Abstract
Abstract Digital pixel sensor (DPS) which includes an A/D converter in each pixel has been developed in the past few years. With pixel level A/D conversion, higher SNR can be achieved, and the small output voltage swing has less impact on imaging quality. Furthermore, by employing A/D converter and memory at each pixel, high speed digital readout can be achieved and therefore provide more rooms for aggressive image-processing applications. Studies adapting multiple imaging capturing or synchronous self-reset schemes were proposed for widening the dynamic range of a linear response pixel. Even though, large pixel area is still the main drawback for most digital pixel sensors, and the massively data processing units required to extend dynamic range increases the complexity of periphery circuits and consumes more dynamic power. In this dissertation, we first presented a clock count output digital pixel sensor which requires only 10 transistors per pixel. The pixel is the clock count output, and is named the CPS. An amplified logarithmic output response similar to the response to light of the human eye is demonstrated by this pixel. The self-offset cancellation scheme of the CPS substantially reduces the FPN caused by the comparator offset drops to less than 2LSB. Moreover, from the estimated results, the dynamic range of the CPS is at 96dB with an 8-bit ADC resolution. Although the CPS can provide a lower supply voltage and a higher dynamic range, the dynamic power consumption and readout speed severely limit the CPS to be applied to high resolution imagers. To improve the dynamic power and readout speed issues, a new architecture of the bit-frame digital pixel sensor is presented. This holds the benefits of the CPS, such as non-linear transfer curve in one time sampling and its dynamic range, which are also proportional to the ADC resolution. The pixel requires 15 transistors per pixel. The readout speed of this pixel is significantly increased and consumes very low power. Moreover, due to its unique characteristics, a simple manner of detecting the object edges of an image is proposed. By applying this edge detection manner, the self-calibration of faulty pixels can also be implemented.