Neuralink Thread
A neural electrode thinner than a hair, sewn into living cortex by a robot — and the company's claims need a careful read.
A Neuralink thread is a polymer-substrate microelectrode array roughly 5 µm thick and 30 µm wide, carrying ~16 platinum-iridium or sputtered-iridium-oxide recording sites along its length. Threads are inserted into the cerebral cortex by a custom robot equipped with a needle that grips, places, and releases each thread to depths up to 8 mm. A typical N1 implant carries ~1,024 channels across ~64 threads. Onboard ASICs amplify and digitize neural signals at ~20 kSa/s per channel and transmit wirelessly. The system is designed for high channel count, low foreign-body response (compared with rigid silicon Utah arrays), and rapid surgical placement.
Each thread is fabricated by photolithography on a polyimide or parylene-C substrate. Recording sites are typically iridium-oxide or PEDOT-coated to lower impedance. The robot, pictured in Musavi et al. 2019 (Neuralink white paper), inserts threads one at a time using a tungsten needle that pierces the dura, deposits the thread tip at the target depth, and withdraws — leaving the thread in tissue. The N1 implant package combines threads, a flexible PCB, ASICs for amplification and digitization (Neuralink-designed CMOS, 16-channel-per-ASIC), an inductively powered battery, and Bluetooth-class wireless telemetry. Inserted threads are flexible enough to move with brain micromotion (~10 µm) without tissue damage on the scale that rigid Utah-array shanks cause. The company demonstrated rodent and primate recordings (2019, 2021) and in early 2024 reported a first human implant ('PRIME study') in a quadriplegic volunteer, with subsequent posts claiming ~85 % cursor-control accuracy after training.
In plain terms
Old-style brain implants are like pushing a tiny fork into pudding — stiff and noisy. A Neuralink thread is more like pulling a single thread of dental floss into the pudding: bendy enough to flex with each heartbeat, with dozens of microscopic ears built into it that listen to nearby brain cells. A specialized robot, working faster than any human surgeon, threads dozens of these into the brain in under an hour.
Also called: polymer microelectrode thread, Utah-array successor.
Why it matters
- Highest published channel-count flexible BCI: ~1,024 sites in a chronic implant, vs. ~96–256 typical for Utah arrays.
- Robotic insertion may scale BCI surgery beyond what neurosurgeons can do by hand — a real engineering claim, though not unique to Neuralink (Paradromics, Precision Neuroscience pursue parallel paths).
- Pushes flexible-substrate electrode chemistry, low-impedance coatings, and on-chip neural compression toward commercial maturity.
- Has accelerated public and regulatory attention to invasive BCIs — both helpfully and, per critics, with marketing claims that outpace peer-reviewed evidence.
Where it lives
- Neuralink N1 implant (PRIME human study, FDA IDE 2023)
- Neuralink R1 surgical robot
- Comparators: Synchron Stentrode, Paradromics Connexus, Precision Layer 7
Key numbers
- Thread thickness: ~5 µm
- Thread width: ~30 µm
- Sites per thread: ~16
- Channels per implant: ~1,024 (Across ~64 threads)
- Sample rate: ~20 kSa/s/channel
- First human implant: Jan 2024 (PRIME study, single subject)
Surprising facts
- Each thread is roughly the diameter of a red blood cell (~5 µm) and yet carries ~16 individually addressable recording sites — placement precision approaching one neuron's reach.
- The R1 robot performs ~64 thread insertions, each penetrating the cortex to a programmed depth, in roughly an hour — faster than any human surgeon can implant a Utah array.
- Compared with rigid silicon Utah arrays, the polymer thread's bending stiffness is ~10⁶× lower, meaning it can flex with cardiac and respiratory brain motion rather than slicing through tissue at each pulse.
Open questions
- Thread retraction in early human subjects: Neuralink reported in May 2024 that some threads in its first human subject had retracted from the cortex, reducing usable channel count. The company adjusted decoding algorithms to compensate. The retraction mechanism — micromotion, gliosis, or surgical placement variance — is not yet publicly characterized in detail. This is the kind of single-subject device behavior that needs replication and peer review before broad claims can be made.
- Performance numbers come from company posts, not peer review: Headline figures such as '~85 % cursor-control accuracy' and 'fastest-ever cursor speed' originate in Neuralink blog posts and X/Twitter demos. As of mid-2026, no full peer-reviewed clinical performance paper has been published. Treating those numbers as established BCI benchmarks would be premature.
A short history
Founded 2016 by Musk and a small team of neuroengineers including Vanessa Tolosa, DJ Seo, Ben Rapoport (later Precision Neuroscience), and Max Hodak (later Science Corp). The 2019 white paper described the polymer-thread approach and the R1 robot. Pig (2020) and primate (2021) demos followed. The FDA granted Investigational Device Exemption in May 2023; the first human implant followed in January 2024. As of 2026, the public record includes a small number of human subjects and an ongoing trial; many performance claims remain company posts rather than peer-reviewed publications.
Sources
- Musk & Neuralink — bioRxiv white paper (2019) (https://www.biorxiv.org/content/10.1101/703801v1)
- Neuralink PRIME study clinical trial registration (NCT06429735) (https://clinicaltrials.gov/study/NCT06429735)
- Reuters — FDA approves Neuralink human trials (2023) (https://www.reuters.com/science/neuralink-says-fda-approval-its-first-human-trial-2023-05-25/)
- Independent BCI review — Cell Reports Medicine (2024) (https://www.cell.com/cell-reports-medicine/)