The papers that moved the field, organised by theme rather than by date. Each one is briefed the same way: the problem, what was done, what changed, how strong the evidence is and what it could enable. Every figure quoted comes from the published abstract.
Natural conversation combines words with gesture, but speech BCIs and motor BCIs have been built and tested separately.
What did they do?
Recorded from a single high-density ECoG array in three participants with paralysis, decoded isolated arm and face movements, then ran speech and gesture decoders in parallel to drive a virtual avatar.
What changed?
Shows that one cortical implant can serve more than one effector at once, and that training on both isolated and simultaneous attempts improves performance across contexts.
How strong is the evidence?
Peer reviewed, three participants. The system was wired to external processors; quantitative results are in the full paper and are not summarised here.
What could this enable next?
Communication BCIs that carry expression as well as words, and a design argument for broad surface coverage over a single small recording site.
Geographic atrophy destroys central vision and no therapy had been shown to restore it.
What did they do?
Implanted a photovoltaic microarray under the retina in 38 participants; camera glasses project near-infrared images onto it.
What changed?
26 of the 32 participants assessed at 12 months improved by at least 0.2 logMAR with the system on.
How strong is the evidence?
Peer reviewed, open-label, multicentre, baseline-controlled. Serious adverse events related to the procedure or device were reported and are detailed in the paper.
What could this enable next?
The first authorised device for restoring form vision in this condition, CE-marked in 2026.
26 of 32 (81%)
Clinically meaningful improvement at 12 months — 95% CI 64–93
High-bandwidth interfaces have required either open surgery or electrodes that penetrate the brain.
What did they do?
Built a 1,024-channel thin-film array, delivered it through a narrow slit in animal models and cadavers, and tested it intraoperatively in five neurosurgical patients.
What changed?
Shows recording and stimulation from the same electrodes at sub-millimetre scale across large areas, with reversible placement.
How strong is the evidence?
Peer reviewed. Human data are from short intraoperative use, not chronic implantation.
What could this enable next?
Modular coverage of several cortical regions with a less invasive procedure.
Attempting to speak is tiring for people with paralysis, and the prospect of decoding private thought raises a real concern.
What did they do?
Recorded multi-unit activity in four participants, compared inner with attempted speech, and decoded imagined sentences in real time.
What changed?
Found a shared representation plus a distinct 'motor-intent' dimension that separates the two, and showed strategies that stop a speech BCI from picking up private inner speech.
How strong is the evidence?
Peer reviewed, four participants.
What could this enable next?
Less effortful speech BCIs, and concrete engineering answers to a mental-privacy question.
Text output loses prosody, and delays between attempt and sound break the flow of conversation.
What did they do?
Decoded activity from 256 microelectrodes in ventral precentral gyrus straight to voice, giving a man with ALS immediate audio feedback of his own synthesised speech.
What changed?
Moves speech BCIs from producing transcripts to producing a voice the user can hear and modulate as they speak.
How strong is the evidence?
Peer reviewed, one participant. Trained without ground-truth speech audio from the participant.
What could this enable next?
Expressive conversation: emphasis, questions, interjections and singing, not only words.
Earlier speech BCIs needed long training and still made errors in roughly one word in four.
What did they do?
Implanted four microelectrode arrays in the left ventral precentral gyrus of a 45-year-old man with ALS and decoded his attempted speech into text, voiced in a synthetic version of his own voice.
What changed?
Useful on the first day after 30 minutes of calibration, then accurate enough for daily conversation over a large vocabulary for months.
How strong is the evidence?
Peer reviewed, one participant, with 8.4 months of follow-up and more than 248 hours of self-paced use.
What could this enable next?
Speech BCIs judged as assistive devices people rely on, rather than as laboratory demonstrations.
99.6%
Accuracy, 50-word vocabulary — First day of use, 25 days after surgery
97.5%
Accuracy, 125,000-word vocabulary — Sustained over 8.4 months
Naturalistic speed and expressivity had been out of reach for speech neuroprostheses.
What did they do?
Used high-density ECoG over speech cortex in a participant with severe limb and vocal paralysis to decode three outputs at once: text, speech audio and facial-avatar animation.
What changed?
Matched intracortical speed with a surface array, and added voice and face as output channels.
How strong is the evidence?
Peer reviewed, one participant. Decoders reached high performance with less than two weeks of training.
What could this enable next?
Embodied communication, and a credible surface-array route to speech that does not penetrate cortex.
Implanted BCIs had required a craniotomy; whether electrodes could be delivered safely through a vein was unknown.
What did they do?
Placed a stent-electrode array in the superior sagittal sinus via the jugular vein in people with severe upper-limb paralysis and followed them for a year.
What changed?
Established the first prospective human safety data for the endovascular route, with home use for computer control.
How strong is the evidence?
Peer reviewed, single-centre, prospective; five enrolled, four included in analyses.
What could this enable next?
A BCI procedure that existing neurointerventional teams can perform.
4
Participants implanted — Five enrolled
0
Serious adverse events at 12 months — No vessel occlusion or device migration
233 Hz
Mean signal bandwidth — Stable throughout the study in all four
In three participants, overlapping stimulation from several electrodes gives stronger, more localisable touch sensations that stay stable over the study.
Greenspon CM et al. · Nature Biomedical Engineering · Intracortical, ICMS