Low-Power LogSync Chip Identifies Cognitive and Emotional States

Engineers at EPFL’s School of Engineering and Neuro-X Institute have developed LogSync, a low-power implantable neural system-on-chip designed to read brain signals and identify cognitive and emotional states for future closed-loop neuromodulation therapies. Led by Mahsa Shoaran, head of the Integrated Neurotechnologies Lab, the team created a single chip that records neural activity, extracts biomarkers from oscillatory brain signals, and analyzes synchronization between different brain regions. A machine-learning algorithm then interprets these network-level features to distinguish mental states and trigger stimulation when clinically relevant patterns appear. The system targets limitations in current deep brain stimulation, a therapy already used for movement disorders such as Parkinson’s disease and being explored for psychiatric conditions including depression, posttraumatic stress disorder and obsessive-compulsive disorder. Many existing devices deliver stimulation using predefined settings or rely on signals from a single brain region. Shoaran’s team argues that psychiatric disorders are increasingly understood as disruptions in communication across distributed brain networks, making multi-region biomarker extraction essential for adaptive treatment. Validated in the IEEE Journal of Solid-State Circuits, LogSync was tested using human brain recordings from prior psychiatric experiments. One dataset captured cognitive engagement during a mental task, while another recorded emotional arousal as participants viewed video clips. The chip successfully distinguished these states from neural data alone. Additional rat experiments at the University of Minnesota demonstrated real-time tracking of oscillations and generation of stimulation triggers at predefined brain phases. A central engineering challenge was maintaining performance within the strict power constraints of an implant. By co-designing hardware and algorithms for implant use, the team achieved network-level processing on a 4-by-2-millimeter chip while consuming less than 200 microwatts. LogSync also uses logarithmic computations to simplify the complex operations required for brain-network analysis, biomarker extraction and decoding. Although the initial validation focused on mood and anxiety disorders, the architecture could support broader applications such as adaptive deep brain stimulation for movement disorders, closed-loop noninvasive stimulation and emerging focused-ultrasound neural interfaces. The laboratory is also pursuing self-supervised machine-learning methods that learn directly from neural recordings rather than predefined biomarkers, advancing through recent ICML 2026 work and the EPFL spin-off Infera Neuro.