University of Twente Student Theses


Recurrent spiking neural networks in FPGA for signal processing applications

Sankaran, A. (2022) Recurrent spiking neural networks in FPGA for signal processing applications.

[img] PDF
Abstract:Inspired by the capabilities of a spiking neural network, this work aims to develop an efficient architecture to run an event-based recurrent spiking neural network on an FPGA platform. The availability of the large and fast on-chip memory enables implementation of networks with a large number of neurons and synapses in a parallel manner. This project focuses on improving the existing deep spiking event-based deep belief network architecture named Gyro. Since the initial version of Gyro’s Edge-AI architecture, the scope of improvement has been broad, including but not limited to enhanced functionalities, reduced memory consumption, reduced complexity of the design and computations, improved energy efficiency and reduced power consumption. The resulting architecture was functionally verified on a Register Transfer Level, making the neuron model fully event-triggered and memory efficient, without any redundant computations. The network also supports recurrently spiking network topology. The change in the network at the fundamental level meant that the network inputs and outputs are less complex. The system also attains higher throughput compared to its previous iteration.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:53 electrotechnology
Programme:Embedded Systems MSc (60331)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page