A new ultra-thin brain-computer interface paves the way for discrete, high-throughput communication with the cortex – with clear therapeutic implications for epilepsy, spinal cord injury, ALS, stroke, and blindness. A research team from Columbia University and NewYork-Presbyterian Hospital, along with partners from Stanford and the University of Pennsylvania, has developed a system that creates a fast wireless connection between the cerebral cortex and external computers, without bulky electronics capsules and without cables passing through the skull. The key lies in extreme miniaturization and monolithic integration: the entire system fits onto a single silicon wafer that can be thinned and bent to lie on the brain surface like paper, thereby reducing the invasiveness of the procedure to a minimum, and raising data throughput to levels required for the operation of advanced machine learning algorithms.

Why BISC is different: from a “can of electronics” to a single wafer

Most medical BCI systems of the latest generation use multiple separate microelectronic modules – amplifiers, A/D converters, control logic, and radio transceivers – housed in a relatively large implanted “can” with cables passing through the skull or to the chest. Such architecture increases invasiveness, creates potential infection pathways, and limits bandwidth. BISC breaks that model. The entire interface – from 65,536 microelectrodes in the array (µECoG), through analog front-ends and A/D converters, to radio transceivers, power supply, and digital control – is located on a single monolithic CMOS wafer thinned to the thickness of a hair. The board is placed subdurally on the cortex, without penetrating the tissue, which reduces reactivity and simplifies surgery.

The implant relies energetically and data-wise on an external, wearable “relay station” that wirelessly delivers power and receives data. The connection is established via ultra-wideband (UWB) radio with an on-chip transceiver; in practice, this enables throughput on the order of hundreds of megabits per second, enough for simultaneous recording of thousands of channels with high temporal resolution and for bidirectional “read–write” interaction (stimulation and recording) on the cortex.

Scalability at the semiconductor industry level

The strength of BISC lies in the fact that it is not produced as a hand-assembled “patchwork,” but by standard semiconductor industry processes. The chip is made in a 0.13-µm BCD process (a combination of bipolar, CMOS, and DMOS processes) that allows the coexistence of precise low-noise analog circuits, digital logic, power management, and powerful outputs for stimulation on the same crystal. This achieves a hitherto unrecorded density of functionality: a microelectrode array of 256×256 contacts, 1,024 simultaneous recording channels, and up to 16,384 programmable stimulation channels, together with radio and power subsystems – all in a volume on the order of cubic millimeters and with a total assembly thickness that can be reduced to a few tens of micrometers. Since everything is made using standard lithography, the platform is suitable for serial production and gradual scaling of channel numbers without changing the basic architecture.

Wireless high throughput: why 100 Mbps matters

In clinical and research BCI applications, it is crucial to “capture” as much information as possible about intentions, perception, and brain states, and this requires both high spatial and high temporal resolution. BISC’s UWB connection declaratively achieves about 100 Mb/s of aggregate throughput, without physical wires between the cortex and the computer. That figure is not just an engineering datum on paper: such bandwidth enables the execution of advanced machine and deep learning algorithms on almost raw signals, without aggressive compression and without losing the finesses needed to decode nuances of movement, speech, or perception. The relay station communicates with the computer via standard wireless protocols, which simplifies integration into existing workflows in the hospital and laboratory.

From laboratory to hospital: study status on December 10, 2025

On the path to clinical application, the team has developed minimally invasive surgery in which the thinned chip is introduced through a small craniotomy and “slipped” into the subdural space, directly onto the cortex. In preclinical models, the stability of long-term recording from the motor and visual cortex has been confirmed. Traveling waves in the gamma band in the visual cortex, which carry an abundance of information about stimuli, have also been recorded, which typical low-resolution systems cannot register. In parallel, short intraoperative studies on humans are being conducted, focused on signal quality, communication robustness, and implant position safety during neurosurgical procedures.

For epilepsy, there are concrete plans: neurosurgical and neurological teams at the NewYork-Presbyterian/Columbia center have registered trials for the use of BISC in patients with drug-resistant epilepsy. The idea is to use high spatiotemporal resolution for precise mapping of foci, seizure prediction, and, in the long term, targeted stimulation that dampens pathological patterns before they grow into a clinical seizure.

Clinical potential: from seizure control to restoration of vision and speech

Although the initial focus is on epilepsy, the technology is applicable to a range of other conditions. In paralysis after spinal cord injury or stroke, high electrode density and throughput open space for decoding movement intent and controlling exoskeletons or functional electrical stimulation. In amyotrophic lateral sclerosis, a BCI system with such resolution can serve as a foundation for fast communication interfaces, and in ophthalmology and neuro-ophthalmology, bidirectional stimulation on the cortex can become the foundation for visual prostheses of higher spatial resolution. Finally, speech decoding technologies with high data stream throughput have already shown that they can reconstruct more natural, expressive speech in real time, which opens an important path toward restoring communication to persons who have lost it.

How BISC works: a short “stack” from electrode to cloud

• µECoG array: 256×256 arrangement of high-density microelectrodes lies on the surface of the cortex and capacitively/ohmically “reads” local field potentials, with minimal penetration into the tissue.


• Analog front-end: each recording channel contains a low-noise amplifier, programmable filters, and A/D converters optimized for bioelectric signals.


• Control and processing: on the chip is a controller with its own instruction set that manages multiplexing, stimulation patterns, and data packetization.


• Power and RF: wireless inductive/capacitive powering and UWB transceiver for bidirectional communication with the wearable relay unit; the relay uses standard 802.11 access toward the computer.


• Software layer: API and machine learning tools that enable decoding of motor skills, speech, and perception, and implementation of closed-loop stimulation.

