EPFL's LogSync implant decodes cross-region brain signals in real time

An implant that watches brain networks, not single regions

Engineers at EPFL have built an implantable system-on-chip, called LogSync, that records activity from multiple brain regions and works out in real time what cognitive or emotional state the brain is in. According to EPFL, the device extracts biomarkers describing both the strength of brain oscillations and the synchrony between signals recorded from different regions, then hands those network-level features to an on-chip machine-learning classifier that can fire stimulation triggers when specific states are detected.

The work, led by Mahsa Shoaran, head of the Integrated Neurotechnologies Lab at EPFL's School of Engineering and Neuro-X Institute, was published in the IEEE Journal of Solid-State Circuits in September 2026. The chip measures roughly 4 by 2 millimetres.

Why cross-region signals matter

The project targets a gap in deep brain stimulation (DBS), the technique that delivers electrical pulses through an implant to modulate neural activity. DBS is already standard for movement disorders such as Parkinson's disease, and researchers are increasingly studying it for severe psychiatric conditions including major depression, post-traumatic stress disorder and obsessive-compulsive disorder.

EPFL notes that most current DBS systems run open-loop: stimulation follows pre-set parameters rather than adapting to the patient's live brain activity. Even adaptive, closed-loop systems often read from a single region — a poor fit for psychiatric disorders, where clinically relevant information is spread across interacting networks. Shoaran frames the research around growing evidence that these disorders stem more from disrupted communication between brain regions than from dysfunction in any one region.

Validation on human data and in rats

The team evaluated LogSync against human brain recordings from earlier psychiatric experiments. One dataset captured shifts in cognitive engagement during a mental task; another captured emotional arousal while participants watched video clips. In both cases the algorithm distinguished the states from the neural recordings alone. In further experiments on rats at the University of Minnesota, the implant tracked ongoing oscillations and generated stimulation triggers at predefined phases of brain activity.

Fitting network analysis into an implant's power budget

Shoaran told EPFL that the hardest constraint was power: implantable systems have very tight energy budgets, and network-level signal processing is computationally demanding. The team's answer was to design the algorithms and the hardware together rather than sequentially, combined with logarithmic arithmetic, which simplifies the operations needed for synchrony extraction and real-time decoding. The result, per EPFL, is a single chip that performs network-level processing and state decoding while consuming under 200 microwatts — a fraction of what many implanted neural-interface systems typically need.

The project sits under a CHF 6.6 million grant from the US National Institute of Mental Health to develop implantable systems that measure and control oscillatory brain networks for severe psychiatric illness, particularly mood and anxiety disorders. Additional support came from a Swiss SERI-funded ERC Starting Grant. The work builds on the lab's earlier research on brain-to-text decoding and closed-loop neuromodulation.

Beyond psychiatry

Although LogSync's initial validation is psychiatric, EPFL says the ability to extract brain-network biomarkers and trigger precisely timed stimulation applies broadly across next-generation neurotechnology: adaptive DBS for movement disorders, closed-loop non-invasive stimulation, and emerging focused-ultrasound interfaces.

The lab is also moving beyond hand-picked biomarkers. A separate ICML 2026 paper from the group uses self-supervised learning to model neural activity across regions, channels and time, and the research feeds into Infera Neuro, an EPFL spin-off Shoaran co-founded.

Why it matters

Closed-loop neurostimulation is widely viewed as the route to psychiatric DBS that adapts to a patient's fluctuating symptoms, but most systems remain blind beyond a single recording site. LogSync demonstrates that network-level analysis — the level at which mood and anxiety disorders appear to operate — can run on silicon small and frugal enough to implant. If the approach holds up clinically, it could make adaptive stimulation viable for conditions where no single electrode placement tells the whole story. It is also an early example of machine-learning inference being squeezed onto sub-milliwatt hardware, a design pattern that is likely to recur wherever AI meets implanted and wearable medical devices.