How Neuralink works: from cortical spikes to wireless digital control

The core problem: intact cortical intent trapped behind injured pathways

In conditions such as cervical spinal cord injury, brainstem stroke, or amyotrophic lateral sclerosis (ALS), the neural cables carrying commands from the brain to the muscles are damaged or degenerating. Yet the computational engine upstream—the primary motor cortex and speech-motor cortex—remains active. When a person imagines moving a thumb, dragging a mouse, or speaking a sentence, populations of cortical neurons fire millisecond electrical impulses called action potentials (or spikes).

An intracortical brain–computer interface creates an engineered digital bypass. Rather than repairing the injured spinal cord biologically, it reads electrical activity directly from cortical tissue and translates that mathematical pattern into external commands. If you are new to the hub, pair this system overview with Inside the N1 implant and R1 surgical robot and PRIME, CONVOY and VOICE human trials.

Step 1: Recording inside the cortex with 64 flexible polymer threads

Every signal-processing chain is bounded by its sensor. Non-invasive electroencephalography (EEG) caps sit outside the skull, where bone and scalp blur electrical activity across centimeters of tissue. To resolve individual neurons, electrodes must sit within tens of micrometers of firing cell bodies inside the cerebral cortex.

Neuralink's N1 implant distributes 1,024 recording electrodes across 64 ultra-thin flexible polymer threads (16 electrodes per thread). Because the brain pulses with every heartbeat and shifts during head motion, flexible polyimide-style threads are engineered to move with cortical tissue rather than tethering cleanly to a rigid skull pedestal. Because these threads are too compliant to push into brain tissue by hand, the custom R1 surgical robot uses a femtosecond-laser-milled needle and optical imaging to thread each ribbon around surface blood vessels.

Official @neuralink Video 3 Dec 2025 · 21:39 Tier 1 & 2 · Technical & Clinical Briefing

Neuralink Overview, Fall 2025

Click to load the privacy-enhanced youtube-nocookie.com player. No third-party cookies or requests are sent until you press play.

▶ Load embedded video (21:39) Watch on YouTube ↗ Read video analysis →

Step 2: On-chip digitization and sealed wireless telemetry

Raw extracellular neural signals are tiny—typically tens to hundreds of microvolts—and vulnerable to electrical noise if carried over long unshielded wires. Inside the N1's hermetically sealed, coin-sized enclosure (which sits in a small recess in the cranium beneath the closed scalp), custom low-power application-specific integrated circuits (ASICs) amplify, filter, and digitize all 1,024 channels right at the recording site.

• Spike detection and LFP extraction: On-chip processors detect threshold crossings and spike waveforms while also capturing lower-frequency Local Field Potentials (LFPs) that reflect synchronized local neural populations.
• Fully implanted wireless link: Processed neural packets stream through the skin via Bluetooth-band telemetry to the Neuralink App on a paired computer, tablet, or phone—no percutaneous cable or transcutaneous plug is used.
• Inductive charging: An internal battery powers the implant during multi-hour daily sessions and recharges wirelessly through an external inductive coil worn briefly over the scalp.

Step 3: Adaptive neural decoding and closed-loop feedback

Once neural feature packets arrive at the receiving device, machine-learning decoders translate high-dimensional firing rates into continuous control vectors or discrete tokens every few milliseconds:

What this architecture enables today vs. what remains investigational

Understanding the physical signal chain clarifies both the breakthroughs and the boundaries of the technology. Reading motor cortex or speech-motor cortex allows a decoder to drive external software (Telepathy cursor control, speech restoration in VOICE, and assistive robotic arms in CONVOY). Writing signals back into the brain—such as stimulating the visual cortex in Blindsight vision research—requires a distinct stimulation safety profile and separate clinical trials.