Invasive BCI

Researchers at imec, Eindhoven University of Technology and Erasmus MC propose a two-stage wireless architecture for intracortical BCIs: a transdural galvanic-coupled body channel link carries data from a free-floating microelectrode array to an intracranial unit, and a transcutaneous link then relays it outside the body. In phantom tests and ex vivo experiments on a human cadaveric head, the transdural link reached 500 Mbps at 20% duty cycling with bit error rates below 10⁻⁵. A built-in send-on-delta encoder (SODA) compresses data by up to 11.4x to cut thermal load, and brain-on-a-chip models showed no unintended neural activity. The study appeared in Communications Engineering on September 1, 2026.

Memory prosthetics — closed-loop brain-computer interfaces that decode hippocampal activity and deliver adaptive stimulation — are moving from animal proof-of-concept toward first-in-human trials, according to a review in iScience. The authors argue that chronically implantable systems require co-design of three subsystems that have been treated in isolation: biocompatible electrode interfaces, on-chip neuromorphic computation, and closed-loop control hardware. The review maps neuroscientific findings such as theta-phase tracking, theta-gamma coupling and sharp-wave ripple detection onto engineering specifications for latency, sampling and charge injection, and onto materials requirements for impedance, switching endurance and chronic stability. It also flags where small-cohort clinical results have been over-generalized.

A proof-of-concept study reports that a non-invasive EEG-based approach can separate hypoglycemic from non-hypoglycemic states in people with type 1 diabetes, with a quadratic discriminant analysis classifier reaching 96.2% accuracy on a limited dataset. Hypoglycemia was accompanied by characteristic changes in the delta and beta bands, which the authors say points to a non-invasive, real-time route to early warning.

Researchers at the University of Naples Federico II in Italy report that the SSVEP-TFFNet deep-learning model outperforms filter bank canonical correlation analysis (FBCCA) at classifying steady-state visual evoked potentials (SSVEP) recorded in extended reality (XR), where headset visuals degrade EEG quality. They used an open XR benchmark dataset of 30 subjects and 1200 trials acquired with Microsoft HoloLens 2. Cutting the montage from 8 channels to 6 or 4 left performance close to the full set, supporting lightweight, wearable XR-BCI designs.

A single-case study of a participant with chronic spinal cord injury found that sense of embodiment was positively associated with frontal theta activity, while simulator sickness was negatively associated with sensorimotor beta activity, during extended reality brain-computer interface (XR-BCI) use. Analyzing 17 XR-BCI sessions with Bayesian correlation and multiple linear regression, researchers at Escola Superior de Saúde do Alcoitão, Universidade de Aveiro and Universidade Católica Portuguesa found simulator sickness to be the only variable independently associated with sensorimotor beta activity, a result they report as robust; the study appeared in Life on August 27, 2026. Different dimensions of subjective experience during XR-BCI operation therefore appear to have partly distinct neurophysiological correlates, a basis for reading user experience from EEG in real time and tuning BCI training and interaction design.

Engineers in the Muller Lab at the University of California, Berkeley have described DustNet, a wireless network of miniaturized ultrasonic implants that acquire and transmit neural signals without wires, in a paper in IEEE Transactions on Biomedical Circuits and Systems. The lab announced the work on its website on August 27, 2026. DustNet follows the lab's earlier MRDust ultrasonic neural interface.

A framework called PQ-NeuroLink adds just 0.45 ms of p95 latency over an unsecured baseline in the most constrained Bluetooth Low Energy single-hop condition, while holding packet delivery at 99.0%. Wireless links between brain-computer interface devices have to be both secure and low-latency, and quantum computers threaten the cryptography they currently rely on. By separating authenticated session establishment from the symmetric streaming path, the framework offers a reproducible communication-layer foundation for secure next-generation BCI deployments.

In cross-subject decoding across simulated multi-level air traffic control tasks, the DMG-GCN model reached 80.30% average accuracy and a 78.63% average F1-score, outperforming state-of-the-art baselines. Built by researchers at Nanjing University of Aeronautics and Astronautics and other institutions, the dynamic microstate-guided graph convolutional network targets the inter-subject variability in controllers' EEG that has held back passive brain-computer interfaces for adaptive automation.

Researchers at Yonsei University in South Korea and the University of Toronto reviewed 64 studies covering 3,103 participants and found wireless EEG increasingly used in research on children with developmental disabilities, with brain-computer interfaces accounting for 28.1% of the included studies. BCI work favored low-channel, dry-electrode, consumer-grade devices, while biomarker-driven studies used higher channel counts and signal fidelity; reporting on data quality was thin, with 85.9% of studies giving no validation against wired EEG and 79.7% not specifying impedance thresholds. Published August 25, 2026 in the Journal of Medical Internet Research, it is the first scoping review to map wireless EEG use across a broad spectrum of developmental disabilities in children.

Arctop has published a companion article to its paper introducing Reinforcement Learning from Brain Feedback (RLbF), a framework that decodes real-time EEG into cognitive states such as workload and stress and uses them as reward signals to train large language models. Unlike RLHF, which depends on sparse, subjective feedback given after the fact, RLbF supplies continuous, involuntary signals that let a model sense in real time how its words land in a listener's brain, which the article says improves communication. It is the first use of brain signals to train a language model and is already running in Arctop's Isaac app, though it adapts on a single dimension, cognitive workload, and technical details are not fully public.

Kandu said its IpsiHand at-home noninvasive brain-computer interface significantly improved upper-extremity motor function in patients with chronic stroke in an NIH-funded randomized controlled trial, with more than half achieving clinically meaningful recovery. The results were published in Stroke: Vascular and Interventional Neurology.

Daniel Palanker, a Stanford professor and affiliate of the university's Wu Tsai Neurosciences Institute, has won the U.S. Defense Health Agency's Outstanding Research Accomplishment award for the PRIMA retinal implant, designed to restore central vision after photoreceptor loss. A 2-mm chip sits beneath the retina and is driven by infrared light from augmented-reality glasses, converting that light into current that stimulates the next layer of neurons. A 2025 paper in the New England Journal of Medicine reported that the device let legally blind patients with advanced dry age-related macular degeneration read letters and words, and the award recognizes its potential for military personnel with laser-damaged retinas.