XNOR-SRAM: In-Memory Computing SRAM Macro for Binary/Ternary Deep Neural Networks

We present an in-memory computing SRAM macro that computes XNOR-and-accumulate in binary/ternary deep neural networks on the bitline without row-by-row data access. It achieves 33X better energy and 300X better energy-delay product than digital ASIC, and also achieves significantly higher accuracy than prior in-SRAM computing macro (e.g., 98.3% vs. 90% for MNIST) by being able to support the mainstream DNN/CNN algorithms.