Research Hub

Reports 12-month follow-up data from the PRIME feasibility study (NCT05829199). Five participants with tetraplegia received the N1 Implant. Participants achieved median 8.1 bps cursor control, with one participant exceeding 15 bps. Battery life, signal stability, and electrode longevity are detailed. No serious adverse device effects observed over the reporting period.

Second PRIME participant demonstrates 7-DOF robotic arm control using the N1 implant with bidirectional haptic feedback delivered through intracortical microstimulation. Functional grasp tasks including object manipulation were achieved within 3 weeks of implantation. Haptic feedback enabled grasp force modulation and reduced object breakage by 62% compared to visual-only control.

First demonstration of bidirectional neural communication using Synchron's Stentrode platform. Implanted via the internal jugular vein, the device delivered sensory feedback while simultaneously recording motor intent signals from participant M1 cortex. The approach eliminates the need for open-brain surgery while enabling closed-loop stimulation.

Describes the next-generation Layer 7 thin-film ECoG array — a 4,096-channel device with 20 um thickness allowing delivery through a linear craniotomy. Building on intraoperative results in 37 cases, this study presents chronic implant data from 6-month ovine studies and first-in-human chronic recording over 90 days. Broadband gamma signals with single-unit resolution detected in 15% of electrodes.

A transformer-based neural decoder trained on intracortical signals from the hand-knob area of motor cortex achieves 62 words per minute in a phoneme-to-text BCI. The model uses population vectors across 256 electrodes (Utah Array) and runs inference in under 10 ms. Evaluated in a participant with ALS over 18 sessions.

Two-year follow-up of the brain-spine digital bridge in chronic complete spinal cord injury. The original participant maintains independent overground walking with the system active. Neurological recovery has persisted and improved even when the digital bridge is deactivated, suggesting activity-dependent plasticity. A second participant enrolled and achieved independent ambulation at 6 months post-implant.

A closed-loop intracortical microstimulation (ICMS) system delivers biomimetic touch feedback in real time, linked to pressure sensors on a prosthetic hand. Participants with cervical SCI achieved naturalistic texture discrimination at 81% accuracy. The system uses adaptive stimulation parameters that adjust based on decoded grasp force.

Building on the 2023 avatar-controlling speech neuroprosthesis, the Chang lab demonstrates an upgraded system achieving 85 words per minute using a 253-electrode ECoG array combined with a fine-tuned large language model decoding pipeline. Emotional prosody classification reaches 74% accuracy across 6 emotion categories, enabling more natural and expressive communication.

Graphene microelectrode array demonstrates impedance stability over 18 months in primate cortex. The transparent electrodes enable simultaneous two-photon optical imaging and high-density electrophysiology. The flexible substrate conforms to cortical curvature, reducing chronic tissue reaction compared to rigid silicon arrays.

Longitudinal analysis of Utah Array (Blackrock Neurotech) performance in the longest-running implanted BCI study to date. Signal quality, firing rate, and decoding accuracy tracked across 15 years in two participants. Initial signal degradation stabilizes at year 3-4, with consistent performance maintained through year 15 in participant T5.

The FDA establishes the first regulatory classification pathway for endovascular brain-computer interface devices. The De Novo classification creates a new product category for minimally invasive neural interfaces, setting performance and safety standards that will apply to future endovascular BCI submissions. This landmark regulatory decision follows successful Synchron COMMAND trial results.

First long-term home-use wireless BCI system deployed for daily use by 3 ALS participants. The fully wireless 200 Mbps intracortical system achieves signal quality equivalent to the wired laboratory system with 48-hour continuous recording demonstrated. Over 2 years, participants averaged 5.3 hours of daily BCI use with 97.2% system reliability under caregiver-managed operation.

A large foundation model pretrained on over 1 million hours of EEG recordings from 10,000 participants enables zero-shot classification of motor imagery tasks across unseen participants. The model achieves 71% accuracy on 4-class motor imagery without any participant-specific calibration, dramatically reducing setup time for non-invasive BCI systems.

The FDA grants Breakthrough Device designation for the Neuralink N1 implant with wireless charging, establishing an accelerated review pathway for pivotal trial expansion to 30 participants. The designation recognizes the device's potential to provide more effective treatment for individuals with tetraplegia and ALS compared to existing assistive technologies.

A flexible photonic probe enables simultaneous optogenetic stimulation and single-unit recording in non-human primate motor cortex. The device delivers patterned light stimulation to specific neuronal populations while recording from neighboring neurons, enabling cell-type-specific interrogation of motor circuits. Chronic stability demonstrated over 9 months with minimal tissue reaction.

A semantic decoder reconstructs continuous natural language from fMRI recordings during story listening and silent speech. The decoder achieves BLEU scores of 0.42, capturing the meaning and structure of perceived and imagined language. The system generalizes across story-listening and silent speech paradigms, suggesting shared neural representations for perceived and produced language.

Building on prior phoneme-decoding work, this study adds a language model postprocessor and demonstrates real-time BCI-assisted speech at 62 words per minute in three participants with ALS. End-to-end latency from neural signal to spoken audio output is under 10 ms. The system handles naturalistic conversation, including interruptions and emotional intonation.

