An 8-Bit in Resistive Memory Computing Core With Regulated Passive Neuron and Bitline Weight Mapping

The rapid development of artificial intelligence (AI) and Internet Things (IoT) increase the requirement for edge computing with low power relatively high processing speed devices. computing-in-memory (CIM) schemes based on emerging resistive nonvolatile memory (NVM) show great potential in reducing consumption AI computing. However, inconsistency NVM may significantly degenerate performance neural network. In this article, we propose a RAM (RRAM)-based CIM core to not only achieve efficiency but also greatly enhance robustness by bit line (BL) regulator BL weight mapping algorithm. simulation results that our proposed 8-bit is 12.6 mW ( $256\times 256$ at 8b). spurious-free dynamic range (SFDR) signal noise distortion ratio (SNDR) 62.64 45.92 dB, respectively. scheme improves top-1 accuracy 2.46% 3.47% AlexNet VGG16 ImageNet Large Scale Visual Recognition Competition 2012 (ILSVRC 2012) mode,