Particle.news
Download on the App Store

EPFL Chip Decodes Brain Networks to Trigger Adaptive Stimulation

Low power use with on-chip machine learning enables network biomarker decoding for clinical testing of network-responsive deep brain stimulation for psychiatric disorders.

Overview

  • The IEEE Journal of Solid-State Circuits paper published Friday reports LogSync, a 32-channel implantable system-on-chip designed to read multi-region brain signals and generate adaptive stimulation triggers.
  • LogSync extracts spectral power and phase synchrony across channels and runs an interpretable neural additive model classifier on-chip to detect cognitive and emotional states in real time.
  • The chip was fabricated in 65-nanometer CMOS, occupies 5.46 mm², achieves about 6.14 μW per channel and keeps total system power below 200 microwatts to meet chronic implant thermal and battery limits.
  • Validation included classification on human intracranial EEG datasets with reported sensitivity and specificity for cognitive and emotional tasks, plus in vivo rodent tests that demonstrated phase-locked stimulation delivery.
  • The project was developed with CHF 6.6 million in NIMH funding and is moving toward commercial translation through the Infera Neuro spin-off, but authors say clinical trials, safety testing, and regulatory review are still required before patient use.