CONVOLUTIONAL NEURAL NETWORK HARDWARE ACCELERATION DEVICE, CONVOLUTIONAL CALCULATION METHOD AND STORAGE MEDIUM

摘要

A convolutional neural network hardware acceleration device, a convolutional calculation method, and a storage medium. The device comprises an instruction processing unit (1), a hardware acceleration module (2), and an external data memory unit (3), wherein the instruction processing unit (1) executes corresponding operations after decoding an instruction set to control the hardware acceleration module (2); the hardware acceleration module (2) comprises: an input buffer unit (21) executing a memory access operation of the instruction processing unit (1) and storing data read from the external data memory unit (3); a data operation unit (22) executing an operational execution operation of the instruction processing unit (1) and processing a data operation of a convolutional neural network, and controlling the data operation process and the direction of data flow according to an operation instruction set; an output buffer unit (23) executing the memory access operation of the instruction processing unit (1) and storing a calculation result that is to be written into the external data memory unit (3) and that is outputted by the data operation unit (22); the external data memory unit (3) storing the calculation result outputted by the output buffer unit (23) and transmitting data to the input buffer unit (21) according to the reading of the input buffer unit (21).