Skip to content
N Neuromorphic Research Nexus

A Brain-Inspired Hardware Architecture for Evolutionary Algorithms Based on Memristive Arrays

A fully memristive spiking neural network with spatiotemporal heterogeneity: each neuron has a distinct, input-adaptive time constant, enabling end-to-end BPTT training without ADCs or comparators, and providing strong robustness to RRAM stuck-at faults.

Evolutionary Memristive Systems Brain-Inspired

View paper View code

Publication

Published in ACM Transactions on Design Automation of Electronic Systems, 2023

Recommended citation: ZiluWang, Xinming Shi, and Xin Yao, “A Brain-Inspired Hardware Architecture for Evolutionary Algorithms Based on Memristive Arrays,” ACM Transactions on Design Automation of Electronic Systems, vol. 28, no. 5, pp. 32, Sep. 2023, doi: 10.1145/3598421.

Abstract

Brain-inspired computing leverages brain principles to develop energy-efficient hardware for sophisticated tasks. This work introduces a hardware architecture for evolutionary algorithms using memristive arrays, enabling sparse and approximate computing through parallel analog computation. Our implementation demonstrates a significant speed improvement of at least four orders of magnitude. Additionally, we investigate fault tolerance and parameter adaptability through grounded simulations. The results show that the system can evolve and adapt to failures or changes in the memristive arrays, highlighting the architecture’s robustness and adaptability.

More projects

Memristive dynamical spiking neural networks with spatiotemporal heterogeneity

2025

A fully memristive spiking neural network with spatiotemporal heterogeneity: each neuron has a distinct, input-adaptive time constant, enabling end-to-end BPTT training without ADCs or comparators, and providing strong robustness to RRAM stuck-at faults.

Spiking Neural Networks Memristive Systems Spatiotemporal Dynamics