Abstract
This paper presents the advanced technologies, including computing-in-memory (CIM) and processing-in-sensor (PIS) techniques, for low-power and low-latency AI edge devices. From system optimization perspective, CIM and PIS techniques are effective by reducing the power dissipation burden of data transmission for the massive computations in convolutional neural network of deep learning model. Furthermore, the realization and performance of CIM and PIS are all strongly relied on the corresponding device development and enhancement.