“…The rapid growth of AI applications such as deep neural networks (DNNs) has made traditional accelerators based on the von Neumann architecture face increasingly prominent problems of high energy consumption and delays, which are mainly caused by the large amount of input, weights, and intermediate data that often move between the processor and memory [1,2,3,4,5,6,7,8,9,10,11]. This problem is known as the von Neumann bottleneck or "memory wall" [5,6,7,8,9,10]. IMC architecture is an effective approach to overcome this problem by effectively improving energy consumption and reducing latency by processing data in memory directly in parallel [6,9].…”