High accuracy Analog Circuit for Neural Networks computation in parallel CMOS Image Sensors

This paper presents a high accuracy four-quadrant analog multiplication and accumulation (MAC) circuit for neural network computation in parallel CMOS image sensors(CIS). In this design, programmable capacitor array controlled by simple digital circuits is used to combine analog multiplication and accumulation operations. The MAC unit is fabricated using UMC 110nm CMOS process and is powered by a single 3.3V power supply. According to the post-layout simulation, the input and output of MAC unit can be negative and positive, and the dynamic range is 1.5V. MAC units can achieve multipliers between -8 and 8, and provide more than 10 bits of accuracy.