Engineers from the Swiss Federal Institute of Technology in Lausanne (EPFL) have developed an implantable chip called LogSync that can read brain signals and recognize emotional and cognitive states. The work was published in the IEEE Journal of Solid-State Circuits. The chip is intended for future treatments of severe mental disorders such as depression, post-traumatic stress disorder and obsessive-compulsive disorder.
Today, deep brain stimulation (DBS) - a method in which electrical pulses are delivered to certain brain areas through an implant - is widely used for movement disorders, for example Parkinson's disease. However, most existing systems operate in the so-called open loop: stimulation follows preset settings and does not adapt to current brain activity. Even adaptive systems often rely on signals from only one area, which limits their use in psychiatry, where information from a whole network of interacting zones matters.
"Our work builds on a growing body of research showing that mental disorders arise more from disrupted communication between brain areas than from dysfunction in any single area," explains Mahsa Shoaran, head of the Integrated Neurotechnologies Laboratory at EPFL.
The LogSync chip records brain activity and extracts biomarkers - measurable indicators reflecting the strength of brain waves (oscillations) and the synchrony of signals from different areas. A machine-learning algorithm then analyzes these network biomarkers to distinguish cognitive and emotional states and trigger stimulation when they are detected.
The researchers tested the system on human brain recordings obtained in previous psychiatric experiments. One dataset reflected changes in engagement during a mental task, the other emotional arousal while watching video clips. The algorithm was able to distinguish these states from neural recordings alone. In additional experiments on rats at the University of Minnesota, the implant tracked ongoing oscillations and delivered stimulation at specified phases of brain activity.
According to Shoaran, the main difficulty was staying within the strict power limits permissible for an implant. "Our entire system was designed jointly: the algorithms and the hardware were developed with the implant's constraints in mind. This allowed us to process and decode neural signals at the network level on a single chip while keeping power consumption below 200 microwatts - a fraction of what many implantable neural interfaces typically require," she says.
Another feature is the use of logarithmic computing, which simplifies the complex operations needed to analyze brain networks and extract biomarkers in real time.
Although the chip was initially tested for psychiatric tasks, the ability to extract network biomarkers and trigger precisely timed stimulation could also be useful in other areas of neurotechnology: adaptive DBS for movement disorders, non-invasive closed-loop stimulation and new interfaces based on focused ultrasound. In parallel, the laboratory is developing machine-learning methods that learn directly from neural recordings rather than from predefined biomarkers. This work, presented at the ICML 2026 conference, supports efforts to bring intelligent neural interfaces closer to clinical use, including through the startup Infera Neuro, which Shoaran co-founded.