Thought to Cursor: How a Neuralink brain implant works

BIOLOGY · 4 MIN

How can someone move a cursor just by thinking?

A man who can't move his hands now plays chess on a laptop with his thoughts. No brain science needed: we'll start with how your body sends messages and zoom all the way down to one tiny brain cell.

8 steps · 9 quick challenges

Step 1 of 8 · The idea

Messages from the brain

Wiggle your fingers. That tiny movement started as a message in your brain. Let's follow it.

Every move you make starts as a plan in your brain. The plan travels down to your muscles as a fast electrical message. If that path breaks, the plan is still there, but it can't get out.

  1. Movement area. A strip across the top of your brain, called the motor area, plans every movement. When you decide to move your hand, cells in this strip start sending messages. Like a control room with one switch for each part of your body.

    Your brain has about 86 billion nerve cells, called neurons. They talk using tiny electrical pulses.

    Each body part has its own patch of the motor strip. The hand's patch is surprisingly large, because hands make such fine movements.

  2. The cable. The messages race down your spinal cord, a thick bundle of nerves inside your backbone. From there, nerves branch out to every muscle. Like a big internet cable running down your back, with smaller wires branching off to each room.

    Messages run down this cable at up to about 100 metres per second, which is why your fingers move the instant you decide.

  3. The break. If the spinal cord is badly hurt, messages can't get past the injury. The brain still makes the plan, but the muscles never hear it. That's called paralysis. Like cutting a phone line: you can still talk into the phone, but nobody hears you at the other end.

    Here's the clever idea: the plan still exists in the brain. If we could listen to it directly, we could skip the broken cable.

    That's what a brain–computer interface does: it listens to the brain and sends the message to a computer instead.

Neurons
≈86 billion
Brain power
≈20 W

Did you know? Your brain uses about 20 watts of power, less than most light bulbs, yet it runs your whole body.

Read more

Your body runs on messages. When you decide to pick up a cup, neurons in your brain fire in a pattern that means 'move the hand like this'.

That pattern travels down the spinal cord and out along nerves to your arm. The muscles squeeze, and your hand moves. It all happens in a fraction of a second.

After a bad spinal injury, the path is blocked. The muscles are fine and the brain is fine, but the messages can't get through. A brain implant tries to catch those messages before they reach the break.

Just imagining a movement activates many of the same neurons as really moving. That's what makes a thought-controlled cursor possible.

Step 2 of 8 · The idea

A coin in the skull

So the plan is still there, inside the brain. To hear it, we need a listener right on top of the motor area.

Neuralink's implant, called the Link, is about the size of a large coin. Surgeons replace a coin-sized piece of skull with it, right above the brain's hand area.

  1. Coin-sized. The Link is a sealed disc about 23 mm across and 8 mm thick. It sits flush in the skull, so once the skin heals over it, nobody can see it. Like swapping a round piece of a dinner plate for a coin that fits the hole perfectly.

    The case is sealed so body fluids can't get in and damage the electronics. The version in the 2019 paper used titanium coated in a thin plastic layer.

  2. Hand area. It goes over the part of the motor area that controls the hand. Hand movements are what you need to steer a cursor, so that's the best place to listen. Like putting a microphone right next to the singer, not at the back of the hall.

    Brain scans before surgery show exactly where each person's hand area is, so the surgeons know where to place it.

  3. Under the skin. Nothing pokes out of the head. No wires, no plugs. The Link charges wirelessly and talks to a computer by radio. Like a phone with wireless charging and Bluetooth, sealed inside your head.

    Older brain implants used a plug sticking out of the skull, with a cable to a computer. The skin around the plug could get infected. Being fully sealed in avoids that.

Size
≈23 mm wide
Visible
No

Did you know? The US FDA allowed Neuralink to start human trials in May 2023. The first person got an implant in January 2024.

Read more

Your skull is a hard shell of bone, thicker than a pencil. Under it, protective layers wrap the soft, wrinkly brain.

To fit the Link, a surgeon removes a coin-sized circle of skull. The implant drops into the hole so its top is level with the bone around it.

Then the skin is closed over it. From the outside, the person looks just as before.

The implant listens to the brain's outer layer. It never goes deep into the middle of the brain.

Step 3 of 8 · The device

Inside the coin

Now let's lift the Link out of its hole and open it up to see what's inside.

The Link is a tiny computer. It has a battery, a charging coil and a custom chip. Underneath hang 64 threads, thinner than a hair, that reach into the brain.

  1. The threads. 64 threads, each thinner than a human hair, dangle from the bottom. Along them sit 1,024 tiny metal pads called electrodes. Each pad is a microphone for brain electricity. Like 64 tiny strings of fairy lights, but each light is a microphone instead of a bulb.

    The threads are a bendy plastic called polyimide, with gold wires inside. In the 2019 paper they were just 4–6 micrometres thick.

    Being soft and bendy matters: the brain wobbles slightly with every heartbeat, and the threads wobble along with it.

  2. The chip. Brain signals are incredibly faint. A custom chip makes them louder, measures them thousands of times every second and spots the moments when a neuron fires. Like a sound engineer who turns up the quiet voices and writes down only the important words.

