An Accuracy-Area Tradeoff for Mixed-Signal Computation in Programmable Smart Image Sensors

Tightly embedded systems often require low power real time processing of acquired sensor data. This paper presents the design and transistor-level implementation of a mixed-signal processing element for highly area-constrained systems in applications where approximate computing is sufficient. The analog processing unit consists of a switched capacitor circuit whose imperfections such as finite gain or offset errors are evaluated here through theoretical analyses and Monte Carlo simulations for a CMOS $0.35\mu m$ technology. Transistor-level simulation results further demonstrate that this processing unit is suitable for image pre-processing tasks such as spatial convolution for edge detection.