Neurointerfaces & brain–computer interfaces

01Overview and value chain

Markers: [EC: EU MDR 2017/745 & BCI neuroethics frameworks | OECD: Biotech & health — neurotechnology & brain–computer interfaces | Regulator: FDA (USA), EMA (EU), NMPA (China)]

Neurointerfaces, or brain–computer interfaces (BCI), form a direct communication channel between the human brain and an external computer, prosthesis or exoskeleton, bypassing the damaged neuromuscular path. The field has moved out of academic labs and into active human trials: Neuralink’s clinical footprint spans five therapeutic programs across four countries with 21 enrolled participants and zero serious adverse device events, Blackrock Neurotech’s Utah Array has been implanted in human research participants for more than a decade of demonstrated intracortical-microstimulation safety, and the FDA cleared Precision Neuroscience’s 1,024-channel Layer 7 cortical interface in April 2025. BCI split into three classes by invasiveness. Invasive intracortical BCI (Neuralink N1 with 1,024 electrodes on 64 flexible threads, Blackrock Utah Array) record single-neuron spikes for the highest spatial resolution. Semi-invasive BCI place electrodes on the cortical surface (electrocorticography, ECoG) or inside a cortical blood vessel (Synchron’s endovascular Stentrode), trading some resolution for surgical safety. Non-invasive BCI use wearable EEG or fNIRS headsets (g.tec, BrainCo) for safe but lower-resolution signals. The decisive technology driver is AI decoding — transformer and recurrent networks that translate chaotic spike trains into text, synthesized speech or smooth robotic trajectories, with published systems reaching above 95% decoding accuracy and over 60 characters per minute. The whole stack is gated by device regulators: FDA Investigational Device Exemption (IDE) trials in the US, EU MDR 2017/745 with notified-body CE certification in Europe, and NMPA registration in China.

The key directions of neurointerfaces & BCI are:

1. Invasive intracortical BCI: micro-electrode arrays implanted inside the cortex (Neuralink N1, Blackrock Utah Array) that record single-neuron action potentials for high-bandwidth control of multi-axis prostheses and fast thought-to-text.
2. Semi-invasive and endovascular BCI: surface ECoG grids and stent-electrodes delivered through the vasculature (Synchron Stentrode, Tsinghua NEO) that balance signal quality against surgical risk without open craniotomy.
3. Non-invasive wearable BCI: EEG and fNIRS headsets (g.tec recoveriX, BrainCo) that read cortex activity through the scalp for rehabilitation, accessibility and consumer use.
4. AI neural decoding: transformer, graph and recurrent neural networks that learn to map spike patterns onto intended movement, speech or cursor trajectories, turning raw neural signal into usable commands.

Sectoral value chain

[Micro-electrode arrays] ──> [Low-power ASIC digitization] ──> [AI intent decoding]
          │                              │                              │
  (MEMS, flexible threads)        (wireless telemetry)         (spike → text / motion)
                                                                         │
[Device actuation] <─── [Closed-loop stimulation] <──────────────────────┘

Value chain levels

Cross-cutting technologies of the sector:

• Robotic micro-thread insertion (“sewing machine”): vision-guided robotic arms that insert flexible polymer electrode threads a few microns wide — thinner than a human hair — directly into the cortex, routing around surface capillaries to avoid micro-haemorrhage.
• Low-noise ASIC biopotential amplifiers: custom integrated circuits that capture and amplify analog neural signals of a few microvolts at the source, digitize them and transmit them wirelessly through the skull by radio.
• Bidirectional closed-loop BCI: systems that not only read cortical activity but write information back through controlled micro-stimulation of sensory cortex, restoring a sense of touch and texture to users of bionic prostheses.

02US

The United States is the epicentre of the commercial BCI breakthrough, with the largest private investment and the most dynamic regulator — the FDA runs an Investigational Device Exemption pathway and a Breakthrough Device programme that have moved invasive BCI into human trials.

Invasive BCI trials, endovascular Stentrode, FDA-cleared cortical layer

• Neuralink (N1 implant, R1 robot): Neuralink’s N1 carries 1,024 electrodes on 64 flexible threads inserted by the R1 surgical robot; its public clinical footprint spans five therapeutic programs across four countries with 21 enrolled participants and zero serious adverse device events, and Nature Medicine has published long-term independent use of an intracortical BCI for speech and cursor control.
• Synchron (endovascular Stentrode): the Stentrode is implanted within the motor cortex via the jugular vein in an endovascular procedure with no craniotomy; Synchron reported positive results from the US COMMAND early-feasibility study and is positioning a 2026 pivotal trial targeting the first FDA commercial approval in motor BCI.
• Blackrock Neurotech (Utah Array) and Precision Neuroscience: Blackrock’s Utah Array underpins the BrainGate program — the longest-running human BCI program — with a published decade of intracortical-microstimulation safety, while Precision Neuroscience received FDA 510(k) clearance for its 1,024-channel Layer 7 cortical interface in April 2025.

03CN

China develops BCI under the state China Brain Project, betting on semi-invasive interfaces and mass deployment of non-invasive BCI in education, medicine and consumer applications, with the NMPA running its own innovative-device track.

China Brain Project, Tsinghua NEO, NeuraMatrix ASIC and non-invasive BCI

• China Brain Project and Tsinghua NEO: the state brain-research programme funds sovereign BCI technology at Tsinghua University and Tianjin University, where the semi-invasive wireless NEO interface (electrodes on the dura, epidural) was implanted in a tetraplegic patient who could then drink independently using a bionic hand.
• NeuraMatrix (BCI ASIC): the Beijing start-up NeuraMatrix (宁矩科技) builds its own low-power, high-channel BCI ASIC chips for neural signal digitization and stimulation, and produces wireless research BCI systems for primates while preparing clinical-grade platforms.
• BrainCo and non-invasive BCI: BrainCo (强脑科技) commercializes non-invasive BCI headsets with NMPA clearance for medical and accessibility use, and has filed for a Hong Kong IPO, leading the Chinese non-invasive segment.

