Genesis: A Spiking Neuromorphic Accelerator With On-chip Continual Learning

2025-09-06

AI Summary

Addressing continual learning on resource-constrained neuromorphic platforms, this work proposes an online spiking neural network accelerator. Methodologically, it introduces an activity-dependent meta-plasticity mechanism wherein meta-parameters are co-located with synaptic weights in memory, mitigating catastrophic forgetting while improving memory access efficiency. Combined with low-precision learning parameters, customized sparse spike transmission, and an on-chip neuromorphic compute architecture, the design enables energy-efficient continual learning. Experiments on the task-agnostic split-MNIST benchmark achieve a 74.6% average accuracy with only 17.08 mW power consumption under a 65 nm process. The key contributions are: (i) the first hardware implementation of meta-plasticity; and (ii) co-location of synaptic weights and meta-parameters in memory—yielding significant improvements in energy efficiency and continual learning robustness.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Hardware-aware ML
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Continual learning, the ability to acquire and transfer knowledge through a models lifetime, is critical for artificial agents that interact in real-world environments. Biological brains inherently demonstrate these capabilities while operating within limited energy and resource budgets. Achieving continual learning capability in artificial systems considerably increases memory and computational demands, and even more so when deploying on platforms with limited resources. In this work, Genesis, a spiking continual learning accelerator, is proposed to address this gap. The architecture supports neurally inspired mechanisms, such as activity-dependent metaplasticity, to alleviate catastrophic forgetting. It integrates low-precision continual learning parametersand employs a custom data movement strategy to accommodate the sparsely distributed spikes. Furthermore, the architecture features a memory mapping technique that places metaplasticity parameters and synaptic weights in a single address location for faster memory access. Results show that the mean classification accuracy for Genesis is 74.6% on a task-agnostic split-MNIST benchmark with power consumption of 17.08mW in a 65nm technology node.

Problem

• Achieving continual learning in resource-limited artificial systems
• Alleviating catastrophic forgetting with neurally inspired mechanisms
• Optimizing memory access and power for spiking neuromorphic accelerators

Innovation

• On-chip continual learning with spiking neuromorphic architecture
• Activity-dependent metaplasticity to prevent catastrophic forgetting
• Memory mapping technique for faster synaptic access