The motor area is a thin folded sheet on the brain's surface. Different patches become busier before different body parts move, so a hand patch is a useful place to listen for cursor signals.

Your brain has about 86 billion nerve cells, called neurons. They do not send one neat 'move hand' wire; thousands fire in patterns, and the pattern changes with direction, speed and clicking.

The hand patch is large because hands make fine movements. That gives the decoder more detail to learn from than a body part with only a few simple motions.

The spinal cord is a protected bundle of nerve fibres inside the bones of your spine. It carries movement commands down and touch signals back up, so it is the body's main message highway.

Those signals are not metal-wire electricity, but moving voltage waves in living cells. The speed depends on the fibre, and the fastest ones are quick enough for near-instant reactions.

A brain-computer interface does not repair this cable. It makes a detour: brain signals go to a computer, and the computer controls a cursor or another device.

A spinal cord injury can block outgoing movement commands even when the brain's movement-planning circuits still work. The person may still intend a movement, but the signal cannot reach the muscles.

The useful clue is that the plan still exists near its source. If electrodes can listen there, software can send a new command around the damaged path to a cursor or other tool.

This kind of brain-computer interface reads patterns linked to intended movement. It does not read private thoughts, memories or words.

The N1 is designed to sit in a round opening in the skull, with its top level with the surrounding bone. That matters because a bump or cable through the skin would be uncomfortable and risky.

Its case has to be sealed because body fluid and electronics are a bad mix. Neuralink's earlier research package used a titanium enclosure with a parylene coating as a moisture barrier.

The implant does not replace brain tissue. It replaces a small piece of skull above the brain and sends flexible threads down into the outer brain layer.

The motor cortex is not a single button. It is a map-like strip where nearby patches are linked to different body parts, although the map is not as neat as a printed diagram.

For cursor control, useful signals come from hand and arm planning areas. Before surgery, doctors can use scans and medical planning to choose a safe target above those areas.

Listening close to the source gives sharper signals than trying to record through the skull. The trade-off is surgery, which is why this is done only in a clinical trial.

Older research brain implants often used a pedestal through the skin with a cable to lab equipment. That can carry lots of data, but the opening through the skin is a long-term infection risk.

Neuralink's PRIME materials describe an onboard battery, wireless communication to an external device and inductive recharging with the N1 Charger. Inductive means power crosses a small gap using magnetism.

That still does not make it a consumer gadget. It is an investigational medical device being tested for safety and usefulness in people with paralysis.

The PRIME implant uses 64 flexible threads with 1,024 electrodes in total. Each electrode is a tiny conducting pad that senses voltage changes near nearby neurons.

Neuralink's earlier paper describes polyimide threads with gold traces inside. Some versions were only 4–6 micrometres thick, far thinner than a human hair.

Being soft matters because the brain is not fixed like wood. It moves slightly with heartbeat and breathing, so flexible threads should tug less than stiff pins.

Electrodes pick up signals measured in microvolts, which are millionths of a volt. The chip first filters out slow drift and noise, then amplifies the fast spike-shaped signals.

In the 2019 Neuralink paper, each recording channel was sampled about 19,300 times per second. On-board electronics digitized the signal so software could detect spikes with low delay.

Sending every raw measurement would waste power and radio bandwidth. For cursor control, spike events and summaries can carry the important timing information in a smaller stream.

Neuralink says the N1 Implant has an onboard battery that is inductively recharged by an N1 Charger. No metal plug has to cross the skin.

Inductive charging uses two coils. A changing electric current in the outside coil creates a changing magnetic field, and that field makes current flow in the coil inside the implant.

The same physics charges many phones, watches and electric toothbrushes. The implant must manage heat carefully, because brain tissue should not be warmed by waste power.

The 2019 paper describes a tungsten-rhenium needle narrowed to about 24 micrometres along the inserted length. It is thin enough to carry a thread loop but stiff enough to enter tissue.

The needle hooks a loop at the end of one flexible thread. It drives the thread into the cortex, then retracts quickly so the thread stays behind along the path.

This solves a hard problem: the thread is too floppy to push in by itself. The needle acts like a temporary stiff guide, then leaves.

The cortex has tiny blood vessels on its surface. Damaging them can cause bleeding and inflammation, so thread targets are chosen to avoid visible vessels.

The robot uses an optical imaging stack with several cameras and coloured lights. Neuralink's paper says this helps locate thread loops, estimate the cortex surface and guide the needle.

The surgeon is still in charge. The paper describes planning insertion sites in software and allowing manual micro-adjustments before each insertion.

Neuralink's 2019 robot had an autoinsertion mode rated at up to 6 threads, or 192 electrodes, per minute. Real surgery can be slower because the surgeon may pause to avoid vessels and adjust targets.