Closed-loop hippocampal stimulation based on a multi-input multi-output (MIMO) model improves short-term memory recall by 35% and long-term memory by 28% in 24 epilepsy patients with implanted depth electrodes. The system records hippocampal activity during encoding, computes the optimal stimulation pattern via the MIMO model, and delivers targeted electrical stimulation to reinforce memory traces.

First pediatric BCI implant study enables communication in 3 children (ages 8-14) with locked-in syndrome from brainstem pathology. A subdural ECoG array over sensorimotor cortex decoded attempted movements for binary communication, achieving reliable yes/no communication within 2 weeks. An adaptive decoding algorithm accounts for the developing brain's plasticity and shifting neural representations.

A GPT-4 class large language model used as a BCI postprocessor reduces word error rate from 5.2% to 0.8% across open-vocabulary conversational speech. The LLM operates on the neural decoder's phoneme probability outputs and applies contextual correction in real time with latency under 50 ms, making it practical for continuous conversation.

Reports the first long-term home-use BCI outcome data: 3 ALS participants used intracortical BCI daily for an average of 5.3 hours per day over 2 years in their own homes. System reliability was 97.2% with all operation managed by trained caregivers. Participants used the BCI for email, web browsing, entertainment, and communication with family members.

A hybrid fNIRS-EEG system combining 256 optical channels and 128 EEG channels achieves 85% accuracy on 4-class motor imagery without any user training. The portable headset weighing 340g makes the system practical for real-world use. Multimodal fusion of hemodynamic (fNIRS) and electrophysiological (EEG) signals provides complementary information that outperforms either modality alone.

Intracortical microstimulation of primary visual cortex (V1) produces stable phosphene patterns in 6 participants with acquired blindness. A 96-electrode Utah Array implanted in V1 generates up to 88 distinguishable phosphenes. Participants identify simple shapes (letters, geometric forms) at 72% accuracy and navigate obstacle courses using the phosphene-based visual display.

A bioresorbable silicon-based neural electrode array provides clinical-grade ECoG monitoring during the critical post-surgical window and then dissolves safely over 6 weeks without requiring a removal surgery. Tested in 8 patients following brain tumor resection, the device detected seizure activity with sensitivity equivalent to standard subdural electrodes.

First published results from the Neuralink PRIME trial. Participant Noland Arbaugh, with C4 complete tetraplegia, achieved 6.7 bits per second cursor control using the fully implanted N1 device with 1,024 electrode threads. At 6 months, 85% of threads remained functional. Thread retraction was observed in the first weeks but stabilized. No serious adverse device effects reported.

Demonstrates that activity-dependent spinal cord stimulation paired with intensive rehabilitation promotes axonal sprouting and synaptogenesis in the injured spinal cord. Three participants with chronic complete SCI showed neurological improvement of 8+ ASIA motor score points. Critically, these improvements persisted 12 months after stimulation was discontinued, indicating structural neuroplastic changes.

Explored the geometry of neural population activity in motor cortex during attempted speech, demonstrating that the neural manifold has consistent geometric structure across sessions and participants. Leveraged this geometric consistency to build a speech neuroprosthesis that calibrates rapidly (minutes rather than hours) and generalizes across days with minimal retraining. Achieved 97.5% accuracy on a 50-word vocabulary with under 4 minutes of calibration data.

Reports pivotal results from the Synchron COMMAND trial. Six ALS participants with the Stentrode endovascular BCI achieved independent texting, email composition, and web browsing. Median daily device use was 4.2 hours at 12 months. No serious device-related adverse events were observed. The endovascular approach via the jugular vein required no craniotomy.

Stable Diffusion conditioned on fMRI-derived latent representations reconstructs viewed images with 78% structural similarity (SSIM). The method maps visual cortex activity patterns into the latent space of a pretrained diffusion model, enabling high-fidelity image reconstruction without task-specific training. First demonstration of photorealistic visual reconstruction from non-invasive brain recordings.

Initial human intraoperative results for the Layer 7 thin-film ECoG array. The device was placed on the cortical surface of 12 neurosurgery patients during planned procedures. All 1,024 channels recorded broadband neural signals with single-unit resolution in select electrodes. The device was deployed and cleanly removed in under 3 minutes, demonstrating the practical surgical workflow.

Non-invasive MEG decoding of perceived speech achieves segment-level accuracy of 73% using a deep neural network trained on 169 hours of MEG data. The model generalizes across 5 languages (English, French, Mandarin, Dutch, German), suggesting it captures language-universal auditory and phonological representations rather than language-specific features.

A self-supervised domain adaptation method enables stable BCI cursor decoding across 30+ consecutive days without any manual recalibration. The decoder uses a stabilization algorithm that detects and corrects distribution shifts in neural population activity caused by electrode drift, allowing sustained high performance despite chronic changes in recorded neural signals.

A MIMO (multi-input multi-output) computational model of hippocampal CA1-CA3 dynamics enables targeted electrical stimulation that improves episodic memory encoding by 37% in 15 epilepsy patients during free recall tasks. The system records neural activity patterns during successful encoding, derives the optimal stimulation pattern, and delivers it during subsequent encoding attempts to reinforce memory traces.

