FNSim: A Device-Circuit-Algorithm Codesigned Simulator for Flash based Neural Network

FNSim, a device-circuit-algorithm codesigned simulator is developed to evaluate and optimize the performance of flash based neural network. The accuracy, layout area, energy consumption and delay of different flash based neural networks can be evaluated by inputting the dataset, network topology, circuit netlist and device model parameters to the simulator. A convolutional neural network of Lenet-5 is designed to classify MNIST digits as a case study to show how to optimize the performance of flash based neural network with the developed simulator.