Title

FPGA based high throughput low power multi-core neuromorphic processor

Author

Yangjie Qi

Date of Award

2015

Degree Name

M.S. in Electrical Engineering

Department

Department of Electrical and Computer Engineering

Advisor/Chair

Advisor: Tarek Taha

Abstract

The interest in specialized neuromorphic computing architectures has been increasing recently, and several applications have been shown to be capable of being accelerated on such platforms. This thesis describes the implementation of multicore digital neuromorphic processing systems on FPGAs. Static and Dynamic routing were used to allow communication between the cores on the FPGA. Several applications were mapped to the system including image edge detection, MNIST image classification, and biometric ECG classification. Given that all the applications were implemented on the same processor (hence same base Verilog code), with only a change in the synaptic weights and number of neurons utilized, the system has the capability to accelerate a broad range of applications.

Keywords

Neuromorphics, Neural networks (Computer science), Adaptive routing (Computer network management), Field programmable gate arrays, Electrical Engineering, Computer Engineering, FPGA, Neural Network, Static Routing, Dynamic Routing, Low Power, High Throughput

Rights Statement

Copyright 2015, author

Share

COinS