First demonstration that a human with tetraplegia could use an intracortical BCI to control a computer cursor, open email, operate a television, and control a robotic hand. Participant Matthew Nagle, paralyzed by a spinal cord injury, used a 96-electrode Utah Array implanted in primary motor cortex (M1). This paper established the feasibility of human intracortical BCI and launched the BrainGate clinical program.

Two participants with long-standing tetraplegia used intracortical BCI (Utah Array in M1) to control a robotic arm for self-directed reaching and grasping. Participant S3 (Cathy Hutchinson, 58 years old, paralyzed for 15 years) successfully grasped a bottle of coffee and brought it to her mouth — the first self-feeding with a neurally controlled robotic arm. Demonstrated multidimensional (3D + grasp) BCI control in a real-world task.

A participant with tetraplegia (BrainGate T5) imagined writing letters by hand while intracortical electrodes in motor cortex recorded the associated neural activity. A recurrent neural network decoded the imagined pen trajectories into text at 90 characters per minute (approximately 18 WPM) with 94.1% raw accuracy (99% with autocorrect). This was more than double the previous BCI typing speed record and demonstrated that attempted handwriting produces rich, decodable neural signals.

Decoded attempted speech from intracortical recordings in motor cortex of a participant with severe dysarthria from ALS. An RNN decoder converted neural activity into phonemes at 62 words per minute with 23.8% word error rate (reduced to 9.1% with a language model). The system decoded a vocabulary of 125,000 words in real time. Established that the speech motor cortex retains rich articulatory representations even after years of impaired speech.

First demonstration of a speech neuroprosthesis: an ECoG array implanted over speech motor cortex of a participant with anarthria (complete loss of speech from brainstem stroke) decoded attempted speech into sentences. The system decoded 50 words at up to 15.2 words per minute with a median word error rate of 25.6%. This proved that the speech motor cortex retains usable representations of intended speech even in a person who has not spoken for over 15 years.

A participant with severe paralysis from brainstem stroke used a 253-electrode ECoG array over speech motor cortex to generate text at 78 words per minute (median 28% WER) and simultaneously control a digital avatar displaying facial expressions and emotional prosody. Integrated a large language model to boost decoding accuracy. Demonstrated that speech BCIs can restore not just words but expressive, embodied communication.

First-in-human results for the Stentrode — an endovascular BCI implanted via the jugular vein into the superior sagittal sinus adjacent to motor cortex. Two participants with ALS used the device to control a computer for texting, emailing, online shopping, and banking without open brain surgery. The Stentrode demonstrated that motor cortex signals can be recorded through the blood vessel wall with sufficient quality for functional BCI control.

A "digital bridge" brain-spine interface enabled a man with chronic spinal cord injury to walk naturally over ground. Cortical implants (Utah Arrays) in motor cortex decoded walking intentions in real time, which wirelessly controlled an implanted epidural spinal stimulator (ONWARD ARC-IM) below the injury. The participant regained the ability to stand, walk, climb stairs, and traverse complex terrain. Remarkably, neurological recovery persisted even when the digital bridge was turned off.

First communication by a patient in a completely locked-in state (CLIS) — total loss of all voluntary muscle control including eye movement. Two Utah Arrays implanted in motor cortex recorded neural signals that the patient learned to modulate via auditory neurofeedback. After extensive training, the patient spelled sentences at approximately 1 character per minute. This demonstrated that BCI communication is possible even when all other communication channels have failed.

Introduced LFADS (Latent Factor Analysis via Dynamical Systems), a deep learning method for extracting low-dimensional neural population dynamics from single-trial neural recordings. LFADS uses a sequential variational autoencoder to denoise neural population activity and infer the latent dynamical system driving it. Demonstrated dramatic improvements in the ability to extract meaningful signals from noisy neural data, with applications across motor cortex, somatosensory cortex, and other brain areas.

Translated the ReFIT (Recalibrated Feedback Intention-Trained) Kalman filter from animal studies to human BCI use. The ReFIT algorithm retrains the decoder using the user's inferred intention (toward the target) rather than the noisy observed cursor trajectory, correcting a fundamental mismatch in standard decoder calibration. In two BrainGate participants with tetraplegia, ReFIT approximately doubled point-and-click performance compared to the standard Kalman filter.

Demonstrated that intracortical BCI users could operate an unmodified Android tablet computer for point-and-click typing, web browsing, email, music streaming, and other everyday computing tasks. Three participants with tetraplegia achieved typing speeds up to 40 correct characters per minute using a standard on-screen keyboard. This moved BCI from custom laboratory software to real-world consumer technology interfaces.

Developed and demonstrated methods for rapid BCI decoder calibration in human participants with tetraplegia. Instead of requiring 10-30 minutes of calibration at the start of each session, the system achieved usable performance within 30-60 seconds using either retrospective target inference or neural replay of imagined movements. This addressed a major practical barrier to everyday BCI use — the lengthy daily calibration process.