#Singapore #AI Innovation #transistor technology #neuromorphic computing

Revolutionizing Transistors: A Leap Towards Advanced AI

Associate Professor Mario Lanza and his team from the Department of Materials Science and Engineering at the National University of Singapore have made a groundbreaking advancement in neuromorphic computing. Their recent paper, titled "Synaptic and neural behaviours in a standard silicon transistor," was published in Nature on March 26, 2025, and has generated significant interest from major semiconductor companies.

The Challenge of Traditional Computing

Traditional computer architectures suffer from inefficiency, as data must be transferred from the memory to the Central Processing Unit (CPU) for processing. This separation often leads to wasted energy and slower processing speeds. In contrast, artificial neural networks, which use electronic neurons and synapses, offer an innovative solution by allowing data to be stored and processed in the same location. However, creating these electronic neurons typically requires using multiple silicon transistors, which can greatly increase size and cost—up to 18 transistors per neuron and 6 per synapse.

A New Approach to Transistors

The revolutionary insight from Professor Lanza's research is that they can mimic the neural and synaptic behaviours within a single conventional silicon transistor. This is accomplished through a process known as "impact ionization," where they adjust the resistance of the transistor's bulk terminal. By fine-tuning this resistance, they can generate current spikes akin to those observed when an electronic neuron is activated, effectively replicating the behaviour of both neurons and synapses.

Impact on Size and Efficiency

The implications of this discovery are vast. By integrating the functionalities of electronic neurons and synapses within a single traditional transistor, the team has reduced the footprint necessary for these components significantly. As a result, electronic neurons can be condensed by a factor of 18 and synapses by 6. Given that millions of neurons and synapses are present in artificial neural networks, this development could lead to a new era of computing systems that process information more efficiently and with less power consumption.

The Neuro-Synaptic Random Access Memory (NSRAM)

In addition to their single transistor innovation, the team has developed a dual-transistor architecture known as Neuro-Synaptic Random Access Memory (NSRAM), which allows the switching between neuron and synapse modes. This versatility means that manufacturers can produce units that perform both functions using the same physical structure, significantly streamlining production and reducing costs. Furthermore, this new design does not require doping the silicon to modify resistance values, simplifying the manufacturing process.

A Democratizing Technology

Interestingly, the transistors utilized for this groundbreaking work are standard 180-nanometer nodes—not the advanced variants produced in regions like Taiwan or Korea. This accessibility suggests that local Singaporean companies can implement this technology, democratizing access to high-performance electronics.

Dr. Sebastián Pazos, the study's first author, emphasizes a shift in semiconductor research from relentless downscaling towards a focus on efficiency. According to him, the pathway to advanced artificial intelligence does not solely lie in producing smaller components but in innovatively leveraging existing technologies for enhanced performance. He asserts that this approach opens doors for various contributors in the field, even those without access to the latest fabrication technologies.

Conclusion

The work of Professor Lanza and his team marks a milestone in the field of artificial intelligence and neuromorphic computing, promising transformative developments in energy-efficient computing systems. With the ability to replicate complex neural behaviors inside conventional transistors, we may soon see a surge in AI capabilities that reshape the technological landscape. As these innovations proliferate, they herald a new era where advanced AI technology is within the reach of a broader spectrum of society, driving progress across various industries.