Two-ring systolic array network for artificial neural networks

摘要

The paper describes an efficient and fast systolic architecture for the implementation of artificial neural networks. The proposed architecture, TORAN (two-in-one ring array network), combines two-ring systolic array networks to perform two sets of weight-data pair multiplication operations in parallel. This increases the hardware by 40% but reduces by 50% the number of cycles to perform the weighted sum calculation compared with a single systolic ring-array network.