Sponsored
⚡ News

Neural Chip Powers Adaptive DBS for Depression

an abstract image of a sphere with dots and lines

In a lab at EPFL’s School of Engineering, Mahsa Shoaran’s team has spent the last several years trying to solve a problem that has quietly stalled psychiatric neurostimulation for a decade: most deep brain stimulation devices cannot tell whether they are helping. They deliver current on a fixed schedule, hour after hour, regardless of whether the patient is in crisis or asleep. The result, documented across multiple clinical trials of DBS for depression, has been inconsistent outcomes and a therapy that works dramatically for some patients and not at all for others. Shoaran’s lab, the Integrated Neurotechnologies Laboratory within EPFL’s Neuro-X Institute, has now published a chip called LogSync that attempts to close that gap, not by stimulating more precisely in one spot, but by listening to how multiple brain regions talk to each other and firing only when that conversation turns pathological.

The work, reported by neurosciencenews.com and published in the IEEE Journal of Solid-State Circuits, lands as part of a CHF 6.6 million research initiative funded by the U.S. National Institute of Mental Health to build smarter, network-aware neural implants for mood and anxiety disorders that have resisted medication, talk therapy, and even electroconvulsive treatment.

Why a single electrode was never going to be enough

Deep brain stimulation already has a two-decade track record in movement disorders. Surgeons implant electrodes in structures like the subthalamic nucleus, and for Parkinson’s disease or essential tremor, a steady current reliably quiets tremor and rigidity because the dysfunction lives in a fairly contained circuit. Psychiatric illness does not cooperate the same way. Depression, OCD, and PTSD show up in brain imaging and electrophysiology studies as disruptions in communication between distant regions, the prefrontal cortex losing its grip on the amygdala, or reward circuits falling out of rhythm with regions that regulate attention and threat response.

That distinction is why open-loop stimulators, which push constant current whether or not a patient needs it, have struggled in psychiatric trials, and why even the newer closed-loop prototypes have had limited success. Those systems typically watch local field potentials from a single implanted region, a design borrowed directly from motor-disorder DBS. Shoaran’s argument, as described in the EPFL reporting, is that a single listening post cannot catch a problem that is fundamentally about coordination between sites.

Psychiatric disorders arise more from disruptions in communication between brain regions than from dysfunction of any single region.

That single sentence, attributed to Shoaran in the EPFL account of the research, is effectively the chip’s design brief. LogSync samples from multiple electrodes at once and calculates two measurements in parallel: oscillatory power in specific frequency bands, and phase synchrony, a measure of how precisely distant regions are firing in step with each other. Clinicians have long suspected that abnormal synchrony, not abnormal activity in any one spot, is what marks a slide into depressive withdrawal or anxious hyperarousal. LogSync is built to catch that slide as it happens.

The engineering problem was power, not intelligence

Building a classifier that can distinguish a pathological brain state from a normal one is, in isolation, not new. Machine learning models that decode mood and cognitive states from intracranial recordings have existed in research settings for years. The harder problem has been running that kind of computation inside a skull. Multiplication-heavy machine learning algorithms draw power fast, and an implanted chip has nowhere to shed that heat and no easy way to recharge a battery buried under scalp and bone.

Shoaran’s team solved this through what the EPFL reporting describes as algorithm-hardware co-design: building the machine learning model and the physical chip together from the start, rather than designing an algorithm and then trying to cram it onto silicon afterward. The key trick is converting the math into the logarithmic domain, so that operations which normally require energy-hungry multiplication can instead be done with simple addition and subtraction. Cross-frequency coupling and synchrony calculations, normally some of the more computationally expensive operations in neural signal processing, become cheap. The entire system, according to EPFL, runs on under 200 microwatts, a fraction of what comparable implantable neural interfaces typically require.

Network-level neural signal processing and decoding on a single chip, while keeping power consumption under 200 microwatts.

That power budget matters beyond battery life. Heat dissipation inside brain tissue is a real constraint on implant design, and a chip that sips power rather than drawing it in bursts gives engineers more room to add electrodes, more room to run continuously, and less risk to the tissue it sits against.

What the chip has actually been tested on

LogSync has not yet been implanted in a person with depression. The validation so far happened in two stages. First, the EPFL team ran the chip’s decoding performance against existing intracranial recordings from human clinical cohorts, in cases unrelated to the chip’s eventual psychiatric target. In one dataset, LogSync distinguished between levels of cognitive engagement as subjects worked through demanding tasks. In another, it classified emotional arousal as subjects watched emotionally charged video clips, using only oscillatory signal patterns as input, no behavioral or self-report data. Second, working with researchers at the University of Minnesota, the team tested the physical chip in rodents, confirming it could track ongoing brain oscillations and fire stimulation pulses locked to a specific phase of the neural wave cycle in real time.

That phase-locked triggering detail matters to engineers working in this space, because timing stimulation to a precise point in an oscillation cycle, rather than just turning current on or off, is one of the more technically demanding feats in closed-loop neurostimulation. It is also one of the clearest signs that LogSync is a genuinely closed-loop system and not simply a smarter on-off switch.

Approach Sensing Stimulation logic
Open-loop DBS (standard clinical devices) None during stimulation Continuous, fixed-parameter current
Single-site closed-loop prototypes One brain region’s local field potential Triggered by local signal threshold
LogSync (EPFL) Multiple regions, power and phase synchrony On-chip classifier, phase-locked trigger, under 200 microwatts

From a chip on a bench to a device in a patient

EPFL’s reporting makes clear that LogSync is a component, not a finished treatment. Getting from a validated system-on-chip to an implantable product means packaging, biocompatibility testing, chronic safety studies, and eventually a clinical trial in patients with treatment-resistant depression or PTSD, a path that for DBS devices historically runs through years of regulatory review. Shoaran’s lab is already working on the next layer up: self-supervised foundation models trained directly on raw neural telemetry, intended to generalize across patients and electrode placements rather than requiring a bespoke classifier for each implant. That work is set to be presented at ICML 2026. Commercialization is running on a parallel track through Infera Neuro, an EPFL spinout Shoaran co-founded to move network-responsive neural interfaces toward actual patients.

The architecture’s reach extends past psychiatry. The same logic, reading synchrony across a network rather than activity at one point and firing stimulation only when needed, applies to adaptive DBS for the motor fluctuations of advanced Parkinson’s disease, to responsive neurostimulation for epilepsy that does not respond to drugs, and potentially to non-invasive approaches like focused ultrasound that do not require opening the skull at all.

What LogSync actually proves, at this stage, is narrower than a cure for depression: that a chip can watch several parts of a brain argue with each other, do the math for that argument in microwatts instead of milliwatts, and decide in real time whether to intervene. Whether that translates into a device a psychiatrist can implant and a patient can live with normally is still an open engineering and regulatory question, one that Shoaran’s NIMH-funded program is now built specifically to answer.

Source: neurosciencenews.com

Credit: Growtika on Unsplash

adaptive DBSadaptive deep brain stimulation chipclosed-loop DBS depressionEPFL neurotechnologyLogSyncLogSync EPFLlow-power neural chipMahsa Shoaran neural implantneural implant chiptreatment-resistant depression
Facebook
Twitter
LinkedIn
Stay charged
The electric pulse of discovery, in your inbox.

One weekly email. The most fascinating stories at the intersection of biology, electricity, and the future. No noise.