The Race to Build a BCI That Doesn’t Require Brain Surgery

In September 2019, Facebook announced it had bought a small New York startup called CTRL-labs, founded by Thomas Reardon, a former Microsoft executive who had gone back to school for a PhD in neuroscience. CTRL-labs made a wristband. It looked unremarkable, a black band studded with electrodes, but it did something that most brain-computer interface labs had spent decades trying to do with needles and craniotomies: it let a person control a cursor, a keyboard, a game, using nothing but the electrical signal of their own intention to move. Reardon called it reading the “final common pathway,” the bundle of motor neurons that fire just before a muscle contracts, picked up through the skin at the wrist rather than through the skull.

No surgeon was involved. No hole was drilled. And that was the point. CTRL-labs had not solved the brain-computer interface problem so much as sidestepped it, by moving the point of contact off the brain entirely and onto a part of the nervous system that surgery never has to touch. It is one of at least four distinct strategies now competing to answer the same question: can a machine read, or write to, the nervous system with meaningful bandwidth, without opening the body to do it.

Move the interface to the wrist, and the skull stops being a problem.

Why the Skull Blurs Every Signal Underneath It

The dominant noninvasive method, electroencephalography, has existed since Hans Berger recorded the first human EEG in Germany in 1924. It has never gone away, and it has never gotten dramatically better, for a reason rooted in physics rather than engineering. The skull is bone, and bone is a poor conductor of electricity. When a patch of cortex fires, the electrical field it generates has to pass through cerebrospinal fluid, skull, scalp, and hair before a scalp electrode ever sees it. Each layer smears the signal sideways, a phenomenon called volume conduction. A scalp electrode picking up activity is really averaging the electrical chatter of tens of millions of neurons spread across several square centimeters of cortex. It cannot tell you which one did what.

That blurring shows up as bandwidth. In 2021, Frank Willett and colleagues in Krishna Shenoy’s lab at Stanford, working within the BrainGate consortium, published a result in Nature in which a paralyzed participant with an intracortical microelectrode array imagined handwriting and had it decoded at roughly 90 characters per minute, competitive with typical smartphone typing speed. Nothing built from scalp electrodes comes close. Classic noninvasive spelling systems built on P300 signals or steady-state visual evoked potentials typically manage single-digit to low double-digit bits per minute, an order of magnitude or more below what an implanted array can pull out of the same brain.

The skull was built to protect the brain, not to talk to it.

That gap is why companies chasing a genuinely noninvasive, high-bandwidth BCI have mostly stopped trying to out-engineer EEG and started looking for entirely different physical channels: light instead of electricity, magnetism instead of electricity, sound instead of either.

The Slow, Warm Signal of Blood

Kernel, founded by entrepreneur Bryan Johnson in 2016, bet on optics. Its headset, Kernel Flow, uses a technique called time-domain functional near-infrared spectroscopy, which shines pulses of infrared light into the scalp and times how long photons take to scatter back out. Active neurons demand more oxygenated blood, and that changes how the tissue absorbs and scatters light, so the device is not measuring electrical activity at all. It is measuring the metabolic aftermath of it, the same neurovascular coupling that functional MRI relies on, just from the surface instead of inside a magnet.

The tradeoff is timing. Blood flow lags neural firing by roughly a second or two, which means fNIRS can localize brain activity reasonably well in space but is nearly useless for anything requiring fast, moment-to-moment decoding, such as controlling a cursor in real time. It is far better suited to slower questions: is this region more active during this task than that one, is this person’s prefrontal cortex working harder under this cognitive load. That has made fNIRS attractive for wellness and cognitive-monitoring products rather than communication devices, and it is why Kernel has marketed Flow toward research and mental-health applications rather than as a control interface.

Reading Magnetic Fields Instead of Electrical Ones

A different fix for the skull’s blurring problem is to stop measuring electricity altogether and measure the tiny magnetic fields that electrical currents in the brain produce instead. Magnetoencephalography has done this since the 1960s, but conventional MEG requires superconducting sensors cooled with liquid helium, housed in a rigid helmet the size of a hair dryer chair, built for one head shape, immobile by design. It has never left the hospital basement.

