Design and characterization of a wide dynamic range CMOS image sensor with on-chip FPN correction capability

Abstract

In this thesis, a wide dynamic range CMOS image sensor (CIS) system operating in a linear-logarithmic mode has been designed and implemented based on a proposed two-step charge transfer operation. For the emerging imaging markets such as the automotive and surveillance applications, the wide dynamic range (WDR) is one of the most important performance factors in the CIS to capture a scene without any loss of information even in extreme illumination conditions. A linear-logarithmic image sensor has attracted a great deal of attention due to the advantages of single image readout and wide dynamic range performance while having a compatibility with the conventional pixel structure. However, fixed pattern noise (FPN) sets the fundamental limit on performance of a linear-logarithmic image sensor. In a linear-logarithm image sensor, two types of the FPN are usually involved. One is the offset-FPN originated from the threshold characteristic variation and the other is the gain-FPN originated from a subthreshold slope characteristic variation of the logarithmic conversion transistor. Therefore relevant post calibration procedures are usually required for both the FPN sources to acquire high quality wide dynamic range images. To reduce the FPN, image sensors usually accompany in-pixel extra components or complex image processing steps, or even both. In this thesis, in order to realize a low-cost and high performance wide dynamic range image sensor system, new schemes of effective offset and gain FPN correction for the linear-logarithmic image sensor are proposed along with the overall image sensor system development. The first is a two-step charge transfer operation method which is simple yet effective for the offset-FPN correction. The proposed operation has compatibility with the conventional 4-transistor APS structure and only the transfer and readout operation is altered. Therefore the proposed image sensor can be realized without the frame memory or any other c...