Low-power-consumption system based on spiking neural network and applied to voice keyword recognition

The low-power-consumption system based on the spiking neural network and applied to voice keyword recognition comprises an external storage, an upper computer and a spiking neural network hardware accelerator, wherein the spiking neural network hardware accelerator comprises a storage module, a controller, a scheduler and a calculation array; the invention provides a two-dimensional coordinate storage structure for storing the spiking neuron state, on one hand, the invalid state of the neuron can be skipped, invalid calculation can be avoided, and on the basis of the sparsity of the spiking neural network, the calculation amount can be greatly reduced, and on the other hand, the time delay for detecting the valid state can be reduced, and the resource utilization rate can be effectively improved; according to the method, a scheme of convolution straight-through pooling and parallel execution is provided, the link of writing back a middle convolution result to be stored in a traditional design is avoided, afte