[10p-Z09-1~20] 13.1 Fundamental properties, surface and interface, and simulations of Si related materials

▲ [10p-Z09-15] A Simulation Study on the System Performance of Neural Networks using Embedded Nonvolatile Memory

〇(M2)Paul Davin Johansen^1,2, Masaharu Kobayashi^1,2 (1.Inst. of Industrial Science, 2.University of Tokyo)

Keywords:neural network, nonvolatile memory

Hardware neural networks (HNNs) can provide faster training/inference latency and reduced power consumption than neural networks implemented via software. While neural networks have been studied extensively, as of yet, there has been little research on how device configuration and variability can affect performance metrics for neural networks on a system level. In this simulation study, several system performance metrics such as accuracy, latency, and energy consumption for HNNs were analyzed.