The Lab Mistake That Might Revolutionize Computing
TL;DR
A lab mistake exposed an odd behavior in a MOSFET: with the bulk terminal mishandled, the transistor produced a sudden current spike and hysteresis, resembling a firing neuron. The crucial piece is the usually overlooked bulk terminal. When charge cannot drain normally, it builds up until a hidden bipolar effect kicks in, then relaxes again. The researchers frame this as NSRAM: MOSFETs could act as both artificial neurons and synapses, using one or two transistors instead of dozens or hundreds of components.
Nauti's Take
This is the kind of breakthrough worth taking seriously, but not worth translating into immediate GPU replacement. The strong part is not just the effect; it is that it appears in ordinary MOSFETs and could fit existing chip manufacturing.
The weak part is the long distance from device demo to useful system. For AI, the bigger point is clear: the next efficiency wave may come less from bigger models and more from hardware that stops pretending a brain is just software running on many transistors.
Briefingshow
AI data centers still depend on GPUs that burn serious power because they simulate neural networks with conventional digital hardware. If standard silicon devices can directly produce neuron-like behavior, this is more than a small efficiency tweak; it points to a different hardware route. The near-term prize is edge AI: smarter battery-powered devices that do less cloud round-tripping.
Sources