    Keeping only the important moments shrinks the data a lot, so it can be sent wirelessly without draining the battery.

  3. Power. A small battery runs it all day. To charge it, a charger outside the head sends power through the skin to a coil of wire inside the Link. Like an electric toothbrush on its stand: power jumps across a tiny gap with no metal touching.

    This is called inductive charging. A changing magnetic field from the outside coil makes electric current flow in the inside coil.

Threads
64
Electrodes
1,024

Did you know? The 2019 chip used only about 6 thousandths of a watt. Ten thousand of them would run on the power of one light bulb.

Read more

An electrode is just a small patch of metal. When a nearby neuron fires, the electric charge around it changes slightly, and the metal patch feels that change.

With 1,024 electrodes, the Link listens to many neurons at the same time. More listeners means a clearer picture of what the brain is planning.

The chip, battery and charging coil are packed in a sealed case that keeps body fluids out, because water and electronics don't mix.

The whole device runs on very little power, so it doesn't noticeably warm up the brain.

Step 4 of 8 · The device

The sewing robot

Threads that thin are far too delicate for a human hand to place. So a robot does the sewing.

A surgical robot, called R1, places each thread into the brain one by one. It uses a needle thinner than a hair and a camera to steer around blood vessels.

  1. Tiny needle. The robot's needle is about 24 micrometres wide at the tip, thinner than a hair. It hooks a thread, pushes it a few millimetres into the brain, then pulls back and leaves the thread behind. Like a sewing machine that leaves the thread in the fabric each time the needle goes up.

    The needle is made of tungsten and rhenium, two very strong metals, so it can be extremely thin without bending.

  2. Dodging vessels. The brain's surface is covered in tiny blood vessels. The robot's camera maps them, and the surgeon chooses spots in between so nothing bleeds. Like planting seeds in a garden while carefully stepping around the hose pipes.

    Rigid pins used in older implants could poke vessels and cause small bleeds. Steering around them is a big part of making implants safer.

  3. Fast and steady. The 2019 robot could place up to 6 threads, or 192 electrodes, per minute. It even adjusts for the brain moving gently as the heart beats. Like threading a needle on a slowly bobbing boat, again and again, without missing.

    In the 2019 tests, about 87% of thread insertions worked first time.

Needle tip
24 µm
Speed
6 threads/min

Did you know? Each thread goes only a few millimetres deep, about the thickness of the brain's outer layer, where the motor neurons live.

Read more

Placing 64 threads by hand would be like threading 64 needles while looking through a microscope, without ever touching a blood vessel.

The robot does the precise part. A camera builds a map of the brain's surface, the surgeon picks the target spots, and the robot inserts each thread exactly there.

Once a thread is in, the soft brain holds it gently in place.

A human hair is about 70 µm wide. The robot's needle tip is about a third of that.

Step 5 of 8 · Listening to neurons

Meet a neuron

Let's follow one thread down into the brain, until we're small enough to see the cells it's listening to.

Down here, the brain is a dense forest of cells called neurons. Each one collects messages from its neighbours and decides whether to pass a message on.

  1. Cell body. The round middle is the cell body. It adds up all the incoming messages. If the total is strong enough, the neuron fires a message of its own. Like someone listening to a crowd and only shouting when enough people say 'go!'.

    The cell body holds the nucleus, which contains the cell's DNA, just like most other cells in your body.

  2. Dendrites. Branches called dendrites spread out like a tree. They catch messages from thousands of other neurons. Like the branches of a tree catching raindrops from the sky.

    A single neuron can receive messages from up to about 10,000 others.

  3. The axon. One long wire, the axon, carries the neuron's message away to other cells. Some axons are tiny; others run all the way from your spine to your toes. Like the one outgoing cable from a house, after many incoming letters arrive.

    Axons that run to your muscles can be over a metre long, even though the cell body is smaller than a grain of sand.

Inputs
≈10,000
Cell body
≈20 µm

Did you know? Your brain has roughly as many neurons as there are stars in our galaxy, and each one is wired to thousands of others.

Read more

Neurons are the cells that do the brain's talking. Each one has three main parts: dendrites to listen, a cell body to decide, and an axon to send.

Messages flow one way: in through the dendrites, through the cell body, and out along the axon to the next neurons.

In the motor area, some neurons fire more when you plan to move left, others when you plan to move up. Their mix of activity is the movement plan.

Neurons that fire for 'move left' and ones that fire for 'move up' sit side by side, all mixed together.

Step 6 of 8 · Listening to neurons

The spike

When a neuron decides to send its message, it fires a tiny burst of electricity. Let's zoom in on one.

A neuron's message is a quick electrical pulse called a spike. It's made by tiny charged particles rushing in and out through the cell's skin.

  1. A quick pulse. A spike lasts about one thousandth of a second. It races along the axon like a wave, from the cell body to the far end. Like a stadium wave: each person stands up briefly, and the wave travels around the crowd.

    Every spike is the same size. Neurons say 'more' by firing more often, not by firing bigger spikes.

