Spiking neural networks (SNNs) are artificial intelligence (AI) models inspired by how biological neurons communicate with each other. While biological neurons exchange information in the form of electrical impulses, SNNs rely on brief signals known as spikes.
SNNs have proved promising for reducing power consumption, as developers can ensure they do not process information continuously, but rather only when meaningful changes occur. This could be highly advantageous, as current AI systems are known to consume large amounts of energy.
While some SNNs introduced in the past achieved encouraging results, they typically struggle to retain useful information (i.e., context) for long periods. This was found to be particularly challenging when the models have only a limited amount of data storage available or are operating under energy constraints.
Researchers at Imperial College London and ETH Zurich recently introduced new co-designed hardware and software that could overcome this limitation of SNNs. Their proposed architecture, introduced in a paper published in Nature Machine Intelligence, was found to tackle long-sequence tasks both effectively and energy-efficiently while also reducing data storage requirements.
“Spiking neural networks excel at event-driven sensing,” wrote Pengfei Sun, Zhe Su and their colleagues in their paper. “Yet maintaining task-relevant context over long timescales both algorithmically and in hardware, while respecting both tight energy and memory budgets, remains a core challenge in the field. We address this challenge through an algorithm–hardware co-design effort.”
A novel brain-inspired dual memory system
Past studies have shown that while some neural processes are extremely fast, others are slow and allow the brain to retain information for longer periods. The architecture developed by Sun, Su and their colleagues was designed to artificially emulate this combination of fast and slow neural processes observed in the human brain.
The team co-designed software and hardware components that complement each other, incorporating principles inspired by the brain’s fast-slow organization. The software components consist of a multilayered SNN that combines a slow memory pathway with fast spiking activity.
“At the algorithm level, inspired by the cortical fast–slow organization in the brain, we introduce a neural network with an explicit slow memory pathway that, combined with fast spiking activity, enables a dual memory pathway architecture in which each layer maintains a compact low-dimensional state that summarizes recent activity and modulates spiking dynamics,” wrote the authors.
“This explicit memory stabilizes learning while preserving event-driven sparsity, achieving competitive accuracy on long-sequence benchmarks with 40–60% fewer parameters than equivalent state-of-the-art spiking neural networks.”
A key advantage of the team’s dual-memory architecture is that it can efficiently process incoming data while also retaining important, task-relevant information in a compact form. The team also designed specialized hardware optimized for running their SNN.
“At the hardware level, we introduce a near-memory-compute architecture that fully leverages the advantages of the dual memory pathway architecture by retaining its compact shared state while optimizing data flow, across heterogeneous sparse-spike and dense-memory pathways,” wrote Sun, Su and their colleagues.
Initial results and possible real-world applications
The researchers evaluated their proposed SNN and co-designed hardware in a series of initial tests. They specifically assessed the speed at which their architecture completed long-sequence tasks (i.e., tasks that entail processing long streams of data), its energy consumption and how much data it could process within a set time.
“Experimental results demonstrate more than a fourfold increase in throughput and over a fivefold improvement in energy efficiency compared with state-of-the-art implementations,” wrote the authors. “Together, these contributions demonstrate that biological principles can guide functional abstractions that are both algorithmically effective and hardware-efficient, establishing a scalable co-design framework for real-time neuromorphic computation and learning.”
The new SNN and hardware developed by Sun, Su and their colleagues could soon be refined further and tested on a broader range of computational tasks. In the future, it could potentially enable the rapid analysis of large amounts of data in real time and under energy constraints. This could be valuable for various real-world applications, ranging from robotics to wearable devices, edge AI and networks of multiple connected sensors. https://techxplore.com/news/2026-06-brain-ai-architecture-faster-power.html
Recent Comments