Physicists Matthew Brookes and Richard Bowtell at the University of Nottingham, working with sensors called optically pumped magnetometers, published a paper in Nature in 2018 demonstrating a wearable MEG system that operates at room temperature using rubidium vapor cells rather than cryogenics. The sensors sit in a 3D-printed helmet molded to an individual head, light enough that a participant can move, nod, even walk on a treadmill while being scanned. Magnetic fields, unlike electrical ones, pass through the skull with far less distortion, so OPM-MEG keeps much of the spatial precision that makes MEG valuable clinically while shedding the cryogenic hardware that made it immobile. It is not yet a consumer product. It requires magnetically shielded rooms to work well, since the fields it measures are billions of times weaker than the Earth’s own magnetic field. But it is the clearest evidence that a wearable, skull-agnostic brain sensor is physically possible, not just a marketing promise.

Betting on Sound Where Light and Electricity Fail

The newest entrant into the field uses neither light nor magnetism but sound. Focused ultrasound can pass through the skull with far less scattering than light does, and functional ultrasound imaging, developed largely in preclinical neuroscience labs studying rodents and non-human primates, has shown it can track blood-flow changes tied to neural activity at a spatial resolution well beyond fNIRS. Mikhail Shapiro’s lab at Caltech has spent years developing tools to make neurons themselves responsive to ultrasound and to genetically encode reporters that ultrasound can detect, part of a broader push to make sound a two-way channel into the brain rather than just an imaging tool.

On the wilder end of the field, Mary Lou Jepsen’s company Openwater has pursued a combination of holographic optics and ultrasound with the explicit goal of building a wearable device with resolution rivaling MRI, without the magnet, the noise, or the seven-figure machine. None of these approaches yet reads individual neurons firing in real time through an intact human skull. What ultrasound offers, in theory, is a path to spatial precision that neither scalp EEG nor fNIRS can currently match, at temporal speeds still too slow for direct thought-to-text communication but potentially fast enough for coarser signals: attention, arousal, the general location of a motor intention.

Every noninvasive route trades bandwidth for safety. None of them yet trades nothing.

What Minimally Invasive Buys That Noninvasive Still Can’t

It is worth being precise about where the line actually sits, because the industry itself keeps moving it. Synchron, led by Tom Oxley, built a device called the Stentrode that is delivered through a catheter into the jugular vein and threaded up to sit against the wall of a blood vessel near the motor cortex, reading electrical activity from inside the vasculature rather than from the scalp. It requires no craniotomy, and Synchron has described it as minimally invasive rather than noninvasive. The first human implant happened in Melbourne, Australia, in 2019, and Synchron implanted its first United States patient at Mount Sinai in July 2022 under an FDA early feasibility study. Precision Neuroscience, cofounded by former Neuralink engineer Benjamin Rapoport, takes a similar middle path: a film of electrodes thinner than a human hair laid on the surface of the brain through a small slit, rather than an array of needles driven into cortical tissue.

Neither device requires surgery in the traditional, skull-opening sense that Neuralink’s array does. Both still require a surgical or interventional procedure, sedation, and a hospital. That places them in a category the field increasingly needs its own vocabulary for, somewhere between “brain surgery” and “wearable,” and it is a reminder that the fully noninvasive technologies described above, EEG, fNIRS, OPM-MEG, ultrasound, are competing against a moving target that keeps getting less invasive rather than staying still.

The Question the Field Keeps Avoiding

Every noninvasive project eventually runs into the same fork. Either it accepts a low-bandwidth signal and finds a product where that is enough, attention tracking for a headset, coarse motor intent for a wristband, mood signals for a wellness app, or it keeps chasing the bandwidth of an implanted array and keeps failing to reach it, because the skull and the scalp are not incidental obstacles. They are doing exactly what they evolved to do, which is keep the outside world from touching the brain directly. Reardon’s wristband succeeded commercially by abandoning the goal of reading the brain at all. Kernel and the OPM-MEG groups are succeeding scientifically by being honest about what their signal actually is: metabolism, or magnetism, not thought.

The honest version of this story is not that surgery is about to become optional for high-bandwidth brain-computer interfaces. Willett’s 90-character-per-minute handwriting decoder still required electrodes inside the skull, and nothing on the noninvasive side is within an order of magnitude of that. What is changing is the shape of the market underneath it. Locked-in patients who need to communicate at the speed of thought will likely need surgery, or at minimum a Synchron-style intervention, for years to come. Everyone else, the people who just want a headset that knows when they are distracted or a wristband that replaces a keyboard, may never need a surgeon at all. The race to build a BCI without brain surgery is not really one race. It is several, running toward different finish lines, and only one of them was ever trying to replace the implant.