Researchers from the National University of Singapore (NUS), led by Associate Professor Mario Lanza, have demonstrated that a single silicon transistor can mimic neural and synaptic behaviors through specific resistance adjustments, enabling phenomena like punch-through impact ionization and charge trapping. This innovation, published in Nature on March 26, 2025, utilizes standard CMOS technology, ensuring scalability and compatibility with existing semiconductor fabrication processes.
The development of a two-transistor NS-RAM cell capable of operating as either a neuron or synapse represents a significant step toward energy-efficient hardware-based artificial neural networks (ANNs), advancing the field of neuromorphic computing.
Researchers from the National University of Singapore (NUS) have made a significant advancement in semiconductor devices for artificial intelligence by demonstrating that a single silicon transistor can mimic both neural and synaptic behaviors. This breakthrough was achieved by operating the transistor in a specific manner, allowing it to replicate fundamental mechanisms of biological neurons and synapses.
The research team led by Associate Professor Mario Lanza utilized standard silicon transistors, adjusting the resistance at the bulk terminal to induce physical phenomena such as punch-through impact ionization and charge trapping. These adjustments enabled the transistor to exhibit behaviors akin to neural firing and synaptic weight changes, essential for neuromorphic computing.
A key innovation from this study is the development of NS-RAM (Neuro-Synaptic Random Access Memory), a two-transistor cell capable of functioning as either a neuron or a synapse. This device leverages commercial CMOS technology, ensuring scalability and compatibility with existing semiconductor fabrication processes. Unlike previous approaches that required complex circuits or novel materials, NS-RAM offers a practical solution for integrating neuromorphic capabilities into conventional electronics.
Experimental results demonstrated that the NS-RAM cell operates efficiently with low power consumption, maintains stable performance over extended cycles, and exhibits consistent behavior across different devices. These attributes make it well-suited for real-world applications in artificial neural networks (ANNs), paving the way for more efficient and responsive AI systems.
Neuromorphic Computing Breakthrough Using Standard Silicon Transistors
The research team at NUS demonstrated that a single silicon transistor can emulate both neural firing and synaptic weight changes by manipulating its bulk terminal resistance. This adjustment triggers specific physical phenomena within the transistor, including punch-through impact ionization and charge trapping, enabling it to replicate key biological processes essential for neuromorphic computing.
NS-RAM, a two-transistor cell, functions as either a neuron or synapse, leveraging existing CMOS technology. This compatibility with conventional manufacturing processes ensures scalability and ease of integration into current electronic systems, facilitating widespread adoption without the need for novel materials or complex circuits.
Experimental results reveal that NS-RAM operates efficiently with low power consumption, maintaining stable performance over extended periods. Its consistent behavior across different devices underscores its reliability, making it suitable for real-world applications in artificial neural networks (ANNs). This advancement holds significant implications for enhancing AI systems by improving energy efficiency and processing capabilities through neuromorphic integration.
Implications for Compact and Power-Efficient AI Processors
The research demonstrates that manipulating the bulk terminal resistance of a single silicon transistor induces specific phenomena such as punch-through impact ionization and charge trapping. These phenomena enable the transistor to emulate neural firing and synaptic weight changes, which are essential for neuromorphic computing.
This development offers a practical solution for advancing ANNs with improved efficiency and responsiveness, potentially revolutionizing AI systems by enhancing energy efficiency and scalability.
More information
External Link: Click Here For More