The same paper reported an average 87.1% insertion success rate across 19 surgeries in animal tests. That is not a promise for every human surgery, but it shows why a robot is useful.

The robot also has to deal with a living brain that moves gently. Cameras and position sensing help it place each thread at the intended depth.

The cell body contains the nucleus, which holds DNA, and it keeps the cell supplied with energy and building materials.

Incoming messages can nudge the neuron toward firing or away from firing. The cell body and nearby axon start area add up those nudges.

If the total crosses a threshold, the neuron fires a spike. If it does not, the neuron stays quiet, but it keeps listening.

Dendrites are the neuron's receiving branches. Other neurons connect to them at tiny junctions called synapses, where chemical signals can open little electrical channels.

A single neuron can receive messages from thousands of other neurons. Some messages push it toward a spike; others calm it down.

This is why one electrode does not record a single tidy thought. It hears nearby cells that are each being pushed and pulled by many inputs.

The axon is the output line. Once a spike starts, it travels along the axon toward the neuron's next partners.

Some axons are wrapped in a fatty insulation called myelin, which helps spikes travel faster. Damage to myelin in some diseases can slow or disrupt messages.

Axons to muscles can be over a metre long, even though the cell body is smaller than a grain of sand. In the cortex, many axons are shorter and connect local circuits.

A spike is a brief voltage swing that usually lasts about 1 millisecond, or one thousandth of a second. It is fast because the membrane gates open and close quickly.

For a given neuron, spikes are roughly all-or-nothing. A stronger message usually means the neuron fires more spikes per second, not that each spike becomes much bigger.

That timing is what the implant cares about. If motor neurons fire more often while the person intends 'move right', the decoder can learn that pattern.

At rest, a neuron keeps more positive charge outside than inside, so the inside is about −70 millivolts compared with the outside fluid.

When the neuron reaches threshold, sodium channels open. Sodium ions rush in because there is both a concentration difference and an electrical pull.

The total swing is roughly 100 millivolts, or one tenth of a volt. Across a cell membrane only nanometres thick, that is a dramatic flip.

Sodium channels close quickly after they open. Then potassium channels open, letting positive potassium ions leave the cell.

That outward flow brings the inside voltage back down. For a tiny moment the neuron is hard to fire again, which helps spikes travel forward instead of echoing backward.

Pumps in the membrane restore the ion balance over time. They use energy, which is one reason the brain needs oxygen and food.

The electrode does not poke into a neuron or read inside it. It sits nearby in salty fluid and senses the voltage change caused by ions moving through the neuron's membrane.

That outside signal is tiny, often tens to hundreds of microvolts, and it fades quickly with distance. Neurons farther away blend into background noise.

That is why flexible threads go into the cortex instead of resting on top of the skull. Close electrodes can hear sharper spike timing.

Each thread carries several electrode pads along its length. Because the pads sit at different depths, one thread can sample a little vertical slice of cortex.

The thread is much narrower than a hair and can fit between cells, but it still displaces some tissue. Making it thin and flexible reduces tugging as the brain moves.

The gold traces carry tiny analog signals back to the implant electronics. There, the signals are amplified before software tries to interpret them.

The chip filters and amplifies the electrode signal, then turns it into digital numbers. In Neuralink's 2019 system, the analog-to-digital converter sampled each channel at 19.3 kHz.

Spike-detection software looks for sharp waveforms that match neuron firing. It does not need to perfectly name every neuron; multi-neuron activity can still control a cursor.

For wireless use, spike times or short spike-count summaries can save power and bandwidth. The laptop needs to know which patches got busier, not every wiggle of raw noise.

Neuralink describes the Link as wirelessly transmitting neural data to the Neuralink Application on an external device, such as a computer. The app then turns that data into actions.

Bluetooth is a short-range radio standard. It is good for low-power links across a room, but it is not meant to blast huge raw laboratory data streams all day.

That is why on-board processing matters. Filtering and summarising spikes before transmission reduces the data that has to cross the skin.

The decoder is trained with examples. The participant imagines or attempts moving while the screen shows known targets, so software can match spike patterns to intended directions.

At first, the computer may guide the task and collect data. As the mapping improves, the person gains more direct control and the decoder updates from successful movements.

The mapping is not permanent. Signals can change as the brain learns, electrodes shift slightly or channels get noisier, so recalibration keeps the cursor controllable.

Noland Arbaugh received the first PRIME Study implant in January 2024 at Barrow Neurological Institute. Neuralink says he could control a laptop cursor from different positions, including lying in bed.

In the weeks after surgery, Neuralink reported that some threads retracted, reducing effective electrodes and lowering cursor speed. The team adjusted recording, cursor translation and interface software, and performance recovered.

Neuralink reported 4.6 bits per second in his first research session and later 8.0 BPS, while engineers using a mouse were around 10 BPS. It is exciting early trial research, not a cure.