniformity raises the question of the CMOS bias contribution in the sensor fixed pattern noise. Considering the classical readout scheme of uncooled infrared bolometers as described in figure 5, Vth dispersion of the NMOS biasing transistor will impact the FPN. From extracted parameters and based on Pelgrom s work [4], we estimate the threshold voltage (Vth) standard deviation to about 5.1 mv, leading itself to 28 mv output standard deviation under similar bias and timing conditions as seen in 2.1. As a result the CMOS induced output dispersion represent more than 4.59K in apparent scene temperature. Even if quadratic to bolometers dispersion itself, this FPN source is clearly to be considered as a contributor to the final dispersion. 2.2 LOWERING CMOS INDUCED PIXEL WISE FPN In conjunction with the constant effort toward even better microbolometer process uniformity, new product projects are pushed to reduce at the same time readout circuit induced signal dispersion. The circuit presented on figure 6 aims specifically at reducing FPN of bolometric infrared sensors. Basically, the active bolometer is biased with an injection NMOS driven by an amplifier. This structure is also known as BDI (buffered direct injection). Though the use of BDI in bolometer circuits has been published in the past [5] for various other reasons, the innovative principle is here to place the amplifier and the injection transistor in the column structure, in order to escape from area constraints. To compensate for bus access resistance, the amplifier feedback loop is closed via a high impedance line, ensuring proper biasing of the bolometer, regardless its location in the FPA. Besides the 90% drop of the CMOS induced pixel to pixel dispersion, the circuit offers many advantages commonly afforded by BDI structures, such as higher injection ratio and lower noise. Figure 6: Schematic of advanced pixel layout 3.1 COLUMN TO COLUMN FPN SOURCES 3 COLUMN TO COLUMN DISPERSION As shown on figure 5, the compensation of focal plane array temperature is performed thanks to a blind or thermalized bolometer and an injection PMOS. There are here again different possible sources of signal dispersion. The first one is the blind bolometer resistance dispersion. The column structure benefits of course from the high uniformity of the amorphous silicon bolometric process. As discussed previously, in similar chip configuration, the simulated output columnar dispersion based on the usual known resistance distribution for the blind bolometers, leads to a static dispersion of about 28 mv. Another source of signal dispersion to be considered is the PMOS Vth induced bias non-uniformity. In this case Pelgrom s law shows a PMOS Vth standard deviation of about 21.8 mv. As a consequence, the CMOS induced columnar signal dispersion cannot be neglected.

5 Theses contributions to column-wise FPN are to be considered as static, for they depend only on geometrical and deterministic factors. However it is known that 1/f noise contribution to sensor FPN drift after shutter actuation becomes significant [6] over extended periods of time. Although much less annoying than the static dispersion in terms of amplitude, the 1/f noise induced FPN is particularly difficult to handle because of its fluctuating, unpredictable nature. As a consequence it overcomes classical table based offset compensation. 3.2 LOWERING COLUMN-TO-COLUMN FPN As shown on figure 2, the column-wise patterns are very annoying to the end-user and therefore need to be considered with great attention. We have therefore developed a hybrid (circuit and operating method) strategy for column to column FPN cancellation. Basically the focal plane array embeds measurement means able to quantify precisely the static and dynamic columnar dispersion. Based on those accurate data, some adapted software correction can be run to drastically reduce column-to-column FPN. In spite of its on-going state of development, this strategy has already achieved conclusive results, namely 50% reduction of the dynamic column to column FPN, as shown in table 1. FPN performances of bolometric test-chips Standard test chip Test chip with column correction Pixel FPN (µv) Column FPN (µv) Table 1: Dynamic FPN cancellation of column FPN compensation algorithm 4 CONCLUSION In this paper we demonstrated various circuits and methods to reduce CMOS induced column wise dispersion as well as pixel-to-pixel dispersion. Despite already excellent results in terms of dispersion, ULIS continues its effort towards even better image quality. The association of proper CMOS circuits and low dispersion ULIS αsi:h process leads to far lower dispersion across the bolometric array response, without requiring extra external components like DACs, ADCs, Memory, processors, etc 5 REFERENCES 1. F. SAFFIH, R. I. HORNSEY, H. R. WILSON, Human Perception of Fixed Pattern Noise in Pyramidal CMOS Image Sensor, Photonics North, Proceeding of SPIE, Volume 5578, pp , E. MOTTIN, A. BAIN, JL. MARTIN, J.L. OUVRIER-BUFFET, S.BISOTTO, J.J. YON, JL. TISSOT, Uncooled amorphous silicon technology enhancement for 25µm pixel pitch achievement, Infrared Technology and Applications XXVIII», SPIE Vol. 4820, B. FIEQUE, A. CRASTES, JL. TISSOT, JP. CHATARD, S. TINNES, 320 x 240 uncooled microbolometer 2D array for radiometric and process control applications, Optical Systems Design Conference, SPIE 5251, MJM. PELGROM, ACJ. DUINMAIJER, APG. WELBERS, Matching properties of MOS transistors, IEEE journal of solid-state circuits, vol. 24, pp , N. ODA, Y TANAKA, T. SASAKI, A. AJISAWA, A. KAWAHARA, S KURASHINA, Performance of 320x240 Bolometer-Type Uncooled Infrared Detector, NEC research and development, vol. 44, pp , A. FRAENKEL, U. MIZRAHI, L. BYKOV, A. ADIN, E. MALKINSON, Y. ZABAR, D. SETER, Y. GEBIL, Z. KOPOLOVICH, Advanced features of SCD s uncooled detectors, Opto-electronics review 14(1), pp 46-53, 2006