Accelerating Neural Networks on FPGAs through High-Level Synthesis

Computer hardware is enjoying a widespread renaissance with the emergence of the compute-intensive and challenging machine learning workloads. ABR develops and maintains NENGO, a biologically-plausible model for neural networks and is keen to develop hardware support for efficient realizations of these networks. FPGA (Field-Programmable Gate Arrays), are an attractive target for this if we can overcome the communication bottlenecks that limit the effectiveness of these designs. This project aims to develop strategies for implementing communication between neuron populations mapped onto an FPGA and deploy them to the cloud for demonstration.

Faculty Supervisor:

Nachiket Kapre

Student:

Xinyang Song

Partner:

Applied Brain Research

Discipline:

Engineering - computer / electrical

Sector:

Professional, scientific and technical services

University:

University of Waterloo