Brain-Computer Interfaces: Neuralink, BrainGate, and Reading Neural Signals

A Paralyzed Man Controlled a Computer Cursor With His Thoughts in 2004

Matthew Nagle, paralyzed from the neck down after a spinal cord injury, became the first person to control a computer cursor, play Pong, and change TV channels using only his brain activity, after receiving a BrainGate implant in June 2004 — a 96-electrode array the size of a baby aspirin placed on the motor cortex of his brain by neurosurgeon Leigh Hochberg at Brown University. Two decades later, that foundational experiment has evolved into a research field that has enabled paralyzed patients to type at 90 characters per minute by imagining handwriting, restored speech to a patient with severe paralysis in 2023, and entered commercial development through companies including Neuralink, Synchron, and Precision Neuroscience.

A brain-computer interface (BCI) is a direct communication pathway between the electrical activity of neurons in the brain and an external device. BCIs bypass the normal neuromuscular pathway — spinal cord, peripheral nerves, muscles — allowing brain signals to control computers, prosthetic limbs, communication devices, or other systems directly.

How BCIs Record Neural Signals

Neural activity generates electrical signals — action potentials (spikes) when neurons fire, and local field potentials (LFPs) reflecting population activity. BCIs capture these signals at different levels of invasiveness and spatial resolution:

BrainGate: The Academic Foundation

The BrainGate consortium — a collaboration between Brown University, Massachusetts General Hospital, Stanford University, and other institutions — has produced the most scientifically rigorous published evidence for implanted BCI performance. Key milestones:

• 2012: A 58-year-old woman with ALS (known as S3) reached out and grasped a bottle with a robotic arm, using a BrainGate Utah Array implant — the first demonstration of BCI-controlled 3D arm movement.
• 2021: A BrainGate participant with paralysis due to spinal cord injury achieved typing speeds of 90 characters per minute by imagining handwriting — decoded by neural population dynamics algorithms — surpassing previous BCI typing speed records by a factor of three (Nature, May 2021).
• 2023: Researchers at UC San Francisco and UC Berkeley reported restoring speech communication to a patient with severe paralysis from a stroke, decoding intended speech from electrocorticography (ECoG) signals with a vocabulary of over 1,000 words (Nature, August 2023).

Neuralink: From Rodents to Human Implants

Neuralink, founded by Elon Musk and Max Hodak in 2016, aims to develop high-bandwidth, chronically implanted BCIs for both medical and eventually cognitive augmentation applications. Its distinguishing technology is a flexible thread-based electrode array (each thread thinner than a human hair) inserted by a neurosurgical robot that avoids blood vessels — intended to reduce the inflammatory response that degrades chronic implants.

• In early demonstrations, Neuralink showed a macaque named Pager playing Pong using a neural implant in April 2021.
• The FDA granted Neuralink Breakthrough Device designation and approved human clinical trials in May 2023.
• In January 2024, the first human patient received a Neuralink N1 implant. The patient, Noland Arbaugh, who is quadriplegic after a diving accident, demonstrated cursor control and computer use within weeks. Subsequent reports noted that a fraction of electrodes had become dislodged from the cortex, reducing performance — though the device remained functional and the patient continued using it.

Synchron's Endovascular Approach

Synchron's Stentrode device takes a fundamentally different approach: instead of brain surgery, the ECoG-like electrode array is delivered via catheter through the jugular vein into the superior sagittal sinus — a blood vessel running over the motor cortex. No craniotomy is required. The trade-off is lower spatial resolution than intracortical electrodes. Stentrode received FDA Breakthrough Device designation in 2020, and human trials (the COMMAND trial) began in the US in 2021. Results published in 2024 showed patients controlling computers and smartphones with the device at home without caregiver assistance.

Signal Decoding: The Machine Learning Layer

Raw neural signals are not directly interpretable as intended actions. Decoding algorithms translate neural population activity into control signals:

• Point process models: Treat spike trains as Poisson processes; historically important for single-unit decoding.
• Kalman filters: Linear models estimating continuous movement from firing rates; used in early BrainGate cursor control.
• Recurrent neural networks (RNNs) and transformers: Deep learning approaches that capture temporal dynamics; enabled the 90-character-per-minute handwriting decoding and the speech restoration results. These models require training on paired neural-behavioral data from the specific patient.

Long-term stability of chronic implants remains the field's most significant unsolved engineering problem. Intracortical electrodes trigger a foreign body response — glial scar formation around the electrode tips — that progressively attenuates signal quality over months to years. Flexible biocompatible materials, coatings releasing anti-inflammatory agents, and closed-loop adaptive decoding algorithms are active areas of research aimed at extending useful device lifespan to decades.