Case Study of Tactile Sensors: System-level Approach to Analog In-sensor Computing

We propose a novel analog in-sensor computing architecture, enhancing the existing fully parallel analog processing system by incorporating analog circuits to store layer outputs. Through simulations using tactile sensors, we demonstrate a 65% reduction in area and superior power efficiency of 660 TOPS/W, twice as efficient as the previous work. Our findings emphasize the cost-effectiveness and power efficiency of analog in-sensor computing for Convolutional Neural Networks(CNNs) model execution.