Nsram: The Artificial Neuron on a Silicon Chip (spectrum.ieee.org)

Researchers have made a groundbreaking discovery by identifying an artificial neuron and synapse embedded in standard CMOS transistors, which could significantly enhance the efficiency of neuromorphic computing. Traditionally, artificial neural networks rely on energy-intensive GPUs that mimic biological neurons through software, consuming up to 1,000 watts each. In contrast, the newly discovered device uses the often-overlooked bulk terminal of a transistor to create neuronic behavior, generating a spike in current when a specific voltage threshold is exceeded, akin to biological neurons firing action potentials. This process not only mimics neuronal behavior but does so with less energy. The significance of this breakthrough lies in its potential to revolutionize AI hardware by enabling the production of scalable, efficient neuromorphic systems. Unlike traditional methods that require extensive circuitry to simulate a single neuron, this innovation allows for individual devices to mimic neural functions, promising a substantial reduction in size, cost, and energy consumption. With the capability to integrate these artificial neurons into existing microchip technology, the research paves the way for future advancements in AI applications, potentially achieving efficiencies comparable to the human brain—approximately one million times more efficient than current GPU-based systems for similar tasks.