Engineering decisions that make a difference

Thickness and flexibility. A wafer thinned to ~50 µm (in some versions even down to ~15–25 µm total thickness with encapsulation) can slide between the dura and the calvaria without creating volumetric pressure on the parenchyma and conformally adapt to the gyri of the cortex. Monolithic integration removes cables and connectors, reduces failure points, and brings significant volume savings. Bandwidth on the order of 100 Mb/s is a qualitative leap – it enables “streaming” of neural data in real time and a foundation for closed-loop stimulation. Energy efficiency and power management are engineered so that dissipation remains within safe limits for neural tissue, with constant monitoring of temperature and signal integrity.

Comparison with other BCI approaches

The BCI ecosystem is diverse: from deep penetrating microneedle arrays to epidural and subdural grids and completely non-invasive systems. Penetrating arrays (e.g., robotically implanted threads) give very high local resolution but carry higher surgical risk and long-term tissue reactivity. Epidural and subdural systems are less invasive but have traditionally offered lower resolution and throughput. BISC combines the advantages of both worlds: subdural, soft contact with the cortex and channel density that enters the territory of high-resolution reconstructions – while avoiding permanent trans-cortical leads and bulky electronics capsules. For clinicians, this means easier placement, for engineers greater scalability, and for patients potentially faster recovery and lower risk of infection.

From DARPA to startup: the commercialization path

The development of BISC was encouraged within frameworks of programs focused on high-resolution interfaces with a large number of recording and stimulation channels, with the aim of making the bridge between the brain and digital systems as wide as possible and safe at the same time. Researchers subsequently initiated industrial transfer through the newly founded company Kampto Neurotech, which produces research-ready versions of the chip and works on further development of the platform to meet requirements for human trials. The industrial approach to wafer-scale production and metrics like electrode density per cubic millimeter stands out as a differentiator compared to existing systems.

Safety, biocompatibility, and regulation

One of the reasons for choosing the subdural configuration is the reduction of tissue reactivity and micromotion that degrade the signal over time. In doing so, flexible dielectrics (e.g., parylene, polyimide) and hermetic encapsulation concepts that do not increase the assembly stiffness are used. Energy densities and thermal flows are engineered so as not to exceed safety limits for neural tissue, while the entire system is conceived to work in a closed loop with signal integrity monitoring and error detection. On the regulatory side, the standard path lies ahead: assessment of material biocompatibility, verification of electromagnetic compatibility, software validation, and clinical studies that must demonstrate safety and utility in clearly defined indications.

AI as the “second half” of the system

A high-throughput BCI without powerful algorithms remains just a “microphone” for the brain. BISC is designed for collaboration with machine learning models that work on local field currents and potentially on connections with visual, somatosensory, and motor representations. In motor decoding, the dense arrangement of contacts facilitates learning the mapping of movement intent into commands for prostheses or exoskeletons. In speech, the combination of dense spatial sampling and a stable wireless connection enables systems to reconstruct continuous, expressive speech in near real time. In perception, large bandwidth and bidirectionality create prerequisites for closed loops in which algorithms not only read but also targetedly stimulate to correct pathological patterns.

What this means for the clinic in the coming years

In epilepsy: finer mapping of networks that generate seizures, predictors based on deep models, and the potential of closed-loop stimulation for preventing seizures. In motor deficits: faster “thought-to-movement” channels and more precise control of multiple degrees of freedom. In communication: more natural, faster speech generated from neural activity, including nuances of expression. In visual disorders: foundations for cortical prostheses with a larger number of stimulation spaces. All this depends on evidence of safety and robustness over longer periods, on the sustainability of wireless powering, and on the system’s ability to maintain a favorable signal-to-noise ratio long-term without scarring changes at the tissue-electrode interface.

Technical specifications and figures (context for experts)

• Electrode density: 256×256 (65,536 contacts) in the µECoG array; simultaneous recording ≥1,024 channels; stimulation up to 16,384 channels with programmable patterns.


• Radio connection: UWB with an aggregate bit rate of approximately 100 Mb/s; the relay station behaves toward the computer as an 802.11 device.


• Fabrication technology: 0.13-µm BCD (monolithic integration of analog, digital, power, and RF).


• Mechanics: thinned crystal of ~50 µm thickness (total with encapsulation ~25–50 µm), flexible for subdural placement; chip area on the order of square millimeters; volume ~3 mm³.


• Software: proprietary “instruction set”, API, and tools for decoding intent, perception, and state; closed-loop stimulation supported.


• Preclinical results: chronic recordings (weeks–months) in pigs and non-human primates with motor and perception decoding; intraoperative recordings in humans in progress.

Open questions and boundaries

How long do electrode impedance and signal-to-noise ratio remain stable? How to ensure long-term hermeticity without increasing stiffness? At what rate does consumption grow in relation to the number of active channels and how does this affect the thermal balance? Can the wireless connection reliably cope with the electromagnetically “polluted” hospital environment? How to incorporate cybersecurity and data privacy from day one? And finally, how to validate decoders so that they are robust to long-term changes in neurophysiology and patient behavior? Answers to these questions will decide whether BISC can cross the boundary from a promising platform to a standard neurosurgical tool.

In sum, with the appearance of a single-component, ultra-thin, wireless BCI that brings the entire “signal chain” onto one chip, the brain gains a high-throughput communication “portal”. If upcoming clinical studies confirm safety and effect in diseases like epilepsy, paralysis, or vision loss, the next wave of neurotechnology could quietly enter neurosurgical theaters from the laboratory and – step by step – into everyday life.