Research on Large Scale SNN Oriented Simulation Optimization Method

Research on neuromorphic computing is becoming popular, which is based on spiking neural networks. There are various spiking neural network simulators available for studying brain mechanisms. However, there is a problem of high GPU memory requirements when simulating large-scale spiking neural networks. This paper proposes a dynamic loading simulation method for large-scale spiking neural networks, which solves the problem of insufficient memory during spiking neural network simulation. This method uses data movement at the sub-network granularity and utilizes the host memory as a larger memory pool to reduce the limitation of GPU memory on the model simulation scale, enabling largescale spiking neural network simulation on a single GPU computer. The pipeline acceleration technique is used to reduce the impact of data movement on simulation speed. Finally, the simulation of a million-scale neural network is achieved in a single GPU experimental environment.