Industries / Frontier science / Brain-computer interfaces
Surgery
Surgery is the work of placing an implant in or on the brain, connecting it, and taking it out again.
6
Milestones achieved, from 1992 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1992
Fired in at speed
Patrick Rousche and Richard Normann at the University of Utah found that the 100 needles of the Utah array could not be pressed into the brain slowly. They built an air-driven inserter that fires the array in, and found in cat cortex that it took at least 8.3 m/s to seat all 100 needles fully and safely at 1.5 mm depth.
Lessona dense bed of needles goes in cleanly only at high speed.
Credit
P. J. Rousche and R. A. Normann
2019
In through a vein
At the Royal Melbourne Hospital, doctors fed Synchron's Stentrode, a stent (a mesh tube of the kind used to hold blood vessels open) carrying electrodes, through a catheter from the jugular vein in the neck to the superior sagittal sinus, a large vein over the brain's movement area, without opening the skull. In the SWITCH study, which began in May 2019, four people with paralysis completed 12 months with no serious device-related adverse events, no blocked vessels and no device movement, and each used it to control a computer.
Credit
Peter Mitchell, Thomas Oxley, Nicholas Opie, Bruce Campbell and colleagues, Royal Melbourne Hospital and Synchron
2020
A safety count for arrays
A University of Michigan team searched the medical literature for complications with Utah arrays in people. Of 433 publications, 76 qualified; they described 48 people with the arrays, 18 of them as long-term implants, and few specific complications, and the longest implant had lasted 1,975 days, about 5.4 years. The authors note that the pedestal, the connector that passes through the scalp, breaks the skin's barrier and may raise the risk of infection. In deep brain stimulation, the established brain implant they also reviewed, infection was the most reported hardware problem, at 4.57%.
Credit
Autumn Bullard, Cynthia Chestek, Parag Patil and colleagues
2021
What removed arrays show
Researchers examined six Utah arrays taken out of two people after 980 days and 182 days. The arrays left in longer were more thickly wrapped in tissue, showed more wear of their platinum tips and gave weaker signals. On one array coated with iridium oxide, brief stimulation stripped the iridium from most of its sites.
Lessontissue build-up and wear of the metal both grow with time, so both limit how long an array works.
Credit
Kevin Woeppel, Jennifer Collinger, Robert Gaunt, Xinyan Tracy Cui and colleagues
2023
Seventeen years of safety data
The BrainGate trial reported on 14 adults who received one or two Utah arrays at seven US sites between 2004 and 2021, for an average of 872 days and 12,203 days in all. There were 68 device-related adverse events, 6 of them serious; the most common was skin irritation around the pedestal. None required removing the device, and there were no infections inside the skull and no deaths or lasting disability caused by the device.
Lessonthe place where the connector crosses the skin was the most frequent source of trouble.
Credit
Daniel Rubin, Leigh Hochberg and the BrainGate team
2024
A robot places the threads
In Neuralink's first human operation, at Barrow Neurological Institute in January 2024, a neurosurgeon opened the scalp and removed a piece of skull and of the dura. Neuralink's R1 robot then inserted the implant's threads into the cortex, and the surgeon set the implant body in the skull opening and closed the scalp. Neuralink says the threads are too thin and flexible to place by hand. Barrow confirms that the first participant received the implant there.
Credit
Neuralink; Barrow Neurological Institute
Watch for
Threads through the dura
Neuralink says it is investigating inserting threads directly through the dura, without removing it, to make the operation less invasive. Elon Musk wrote on X on 31 December 2025 that Neuralink would move to "a streamlined, almost entirely automated surgical procedure" in 2026.
Achieved whena published report or regulator record shows implants placed through intact dura in people, with their complication rates.
Claimed by
Neuralink and Elon Musk, December 2025 and January 2026
No serious device harms
Neuralink says 21 participants are enrolled in its trials worldwide and that it has a "current record of zero serious device-related adverse events".
Achieved whenthe trial's safety data are published in a peer-reviewed journal or by a regulator.
Claimed by
Neuralink, January 2026
Open
01
Scored thread insertion in gel
An open-design robot that inserts sham flexible threads (thin strips of plastic film with no electronics) into a clear gel stand-in for brain, steering around a pattern of dyed "blood vessels" on the gel's surface with a camera, and is scored on a fixed, repeatable test.
The briefCredit · unclaimed02
A see-through vein model
A clear silicone copy of the superior sagittal sinus, the large vein over the brain's movement area, with a pump pushing a blood-like fluid through it. Stent-mounted electrode prototypes are deployed in it by catheter and photographed to measure how closely they press against the wall and how much they slow the flow.
The briefCredit · unclaimed
Open 01 · the brief
Scored thread insertion in gel
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-design robot that inserts sham flexible threads (thin strips of plastic film with no electronics) into a clear gel stand-in for brain, steering around a pattern of dyed "blood vessels" on the gel's surface with a camera, and is scored on a fixed, repeatable test.
Why it is still open
Neuralink's robot is proprietary. Its 2019 paper reported a robot that inserts up to six threads (192 electrodes) a minute, with an average insertion success of 87.1% (standard deviation 12.6%) over 19 surgeries in rats, but released no design files. Academic groups test how flexible probes enter gel, for example a 2026 study of coated probe arrays in agarose gel and rat brain, but this research found no open design for a thread-inserting robot with a published success tally.
Built with what exists
A motorised three-axis stage or micromanipulator; a fine tungsten needle; sham threads laser-cut from polyimide film by a commercial service; agarose gel blocks, which the 2026 study chose because their mechanical properties are close to brain's; a USB microscope camera; an open image-segmentation model such as Meta's Segment Anything Model 2 to map the dyed vessels; Python; design files on GitHub.
First step
Cast five agarose blocks with lines of dyed gel on top and make 20 hand-guided insertions with a needle on a manual micromanipulator to set a baseline.
Done looks like
200 filmed insertions into fresh gel blocks, with a published tally of threads that reach their target depth within 100 µm, the average placement error measured from microscope images, and the number of dyed vessels struck, set against the 87.1% figure Neuralink reported in rats.
Open 02 · the brief
A see-through vein model
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A clear silicone copy of the superior sagittal sinus, the large vein over the brain's movement area, with a pump pushing a blood-like fluid through it. Stent-mounted electrode prototypes are deployed in it by catheter and photographed to measure how closely they press against the wall and how much they slow the flow.
Why it is still open
Published work on how a stent electrode sits in this vein relies on computer models and sheep. A 2024 University of Melbourne study modelled blood flow around a stent-mounted interface in idealised human and sheep vein shapes, compared it with tissue growth in sheep, and concluded that how much the stent is oversized for the vein must be considered. This research found no physical bench model with published deployment results.
Built with what exists
Vein dimensions from that study and the sources it cites; a resin 3D printer for the mould; clear silicone; a peristaltic pump; a glycerol and water mix that flows like blood; bare self-expanding nitinol stents of matching size with flexible printed electrode strips bonded on; a camera and a pressure sensor; Python for measuring the images.
First step
Print a mould of the idealised human vein from the 2024 study's dimensions and cast one silicone channel.
Done looks like
20 deployments at 3 oversizing ratios (stent diameter divided by vein diameter), with the measured gap between stent and wall and the pressure drop for each, published with the mould files and compared with the 2024 study's flow model.