Getting your Trinity Audio player ready...
Researchers at the National University of Singapore (NUS) have made a significant breakthrough in neuromorphic computing by demonstrating that a single silicon transistor can mimic neural and synaptic behaviours.
This advancement, led by Associate Professor Mario Lanza from the Department of Materials Science and Engineering at the College of Design and Engineering, brings brain-inspired computing closer to reality and offers a scalable, energy-efficient solution for artificial neural networks (ANNs).
The human brain, with its approximately 90 billion neurons and 100 trillion synaptic connections, is an extraordinarily efficient computing system. Unlike conventional electronic processors, which consume vast amounts of energy, the brain achieves remarkable computational power with minimal energy consumption.
This efficiency stems from synaptic plasticity, where synapses adjust their strength over time, enabling learning and memory. For decades, researchers have sought to replicate this efficiency using ANNs.
While software-based ANNs have driven significant AI advancements, they require substantial computational resources, making them energy-intensive and impractical for many applications.
Neuromorphic computing, which aims to mirror the brain’s efficiency, involves designing systems that integrate memory and computation within the same hardware – also known as in-memory computing (IMC). However, current neuromorphic solutions often rely on complex multi-transistor circuits or novel materials that are difficult to scale for mass production.
Professor Lanza’s team has demonstrated that a single, standard silicon transistor can replicate both neural firing and synaptic weight changes – the fundamental mechanisms of biological neurons and synapses. This was achieved by adjusting the resistance of the bulk terminal, which controls two key physical phenomena within the transistor: punch-through impact ionisation and charge trapping.
Additionally, the team developed a two-transistor cell, called Neuro-Synaptic Random Access Memory (NS-RAM), capable of operating in both neuron and synaptic modes. By leveraging established semiconductor technologies, the research presents a highly scalable and energy-efficient approach to hardware-based artificial intelligence.
Other approaches require complex transistor arrays or novel materials with uncertain manufacturability, but the method developed at NUS makes use of commercial CMOS (complementary metal-oxide-semiconductor) technology—the same platform found in modern processors and memory chips. This ensures scalability, reliability and compatibility with existing semiconductor fabrication processes.
The research offers a viable alternative to existing hardware designs, which often demand intricate architectures and higher power consumption. By reducing the number of components required for neuromorphic processing, this discovery paves the way for simpler, more efficient AI chip designs that can be seamlessly integrated into current manufacturing practices.
Through rigorous experimentation, the NS-RAM cell demonstrated low power consumption, stable performance over multiple cycles and consistent, predictable behaviour across different devices – essential qualities for building reliable ANN hardware. This breakthrough marks a major step in the development of compact, power-efficient AI processors, potentially enabling faster and more responsive computing systems.
As AI applications continue to expand into areas such as edge computing, robotics and real-time decision-making, energy-efficient neuromorphic chips could play a crucial role in reducing the power demands of AI-driven technologies.
This research brings AI hardware closer to mimicking the efficiency of the human brain, a long-standing goal in the field of neuromorphic engineering. By harnessing standard silicon-based transistors, the work at NUS demonstrates a practical and scalable approach to achieving brain-inspired computing.
Future developments could further refine these transistor-based systems, unlocking new possibilities for AI applications that require both high efficiency and minimal energy consumption.
The ability to create neuromorphic processors using well-established semiconductor technology also means that commercial adoption could happen more swiftly than with alternative approaches.
This advancement positions neuromorphic computing as a compelling solution to the growing computational demands of artificial intelligence, opening new pathways for efficient and intelligent hardware development.
