Neural Network with Cascaded Model Dendritic morphologic and FPGA Implementation

The implement of efficient neuron models and neural networks is one of the crucial issues in the field of brain-inspired computing. Considering the state-of-the-art, LIF is a commonly used phenomenological model in brain-inspired computing and HH is a common physiological model. The neuron computing model based on the hierarchical linear model (HLN) can not only reproduce the physiological behavior of neurons, but also consider the simplicity of calculation, which solves the contradiction between high physiological credibility and lightweight calculation of LIF model and HH model. In addition, HLN model can also realize fully-paralleled computing, which can improve the computing efficiency and throughput of neural network. In view of the aforementioned advantages, HLN model is a promising solution to realize a large-scale brain-inspired computing system. In this work, we present a modular design of HLN neurons to implement a cascaded spiking neural network with full parallelism and also a multiplier-less implantation method to minimize the resource consumption. The cascade network consists of HLN neurons are implemented on an Field Programmable Gate Array (FPGA) platform. The results show that HLN model can simulate the dynamic characteristics of the membrane potential in dendrites and soma. Compared with the model before optimization, the multiplier-less and modular implementation proposed in this paper can save 100% DSP resources and 21% logic resources on FPGA, which proves that the multiplier-less HLN model has outstanding performance advantages and is helpful to build a large-scale brain-inspired computing platform.