Neuralink’s implant reads the brain through threads of wire pushed into the cortex, and every thread displaces tissue. A patent the company was granted on 17 November describes a machine whose sensors are living neurons, grown in a dish and laid on the surface of the brain, where they grow their own connections down into it.
US Patent 12,797,514, “Cell-based Brain-machine Interface”, is assigned to Neuralink Corp. of Fremont, California. It names 12 inventors. Two of them, Max Hodak and Alan Mardinly, founded Science Corporation together in 2021. Hodak co-founded Neuralink and was its president. Science has since raised more than $900 million, including a $400 million round announced on the same day as this grant, and it is building a brain interface of its own from living neurons.
What the patent describes
The cells are cortical neurons prepared outside the body. They are placed on the cortex above its outermost existing layer, forming a new one the patent calls layer zero. Over the following weeks they send axons and dendrites down into the host brain and form synapses with it, which is what lets signals pass in both directions.
Reading and writing happen at the graft, on the surface. Because the cells can be made to glow when active, a camera placed above them can watch which ones are firing. Because they can be made responsive to light, a light source above them can switch them on, and they pass that on to the brain through the connections they have grown. A second route in the patent tethers the cells to electrodes.
The claims
The patent has 18 claims. Claim 1 covers the whole loop: neurons from outside the body grafted on the cerebral cortex, a camera that images the graft to read from it, a light source that illuminates the graft to write to it, and processing circuitry joining the two. Four more independent claims cover parts of that loop: a graft read with a camera, a graft written to with light, and two in which the cells are tethered to electrodes. Every one of them requires the cells, sitting at the outermost cortical layer, to penetrate the brain by growing processes into it.
The dependent claims set out the range. The graft can be about 50 microns to about 1 mm thick. It can go over the frontal, parietal, temporal or occipital lobe. The cells can be stem cells grown in culture, and can be given their new properties by electroporation or by a virus. The mammal can be a human, including one whose movement, sight, hearing, taste or smell is impaired.
Those claims are narrower than the ones Neuralink first sought. In the application as published in March 2023, claim 1 covered any graft of neurons on the cortex that responds to an outside stimulus with a detectable signal, with no camera, no light source and no requirement that the cells grow into the brain.
What the animal work showed
The experiments in the patent are in mice. Cortical neurons were taken from mouse embryos, given a gene that makes them fluoresce when active, and deposited on the brains of adult mice under a glass window.
The graft doubled in thickness over the first two weeks and then held steady. Its activity started as slow synchronised waves across the whole population and, by about four weeks, looked like ordinary cortical activity.
Eleven mice were trained on a task in which the activity of selected grafted neurons, watched through a microscope, moved a tone up and down, with water as the reward. As a group they beat chance by the third day of training. Writing information in worked less often: with grafted cells made light-sensitive, three of seven mice reached the performance criterion, after 21 days of training on average. The patent attributes the failures to how few of the transplanted cells took up the light-sensitive gene, and says the result still shows the layer carrying information into the brain.
The same idea at Science Corporation
Science calls its version biohybrid. In a preprint posted in November 2024, its researchers loaded about 90,000 mouse neurons into a grid of wells 10 microns across, one cell per well, bonded to a glass coverslip and placed cells-down on the brain. About half the wells still held labelled neurons three weeks later, and the cells sent processes into the cortex below. Five of nine mice learned to report when those neurons were switched on with light, as fast as mice whose own brain cells had been made light-sensitive. The preprint has not been peer reviewed.
Science’s published design puts each neuron in the device between a microLED that stimulates it and an electrode that records it, with the cell’s own fibres reaching into the brain. Hodak told Bloomberg this week that the company hopes to test such a device in a person within 16 months and has picked a first disease without naming it. TechCrunch reported that Science is designing those trials with Yale neurosurgery professor Murat Günel.
Signal: Why it matters
Electrode counts in brain implants are limited by what the brain will tolerate being pushed into it. Both of these approaches try to move the sharp part of the problem out of the brain by letting cells, which the brain already accommodates, do the reaching. A granted patent is one marker of how far that idea has travelled from a lab result toward a product: Neuralink filed the first application in September 2021 and now holds claims that name humans, name the lobes and name the hardware.
The patent also shows how closely the histories of the two companies run together. The work behind it was done at Neuralink, which owns it. Two of the inventors named on it went on to found Science, which is now building a living-neuron interface of its own and talking publicly about a first test in a person.
Not disclosed
Whether Neuralink is still working on cell-based interfaces, or has said anything about this grant. Whether Neuralink has licensed the patent or the wider family it belongs to. Whether the mouse results in the patent have been published or peer reviewed. The date the patent office set for the patent to expire, which depends on how long the examination took.
Forward Radar: What to watch
Whether Neuralink says anything publicly about the patent or about living-neuron interfaces.
Related applications from the same September 2021 filing, including the international one, and any continuations Neuralink has pending.
Whether Science names the first disease for its biohybrid device, and whether the human test Hodak described stays on his 16-month timeline.
Sources
Primary:
- USPTO Patent Public Search: US 12,797,514 B2, Cell-based Brain-machine Interface, published 17 November 2026
- Google Patents: US20230077899A1, the published application, filed 8 December 2021
- Google Patents: WO2023038829A1, the international application
- bioRxiv: Brown et al., Optogenetic stimulation of a cortical biohybrid implant guides goal directed behavior, November 2024
- Science Corporation: Biohybrid
- Science Corporation: Science Series D, 17 November 2026
Coverage:
- IEEE Spectrum: Biohybrid BCI adds more neurons to the brain, 19 February 2025
- TechCrunch: Science Corporation’s brain tech is worth $3.5 billion after new funding, 17 November 2026
- Investing.com, summarising Bloomberg: Science Corp valuation hits $3.5B on retina chip funding, 17 November 2026
Cross-reference to prior InsideBCI coverage: