Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Quarter of a Century of Neuromorphic Architectures on FPGAs -- an Overview

Wiktor J. Szczerek, Artur Podobas

TL;DR

The paper surveys two decades of FPGA-based digital neuromorphic architectures, introducing a taxonomy of eight NMAs classes defined by three optional traits (fully parallel, collocated memory/computation, asynchronous update). It analyzes 129 FPGA-based NMAs, extracting neuron and synapse models, topologies, and performance trends to reveal design priorities (e.g., Class 2 and Class 4 prevalences, LIF/IZH neuron usage, and on-chip memory emphasis). The study highlights growth in parallel, on-chip architectures and online learning, discusses scaling toward brain-like sizes, and argues for a formal design-space exploration framework to guide future FPGA-NMA development. The findings illustrate the trade-offs between biology-inspired goals and hardware constraints, underscoring the role of FPGAs in advancing scalable, energy-efficient neuromorphic computation for tasks ranging from image inference to neuroscience research.

Abstract

Neuromorphic computing is a relatively new discipline of computer science, where the principles of biological brain's computation and memory are used to create a new way of processing information, based on networks of spiking neurons. Those networks can be implemented as both analog and digital implementations, where for the latter, the Field Programmable Gate Arrays (FPGAs) are a frequent choice, due to their inherent flexibility, allowing the researchers to easily design hardware neuromorphic architecture (NMAs). Moreover, digital NMAs show good promise in simulating various spiking neural networks because of their inherent accuracy and resilience to noise, as opposed to analog implementations. This paper presents an overview of digital NMAs implemented on FPGAs, with a goal of providing useful references to various architectural design choices to the researchers interested in digital neuromorphic systems. We present a taxonomy of NMAs that highlights groups of distinct architectural features, their advantages and disadvantages and identify trends and predictions for the future of those architectures.

A Quarter of a Century of Neuromorphic Architectures on FPGAs -- an Overview

TL;DR

The paper surveys two decades of FPGA-based digital neuromorphic architectures, introducing a taxonomy of eight NMAs classes defined by three optional traits (fully parallel, collocated memory/computation, asynchronous update). It analyzes 129 FPGA-based NMAs, extracting neuron and synapse models, topologies, and performance trends to reveal design priorities (e.g., Class 2 and Class 4 prevalences, LIF/IZH neuron usage, and on-chip memory emphasis). The study highlights growth in parallel, on-chip architectures and online learning, discusses scaling toward brain-like sizes, and argues for a formal design-space exploration framework to guide future FPGA-NMA development. The findings illustrate the trade-offs between biology-inspired goals and hardware constraints, underscoring the role of FPGAs in advancing scalable, energy-efficient neuromorphic computation for tasks ranging from image inference to neuroscience research.

Abstract

Neuromorphic computing is a relatively new discipline of computer science, where the principles of biological brain's computation and memory are used to create a new way of processing information, based on networks of spiking neurons. Those networks can be implemented as both analog and digital implementations, where for the latter, the Field Programmable Gate Arrays (FPGAs) are a frequent choice, due to their inherent flexibility, allowing the researchers to easily design hardware neuromorphic architecture (NMAs). Moreover, digital NMAs show good promise in simulating various spiking neural networks because of their inherent accuracy and resilience to noise, as opposed to analog implementations. This paper presents an overview of digital NMAs implemented on FPGAs, with a goal of providing useful references to various architectural design choices to the researchers interested in digital neuromorphic systems. We present a taxonomy of NMAs that highlights groups of distinct architectural features, their advantages and disadvantages and identify trends and predictions for the future of those architectures.

Paper Structure

This paper contains 21 sections, 15 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Motivation
  3. Method
  4. Background
  5. Field-Programmable Gate Arrays
  6. Spiking Neural Networks
  7. Neuron models
  8. Synapse Models
  9. Proposed taxonomy for Neuromorphic Architectures
  10. Overview of the existing NMAs
  11. Class 0 (Traits: none)
  12. Class 1 (Traits: fully parallel)
  13. Class 2 (Traits: collocated computation and memory)
  14. Class 3 (Traits: asynchronous network update)
  15. Class 4 (Traits: fully parallel, collocated computation and memory)
  16. ...and 6 more sections

Figures (15)

  • Figure 1: Abstract view of an FPGA, showing reconfigurable logic (blue), DSP-blocks (green), BRAM (red), and hardened memory controller and processor blocks. Illustration based on kuon_fpga_2008.
  • Figure 2: Proposed taxonomy - classes.
  • Figure 3: Simplified diagram of a Class 0 NMA.
  • Figure 4: Simplified diagram of a Class 1 NMA.
  • Figure 5: Simplified diagram of a Class 2 NMA.
  • ...and 10 more figures