Architecture Design of a Convolutional Neural Network Accelerator for Heterogeneous Computing Based on a Fused Systolic Array.

Convolutional Neural Networks (CNNs) generally suffer from excessive computational overhead, high resource consumption, and complex network structures, which severely restrict the deployment on microprocessor chips. Existing related accelerators only…

Preview only. Read the full abstract at the source