March 28, 2025
The GIST Editors' notes
This text has been reviewed in response to Science X's editorial course of and insurance policies. Editors have highlighted the next attributes whereas making certain the content material's credibility:
fact-checked
peer-reviewed publication
trusted supply
proofread
Advancing semiconductor gadgets for AI: Single transistor acts like neuron and synapse
Researchers from the Nationwide College of Singapore (NUS) have demonstrated {that a} single, normal silicon transistor, the basic constructing block of microchips utilized in computer systems, smartphones and virtually each digital system, can perform like a organic neuron and synapse when operated in a particular, unconventional method.
Led by Affiliate Professor Mario Lanza from the Division of Supplies Science and Engineering on the School of Design and Engineering, NUS, the analysis workforce's work presents a extremely scalable and energy-efficient resolution for hardware-based synthetic neural networks (ANNs).
This brings neuromorphic computing—the place chips might course of info extra effectively, very similar to the human mind—nearer to actuality. Their research was revealed within the journal Nature.
Placing the brains in silicon
The world's most subtle computer systems exist already inside our heads. Research present that the human mind is, by and huge, extra energy-efficient than digital processors, due to virtually 90 billion neurons that kind some 100 trillion connections with one another, and synapses that tune their energy over time—a course of often called synaptic plasticity, which underpins studying and reminiscence.
For many years, scientists have sought to copy this effectivity utilizing synthetic neural networks (ANNs). ANNs have not too long ago pushed exceptional advances in synthetic intelligence (AI), loosely impressed by how the mind processes info.
However whereas they borrow organic terminology, the similarities run solely pores and skin deep—software-based ANNs, reminiscent of these powering giant language fashions like ChatGPT, have a voracious urge for food for computational assets, and therefore, electrical energy. This makes them impractical for a lot of functions.
Neuromorphic computing goals to imitate the computing energy and power effectivity of the mind. This requires not solely redesigning system structure to hold out reminiscence and computation on the similar place—the so-called in-memory computing (IMC)—but in addition to develop digital gadgets that exploit bodily and digital phenomena able to replicating extra faithfully how neurons and synapses work.
Nevertheless, present neuromorphic computing programs are stymied by the necessity for classy multi-transistor circuits or rising supplies which might be but to be validated for large-scale manufacturing.
"To allow true neuromorphic computing, the place microchips behave like organic neurons and synapses, we want {hardware} that’s each scalable and energy-efficient," mentioned Professor Lanza.
The NUS analysis workforce has now demonstrated {that a} single, normal silicon transistor, when organized and operated in a particular method, can replicate each neural firing and synaptic weight modifications—the basic mechanisms of organic neurons and synapses.
This was achieved by adjusting the resistance of the majority terminal to particular values, which permits controlling two bodily phenomena going down within the transistor: punch by influence ionization and cost trapping. Furthermore, the workforce constructed a two-transistor cell able to working both in a neuron or a synaptic regime, which the researchers have referred to as "Neuro-Synaptic Random Entry Reminiscence," or NS-RAM.
"Different approaches require advanced transistor arrays or novel supplies with unsure manufacturability, however our methodology makes use of business CMOS (complementary metal-oxide-semiconductor) expertise, the identical platform present in trendy laptop processors and reminiscence microchips," defined Professor Lanza.
"This implies it's scalable, dependable and suitable with current semiconductor fabrication processes."
Via experiments, the NS-RAM cell demonstrated low energy consumption, maintained steady efficiency over many cycles of operation and exhibited constant, predictable habits throughout completely different gadgets—all of that are desired attributes for constructing dependable ANN {hardware} suited to real-world functions.
The workforce's breakthrough marks a step change within the improvement of compact, power-efficient AI processors that might allow sooner, extra responsive computing.
Extra info: Sebastian Pazos et al, Synaptic and neural behaviours in a regular silicon transistor, Nature (2025). DOI: 10.1038/s41586-025-08742-4
Journal info: Nature Supplied by Nationwide College of Singapore Quotation: Advancing semiconductor gadgets for AI: Single transistor acts like neuron and synapse (2025, March 28) retrieved 28 March 2025 from https://techxplore.com/information/2025-03-advancing-semiconductor-devices-ai-transistor.html This doc is topic to copyright. Aside from any truthful dealing for the aim of personal research or analysis, no half could also be reproduced with out the written permission. The content material is supplied for info functions solely.
Discover additional
Crew presents brain-inspired, extremely scalable neuromorphic {hardware} 47 shares
Feedback to editors
