Digital cameras are fast replacing film cameras as the dominant image capture devices. The design of digital cameras, especially the image sensor part, is experiencing rapid advances aimed at improving image quality and reducing camera cost and power -- with the ultimate goal of replacing film. An important trend is the use of CMOS technology to design single chip digital cameras. This involves the design of CMOS image sensors to replace CCDs and rethinking the camera design to exploit the integration of capture and processing.

The Programmable Digital Camera (PDC) project at Stanford has been exploring new CMOS image sensor and digital camera designs and investigating the performance limits of digital imaging. In this talk, I will focus on a recent study that we performed to determine the "optimal" pixel size for a digital camera image sensor. The image sensor size is typically dictated by the size of the imaging lens used. In this fixed sensor area, we have the flexibility to choose the pixel size and thus the number of pixels (resolution) of the sensor. Using a large pixel is desirable since it results in better individual pixel performance in terms of SNR and dynamic range. On the other hand, using a small pixel results in better sensor spatial resolution and MTF. Therefore it is plausible that an optimal pixel size exists. It is difficult to determine such an optimal pixel size analytically, however, since it not only depends on the sensor physical parameters and the imaging optics, but also on the human perception of image quality. I will first describe the image sensor model we use and show how SNR, dynamic range and MTF vary with pixel size. I will then describe a methodology using a Matlab-based digital camera simulator and the S-CIELAB image quality metric to determine the optimal pixel size. I will demonstrate the methodology for CMOS Active Pixel Sensor (APS) designed in technologies with feature sizes down to 0.18 micron.