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

2025-02-23

AI Summary

The absence of a unified taxonomy for neuromorphic computing on FPGAs has led to fragmented architectural designs, non-comparable evaluations, and poor reproducibility. Method: This paper systematically surveys over 100 FPGA-based neuromorphic architecture studies published between 1999 and 2024, and proposes the first two-dimensional taxonomy—structured along platform heterogeneity and application scenarios—that abstracts core architectural features (e.g., neuron models, synaptic mechanisms, communication paradigms, and mapping strategies). It further defines a quantifiable, cross-platform evaluation metric suite. Contribution/Results: The framework enables standardized representation and fair comparison of FPGA neuromorphic systems, significantly improving reproducibility and collaborative efficiency. It establishes a consensus foundation for brain-inspired chip design, toolchain development, and benchmarking—bridging long-standing gaps in methodology and evaluation across the field.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Machine Learning: Deep Neural Architectures and Foundation Models
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

FPGA-based neuromorphic architectures pose a promising viewpoint on the future of computing, but the efforts in this domain are disorganized, without a clear classification scheme for implemented structures. This results in the lack of consensus on what is important for specific groups of neuromorphic systems, e.g., based on the platform they are implemented on or the intended use case. Here, we review the approach to implementing such architectures on FPGAs over the last 25 years. We propose a new taxonomy for those systems that could help the researchers better understand the closest environment their designs reside in and devise new metrics that could fairly grade them against the other ideas.

Problem

• Lack of organized classification for FPGA-based neuromorphic architectures
• Absence of consensus on key aspects for specific neuromorphic systems
• Need for a new taxonomy and metrics to evaluate neuromorphic designs

Innovation

• Proposes new taxonomy for FPGA neuromorphic systems
• Reviews 25 years of FPGA-based neuromorphic architectures
• Introduces metrics for fair system comparison