The Lab Mistake That Might Revolutionize Computing

2026-06-29 · Source: IEEE Spectrum · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Neuromorphic Computing · Depth: Expert, long

Summary

A recent accidental discovery in 2024 by Mario Lanza's lab revealed that a standard MOSFET, fabricated using 180-nanometer node technology, can function as both an artificial neuron and a synapse. This breakthrough, termed neurosynaptic random-access memory (NSRAM), addresses the high energy consumption of current AI systems, where GPUs can draw up to 1,000 watts. The key to this dual functionality lies in manipulating the MOSFET's often-ignored bulk terminal. For neuron-like behavior, a floating or controlled-resistance bulk terminal enables a sudden, nonlinear current spike with hysteresis and self-relaxation, mimicking biological neurons. For synapse-like behavior, manipulating the bulk-source voltage allows for stable, adjustable conductance by trapping charge in the gate dielectric. This single-device approach significantly reduces the component count, replacing dozens or hundreds of MOSFETs per neuron/synapse, and is fully compatible with existing silicon manufacturing lines, demonstrating 100% yield over 10 million cycles. This innovation promises more energy-efficient AI chips, particularly for edge-AI applications.

Key takeaway

For AI Hardware Engineers designing energy-efficient neuromorphic systems, this discovery fundamentally alters component selection. You should investigate integrating single-MOSFET NSRAMs into your designs, as they offer 100% yield and drastically reduce component count compared to multi-transistor implementations. This approach is compatible with existing silicon manufacturing, enabling scalable, low-power solutions for edge-AI and potentially competing with GPUs long-term.

Key insights

A single, modified MOSFET can emulate both neuron and synapse functions, offering a highly energy-efficient and scalable neuromorphic computing solution.

Principles

• Neuromorphic computing seeks brain-like energy efficiency.
• Standard MOSFETs possess hidden neural capabilities.
• Bulk terminal manipulation enables neuron and synapse functions.

Method

By controlling the bulk terminal's resistance or voltage, a single MOSFET can be configured to exhibit neuron-like spiking and hysteresis, or synapse-like adjustable conductance through charge trapping in the gate dielectric.

In practice

• Implement single-MOSFET neurons for edge-AI.
• Design NSRAM circuits for energy-efficient processing.
• Utilize 180-nanometer node MOSFETs for neuromorphic hardware.

Topics

• Neuromorphic Computing
• MOSFET
• Artificial Neurons
• Artificial Synapses
• NSRAM
• Energy Efficiency

Best for: AI Hardware Engineer, AI Scientist, Research Scientist

Related on AIssential

• Indian Researchers Develop Molecular Memristor for Neuromorphic Computing — Molecular memristors offer high-resolution, energy-efficient neuromorphic computing for edge AI by performing in-memory analog computation.
• Exploring the Convergence of Artificial Intelligence and Neuroscience — AI and neuroscience share a symbiotic relationship, driving mutual advancements in understanding cognition and developing intelligent systems.
• Benchmarking the Energy Cost of Assurance in Neuromorphic Edge Robotics — Neuromorphic defenses can enhance AI robustness while simultaneously reducing energy consumption due to induced sparsity.
• Brain-inspired phototransistor could cut AI energy use by sensing and storing data — Brain-inspired phototransistor integrates sensing and memory to reduce AI energy consumption.
• Signed Spiking Neuron Enabled by an Orthogonal-Easy-Axis Magnetic Tunnel Junction — A compact MTJ-based neuron enables signed leaky integrate-and-fire operation, enhancing information density in spiking neural networks.
• Brain-inspired chip runs near absolute zero and could transform quantum computing — A brain-inspired SiC chip operates near absolute zero, enabling integrated quantum computing control and deep space applications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.