  2. Salt rushes in. Charged particles of salt, called sodium ions, rush into the axon through tiny doors. This flips the voltage inside from negative to positive. Like opening a dam gate: water floods through the moment it opens.

    Ions are atoms with an electric charge. Sodium (Na⁺) is the same sodium that's in table salt.

  3. Then resets. Right after, potassium ions flow out through other doors. That swings the voltage back down, ready for the next spike. Like a mousetrap that snaps shut, then gets re-set for the next time.

    The voltage swing is about a tenth of a volt, tiny compared to a 1.5 V battery, but huge for something so small.

Length
≈1 ms
Size
≈0.1 V

Did you know? Spikes in some neurons travel at over 100 metres per second, faster than a racing car.

Read more

Every cell is wrapped in a thin skin called a membrane. A resting neuron keeps its inside slightly negative, about −70 thousandths of a volt.

When a neuron fires, little doors in the membrane open. Sodium ions rush in and the inside briefly turns positive. Then potassium ions flow out and bring it back down.

This flip happens at one spot, which opens the doors next to it, and so on. That chain reaction is the spike travelling down the axon.

Your neurons fire billions of spikes every second, even while you sleep.

Step 7 of 8 · Listening to neurons

Catching the spark

Each spike leaves a tiny electrical ripple around the neuron. And right next to it, one of Neuralink's threads is waiting.

When a neuron fires, a faint electric ripple spreads out around it. The nearest electrode on the thread feels the ripple, and the chip writes down: 'spike, now'.

  1. Neuron fires. Each spike makes a ripple of electricity spread through the salty fluid around the neuron. It fades fast, so the electrode has to be very close. Like dropping a pebble in a pond: the ripples are strongest right next to where it fell.

    An electrode can only hear neurons very close to it, roughly within a tenth of a millimetre. Further away, the ripple is too weak.

  2. The thread. The thread is a thin, bendy ribbon with gold wires inside. Along it sit electrode pads, and each one carries what it hears up to the chip. Like a string of tiny microphones hung in a forest, each one wired back to a recorder.

    The thread is narrower than a neuron's cell body, so it slips in between cells.

  3. Spike spotted. The ripple is tiny, millionths of a volt. The chip makes it much louder, measures it 19,300 times a second and flags each spike. Like a microphone so sensitive it can hear a whisper from across a busy room, and write down every word.

    Each electrode may hear a few nearby neurons. The chip counts their spikes so the computer knows how busy each patch of brain is.

Samples
19,300/s
Signal
µV (tiny!)

Did you know? In the 2019 tests, up to 70% of the electrodes picked up spikes from nearby neurons.

Read more

The fluid around neurons is salty, so it conducts electricity. When a neuron fires, ions moving through its membrane create a tiny voltage change in the fluid nearby.

A metal electrode sitting close by picks up that change. It's far too small to use directly, so the chip amplifies it, turns it into numbers, and looks for the sharp shape of a spike.

Only the spikes are sent on. That's far less data than the raw signal, which saves battery and radio power.

Each electrode is measured 19,300 times every second. Across 1,024 electrodes, that's millions of measurements a second.

Step 8 of 8 · Thought to cursor

Thought to cursor

Thousands of spikes every second, from 1,024 electrodes. Let's zoom back out and see what the computer does with them.

The Link sends its spike counts to a laptop by Bluetooth. Software learns which patterns mean 'left', 'right', 'up' or 'click', and moves the cursor to match.

  1. Over the air. The Link beams its spike counts out through the skin by Bluetooth, the same kind of radio your wireless earbuds use. Like your phone sending music to earbuds, but in the other direction: from the head to the laptop.

    Because it only sends spike counts, the radio link doesn't need much power.

  2. Learning patterns. At first, the person imagines moving a cursor to targets. Software watches which neurons fire for each direction and learns the patterns. Like learning a friend's handwriting: after enough examples, you can read it instantly.

    This software is called a decoder. It's re-tuned over time, because the signals change a little from day to day.

  3. Real results. The first person to get the Link, Noland Arbaugh, is paralysed from the shoulders down. He used it to play chess and video games, and to browse the web, just by thinking. Like getting your hands back on the keyboard, without moving a finger.

    Some threads pulled back out of his brain in the first weeks. Better software helped get his control back.

    His cursor speed reached 8 bits per second. A person using a mouse manages roughly 10. This is still early research.

Best speed
8 bits/s
Mouse user
≈10 bits/s

Did you know? Before his implant, Noland used a stick held in his mouth to tap a tablet. With the Link, he could use his laptop for hours, even lying in bed.

Read more

The computer never reads words or memories. It only sees how busy each electrode's neurons are, many times every second.

During training, the person imagines moving toward targets on the screen. The software matches each direction to the pattern of activity it sees. After that, it can turn new patterns into cursor movement in real time.

Brain–computer interfaces are still experimental and only used in research trials. But they already show how much a person can do once messages can skip a broken spinal cord.

In his very first research session, Noland set a new world record for brain-controlled cursor speed: 4.6 bits per second.