04EU

The European Union emphasizes patient safety, long-term neurorehabilitation and the normative regulation of “neurorights”, pairing its non-profit research centres with the strict MDR 2017/745 device regime.

Wyss Center, Clinatec/CEA, g.tec rehabilitation and neuroethics

• Wyss Center (Geneva): the non-profit Wyss Center develops the fully-implanted Wyss Able ECoG system for home (not only lab) use in locked-in syndrome, and a 5-year follow-up of a fully implanted BCI in a spinal-cord-injury patient was published in 2025; its academic partners have demonstrated a streaming brain-to-voice neuroprosthesis.
• Clinatec / CEA (Grenoble) and g.tec (Austria): Clinatec’s wireless WIMAGINE ECoG implant has driven a four-limb robotic exoskeleton in fully paralyzed patients, while Austria’s g.tec sells the recoveriX non-invasive BCI rehabilitation system — deployed in hospitals for stroke, multiple-sclerosis and Parkinson’s therapy — and runs the BrainSTEP gait-rehabilitation trial.
• MDR and neuroethics: full EU MDR 2017/745 enforcement requires notified-body CE certification, clinical evidence and UDI traceability for any implanted BCI, and EU-funded ethics frameworks address mental-privacy and consent questions specific to neurotechnology.

05Leading companies and research institutes

06Tech stack and innovations

The BCI stack joins a micro-fabricated electrode front end, a hermetically packaged low-power electronics core, and an AI decoding layer that turns raw neural voltage into intent.

1. MEMS flexible-film electrodes and ASIC signal chips:
   • Flexible polyimide threads carry photolithographically defined gold or platinum traces with PEDOT:PSS-coated contacts that lower electrode impedance by orders of magnitude, enabling microvolt-scale recording at the neuron interface.
   • Custom low-power ASIC chips amplify, filter and digitize the microvolt neural signal at the source, then transmit it wirelessly through a hermetic titanium and Parylene C package designed to survive decades in the body.
2. Wireless transcutaneous telemetry:
   • An inductive coil in the implant receives power from an external antenna held over the scalp by magnets, while digitized neural data is sent back out on an ultra-high-frequency radio link, eliminating percutaneous wires that are the main infection route.
   • Endovascular BCI (Synchron Stentrode) instead places the electrode mesh inside a cortical blood vessel, so the signal path avoids craniotomy entirely.
3. AI neural decoding:
   • Transformer, graph and recurrent neural networks learn to map the spike trains of hundreds of neurons onto intended movement or speech, decoding text at above 95% accuracy and above 60 characters per minute in published intracortical systems.
   • Closed-loop variants feed decoded intent back as controlled micro-stimulation of sensory cortex, restoring a sense of touch for users of bionic prostheses.

07Value chains and production pipelines

Production pipeline of an intracortical BCI implant (ISO 13485 / GMP)

┌───────────────────────────┐      ┌───────────────────────────┐
│ 1. MEMS fabrication of    │ ───> │ 2. Laser dicing &         │
│    flexible electrode     │      │    Parylene C coating     │
│    threads                │      │                           │
└───────────────────────────┘      └───────────────────────────┘
                                                  │
                                                  ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 4. EtO sterilization &    │ <─── │ 3. ASIC assembly &        │
│    impedance calibration  │      │    hermetic Ti packaging  │
└───────────────────────────┘      └───────────────────────────┘
               │
               ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 5. Robot-assisted surgical│ ───> │ 6. AI decoder training &  │
│    implantation           │      │    neurorehabilitation    │
└───────────────────────────┘      └───────────────────────────┘

Stage 1: MEMS fabrication of flexible electrode threads

In an ISO 4 cleanroom a liquid polyimide layer is spun onto a silicon carrier and cured, then titanium and gold are sputtered and photolithographically patterned into 16 or 32 micro-channels a few microns wide; a reactive-ion etch opens the circular electrode contact windows.

Stage 2: Laser dicing and Parylene C barrier coating

A picosecond UV laser cuts the individual thread outlines (20–50 micron wide, looped for elasticity), the threads are released in a deionized-water bath, and a CVD Parylene C layer about 1 micron thick is deposited to protect the polyimide from hydrolysis.

Stage 3: ASIC assembly and hermetic titanium packaging

The polymer threads are wire-bonded to the digitizing ASIC, which is sealed with an inductive coil inside a Grade-5 titanium housing by laser welding under helium; a helium leak test verifies hermeticity and the thread exit points are potted with medical silicone.

Stage 4: EtO sterilization and impedance calibration

The assembled device is cleaned ultrasonically, sterilized with ethylene oxide (EtO), degassed, and every electrode’s electrical impedance is measured in sterile saline before double-pouch sterile packaging.

Stage 5: Robot-assisted surgical implantation

A minimally invasive 25 mm burr hole is made over the target cortex, a robotic neurosurgical system (e.g. Neuralink R1) maps surface vasculature and inserts the electrode threads into cortical layers IV–V at roughly 10 threads per minute, and the titanium implant is seated flush with the skull.

Stage 6: AI decoder training and neurorehabilitation

About ten days after surgery the patient wears an external wireless transceiver, and calibration sessions train the decoder network to recognize the spatial spike vectors of intended movement; over roughly two weeks decoding accuracy reaches above 95%, enabling text entry above 60 characters per minute or control of a bionic manipulator.