MTJ-Based Nonvolatile Ternary Logic Gate for Quantized Convolutional Neural Networks

摘要

A ternary logic gate based on MTJ-based nonvolatile logic-in-memory (NV-LIM) architecture for ternary neural networks (TNNs) is proposed. The NV-LIM-based implementation achieves reduced computational cost and data transfer cost related to the inference function of deep neural networks. Through an experimental evaluation of a basic component of TNNs, its impact on the energy, delay and area overhead reduction is demonstrated.