A Highly Parallel and Energy Efficient Three-Dimensional Multilayer CMOS-RRAM Accelerator for Tensorized Neural Network

It is a grand challenge to develop highly parallel yet energy-efficient machine learning hardware accelerator. This paper introduces a three-dimensional (3-D) multilayer CMOSRRAM accelerator for atensorized neural network. Highly parallel matrix-vector multiplication can be performed with low power in the proposed 3-D multilayer CMOS-RRAM accelerator. The adoption of tensorization can significantly compress the weight matrix of a neural network using much fewer parameters. Simulation results using the benchmark MNIST show that the proposed accelerator has 1.283× speed-up, 4.276× energy-saving, and 9.339× area-saving compared to the 3-D CMOS-ASIC implementation; and 6.37× speed-up and 2612× energy-saving compared to 2-D CPU implementation. In addition, 14.85× model compression can be achieved by tensorization with acceptable accuracy loss.