IP개요 |
Computing-in-memory (CIM) is a promising approach to reduce latency and improve the energy efficiency of deep neural network (DNN) for artificial intelligence (AI) edge processors. This design presents an energy-efficient static random access memory (SRAM) – CIM unit-macro using: 1) 6T SRAM cells for weighted CIM multiply-and-accumulate (MAC) operation to reduce area overhead and vulnerability to process variation; 2) a current steering digital to analog converter (DAC) and charge sharing multiplication scheme to increase the linearity; 3) a successive approximation analog to digital converter (SAR-ADC) within cell array to reduce area overhead and power consumption. |