| CMOS image sensors provide the potential of integrating image capture, analog-to-digital (A/D) conversion, color processing, image enhancement and compression, control, and interface onto a single chip. This can greatly reduce camera cost, power consumption, and size. This thesis has focused on CMOS image sensors with digital pixels. In a digital pixel, A/D conversion is integrated at the pixel level and the pixel outputs digital data. Digital pixels offer many potential advantages over other approaches (namely, analog pixels), such as higher SNR and faster readout. Unfortunately, none of the well established A/D conversion techniques meets the stringent area and power requirements for a pixel level implementation.; In this thesis, we introduce the first viable Nyquist-rate pixel level ADC technique, which we call Multi-Channel Bit Serial (MCBS) ADC. It uses successive comparisons to output one bit at a time simultaneously from all pixels. It can be implemented using simple robust circuits. In addition to overcoming the shortcomings of our previously developed sigma-delta ADC, the MCBS ADC offers several important advantages. The ADC is fully electrically testable without the need for any optics. The signals needed to operate the ADC are globally distributed, thus significantly reducing fixed pattern noise. The ADC can be programmed to implement any quantization table, e.g., corresponding to gamma correction. The most important advantage, however, is the high readout speed, which makes it possible to programmably enhance the sensor's dynamic range via multiple sampling. This is particularly important because of the inherently low dynamic range of CMOS image sensors. As another demonstration of the programmability of our digital pixel CMOS image sensors, we show how we operate MCBS ADC as multiplying ADCs for pixel-parallel separable transforms (e.g. DCT) or convolution.; This thesis describes a 640 x 512 pixel CMOS image sensor with 8-bit pixel level MCBS ADC implemented in 0.35 μm digital CMOS process. We show how by sampling at exponentially increasing exposure times binary floating point output resolution can be obtained. We report measured results from the sensor including QE, FPN, sensitivity, and dynamic range exceeding 16 bits. |
