Weekly Neurotech & BCI Digest — August 31, 2026
Intro
The last week of August marks a quiet but structurally significant moment for BCI engineering: research systems are crossing over into real-world deployment at scale. Two major milestones this week — one clinical, one computational — point toward a field shifting from proof-of-concept to practical infrastructure (and toward the hard work of handling neural drift in long-running systems).
Research Highlights
Precision Neuroscience Layer 7: Non-Penetrating BCI Achieves Real-Time Computer Control
Precision Neuroscience announced that its Layer 7 Cortical Interface — an ECoG-style surface array that rests on the brain without penetrating tissue — has enabled participants to make phone calls, send messages, and play Mario Kart in real time using only neural intent. The company is billing this as the first demonstration of full thought-based computer control from a non-penetrating BCI.
The Layer 7 array records from the cortical surface at high density; its minimally invasive profile (inserted through a 2 mm cranial slit) allows temporary intra-operative recordings as well as longer-term chronic use. Precision has been systematically collecting neural data across hundreds of surgical cases, fueling the AI decoders behind this result. The company also appointed BrainGate pioneer John Donoghue as scientific advisor, signalling a push toward rigorously benchmarked chronic performance.
Why it matters for engineers: Non-penetrating ECoG trades single-unit resolution for longevity and reduced gliosis — this result suggests the bandwidth from surface recordings may be sufficient for high-utility everyday computer control, opening a design path that avoids the chronic inflammatory response of intracortical arrays.
UC Davis / Nature Medicine: 19 Months, 3,800 Hours of Independent BCI Use at Home
A paper published in Nature Medicine from UC Davis, Brown University, and Mass General Brigham reports the longest and most intensive independent home use of an intracortical BCI to date. Participant T15, an ALS patient with severe paralysis and dysarthria, used a multimodal BCI for speech decoding and cursor control nearly every day for 19 months — accumulating over 3,800 hours without researcher intervention. The system combines an updated brain-to-text decoder with ventral motor cortex cursor control, achieving stable performance with minimal recalibration.
The decoding architecture builds on prior BrainGate work but adds drift-robust calibration strategies critical for long-term deployment — the same class of “don’t recalibrate constantly” challenges we’ve written about in Conductor. The participant independently used the system to communicate, work, and browse the internet.
Why it matters for engineers: Long-term stability is the field's hardest unsolved problem. 3,800 h of real-world use with minimal recalibration is a new empirical benchmark; the calibration approaches in the supplement are worth reading closely for anyone building production-grade decoders.
Hardware & Devices
ABILITY Neurotech: IMDD Approval for Chronic ECoG Implantation in ALS Patients
Geneva-based ABILITY Neurotech received ethics committee approval (IMDD/MREC NedMec) from the Netherlands to begin a chronic implantation study of its fully wireless ECoG BCI in ALS patients — transitioning from the brief intra-operative recordings performed at TUM Munich earlier this year into months-long continuous use.
The key hardware differentiator: ABILITY's implant uses a transcutaneous optical (infrared laser) link to stream up to 50 Mb/s of raw neural data out of the body for external decoding, rather than performing on-chip compression. The system records from 128 ECoG channels on the cortical surface. The Netherlands trial, coordinated with University Medical Center Utrecht, will have participants train at home over approximately a year, providing a rich dataset for studying decoder adaptation in naturalistic conditions.
Tooling & Datasets
EEG Foundation Model Benchmarks Converge — Three Frameworks at Once
Three independent EEG foundation-model benchmarks landed in rapid succession: NeuroAtlas (evaluating FMs across 17 clinical EEG and BCI datasets), OpenEEGBench (a live community-driven leaderboard via HuggingFace for parameter-efficient fine-tuning, presented at EUSIPCO 2026), and OmniEEG-Bench (standardised evaluation across paradigms). All three address the same gap: EEG-FM papers have been reporting results on inconsistent subsets, making cross-paper comparison unreliable.
Why it matters for engineers: Until now, claiming SOTA on EEG decoding meant picking a friendly dataset. These benchmarks — especially OpenEEGBench's living leaderboard — create shared accountability. If you're building or evaluating an EEG backbone for BCI, these are now the references to track and beat (and if you plan to deploy the winning pipeline on a live stream, make sure it’s built with causal, streaming-safe preprocessing from day one).
Industry & Ecosystem
Neurotech H1 2026: $2.64B Across 81 Funding Events
Neurofounders' H1 2026 funding review tracked 10M rounds funds neuromodulation, diagnostics, and infrastructure. Mid-stage companies in the $20–80M range are thinning — a pattern consistent with the "valley of death" dynamic noted in medtech more broadly.
This week also saw two smaller but notable closes: Aurenar raised $5.7M to develop a neuromodulation device targeting stroke recovery, and ReVision Implant closed €4M (oversubscribed) to advance visual prosthetics toward a new generation of cortical stimulation.
Aleph Neuro Bets on Functional Ultrasound as the Next BCI Modality
Aleph Neuro is pursuing functional ultrasound (fUS) as a non-invasive BCI pathway with higher spatial resolution than EEG. fUS maps cerebral blood flow at ~100 µm resolution — orders of magnitude finer than scalp EEG — but requires a cranial window, placing it between non-invasive and fully invasive modalities. Several groups are now racing to make fUS practical for chronic deployment.
Conclusion
This week's pattern is hard to miss: the field is moving from validating that BCIs work to validating that they work reliably, over time, outside the lab. Precision's non-penetrating result and UC Davis's 19-month home use study both press on the chronic performance question from different angles. The EEG benchmark convergence mirrors a similar maturation in the ML layer. The near-term engineering frontier is decoder stability, hardware longevity, and the calibration infrastructure needed to support months-long autonomous use — especially adaptive, safety-governed updates and confidence-gated decision policies that prevent silent degradation.
📄 Paper of the Week — Long-term independent use of an intracortical BCI for speech and cursor control, UC Davis et al., Nature Medicine (June 2026). The 3,800-hour home-use dataset is likely to become a key benchmark for longitudinal BCI decoder research.
🛠️ Tool Worth Exploring — OpenEEGBench — live HuggingFace leaderboard for EEG foundation model fine-tuning. Open submissions, code at braindecode/OpenEEGBench.
❓ Open Question for Next Week — Precision's Layer 7 result raises a practical question: at what channel count and cortical coverage does a non-penetrating ECoG array saturate for language decoding? The intracortical vs. surface resolution tradeoff now has a real-world performance point to anchor the debate.