A wide-dynamic-range time-based CMOS imager

Abstract

This thesis describes a novel dual-threshold time-based current sensing algorithm suitable for use in wide-dynamic-range CMOS imagers. A prototype 150 x 256 pixel imager employing this algorithm experimentally achieves 95.5 dB dynamic range, 37 dB peak signal-to-noise ratio (SNR), and a highly-linear transfer characteristic while consuming 1.79 nJ/pixel/frame, making it one of the most energy-efficient wide-dynamic-range imagers reported. The individual pixels experimentally achieve 98.8 dB dynamic range and 44 dB peak SNR. The array performance lags slightly behind that of the individual pixels due to the additional noise power contributed to the array data by pixel-to-pixel mismatch effects, attributed primarily to gain and dark-current fixed pattern noise (FPN). The dualthreshold algorithm is also shown to improve low-illumination SNR by 6.1 dB and overall array dynamic range by more than 6.0 dB compared with auto-zeroing alone. The prototype imager implements pixels and their associated 18-bit timing memories in separate on-chip arrays linked by a 200 MHz time-domain-multiplexed communication bus, enabling a pixel pitch of 12.5 pm with 42.7% fill factor in a 0.18-lm 1.8-V CMOS process. Four innovations are contributed by this thesis over previous work, leading to the performance outlined above. First, a novel dual-threshold time-based current sensing algorithm is proposed that forces each single-slope integrating pixel to cross two threshold levels per frame - once just after reset and a second time after a near-optimal amount of photogenerated charge has been collected. This differential measurement technique eliminates offset FPN and pixel reset noise, and reduces comparator 1/f noise. Second, synchronous threshold detection is employed, yielding significant power savings compared with asynchronous approaches in this application, and the resulting time-domain quantization noise introduced by the synchronous detection is analyzed.

(cont) Third, a method of optimizing the global dual-threshold waveform and associated pixel threshold-detection times is presented. The method ensures that the quantization noise introduced by the algorithm remains negligible compared to the intrinsic pixel noise floor, while simultaneously minimizing the number of threshold detections employed, and thus energy consumed. Fourth, a novel capacitively-coupled pixel topology is introduced that enables highly-linear responses to be achieved with this algorithm while minimizing the common-mode input range of the pixel comparator, simplifying its design. Together, these innovations result in energy-efficient wide-dynamic-range pixel operation. The imager is thus suitable for use in portable applications in environments that are challenging for conventional imagers, e.g., when indoor or shadowed lighting and outdoor lighting are simultaneously present in an image.