Neural chip decodes brain networks on 200 microwatts

Most closed-loop brain implants monitor activity in a single area, but psychiatric disorders rarely originate from a single spot. Researchers at EPFL designed an implantable system-on-chip, reported in an EPFL research announcement, that tracks communication across multiple brain regions at once. Dubbed LogSync, the 4-by-2 millimeter silicon chip measures synchrony across brain waves, classifies cognitive and emotional states, and triggers therapeutic electrical stimulation within a power budget under 200 microwatts.

The primary engineering constraint for any neural implant is heat dissipation, where cooking off even a few milliwatts risks damaging adjacent brain tissue. To run feature extraction and on-chip machine learning under that thermal ceiling, the team co-designed the algorithms with logarithmic arithmetic. Converting complex network calculations into the logarithmic domain replaces expensive digital multiplication with simple addition. That keeps the entire 32-channel system running at 196 microwatts, an order of magnitude below conventional multi-region recording setups.

While the machine learning pipeline separated emotional arousal and cognitive task engagement on pre-recorded human data, real clinical deployment is a much steeper hill. Biological neural networks drift over months and electrode-tissue interfaces degrade. Mapping cross-region synchrony to mood episodes outside a laboratory setup remains a challenge. Still, this is a strong demonstration of edge computing where the physical constraints are absolute. Progress in intelligent implants will come from this kind of hardware-level frugality rather than waiting for batteries to get twice as dense.