Developing a Custom Processor for Efficient Neural Network Operation

Started Mar. 7, 2022

Abstract or project description

The project will be to develop a RISC-V processor that supports efficient neural network operations. Using ISA extensions and/or a co-processor design, we will support vector/matrix operations and activation functions natively. Low-precision operations may be a goal as well. We would like to emulate the processor on Amazon Web Services, using F1 FPGA instances. Along the way, we hope to develop skills on writing good RTL and understand the basics of computer architecture, deep neural networks, and logic design/synthesis. The final goal will be to demonstrate our processor running a complete neural network end-to-end natively and efficiently.