TechnologyHow do neural interfaces work?

How we interface with the nervous system

Every neural interface trades access, resolution and invasiveness against each other. This explorer places 32 of them on one map, by what they do and how far they have come, so that an EEG cap, a deep brain stimulator and an intracortical array can be read on the same terms.

BCI is evolving from decoding the nervous system to interacting with it.

Reading
Writing
Closed loop
Biohybrid

Clinical ECoG grid

Layer 7

Neuropixels 1.0

Neural interfaces, measuredElectrode size and pitch at true scale, and where each interface sits, on the dura, under it or in the cortex, and how it gets there. Each square above is 1 mm of cortex.

AccessMechanismTargetMaturityUse case

32 interfaces

Access ↓

Read

Sense neural activity

Bidirectional

Sense and stimulate in closed loop

Write

Influence neural activity

Non-invasive

Around the head

On the skin

Minimally invasive

Vessel or skull window

Invasive

Nerve or sense organ

On the cortex

In the cortex

Deep brain

Biohybrid · Adjacent frontier

Living neural tissue outside the body. Not an interface with a person.

Fig. 01The interface design spaceSelect up to 4 interfaces to compare.

Colour · function

Read
Write
Bidirectional
Biohybrid

Fill · maturity as a clinical modality

Established
Human feasibility
Early human
Preclinical
Research frontier
Adjacent frontier

ComparingECoG · sEEG · Intracortical

Compare mode

Two to four, side by side

Select interfaces on the map, or start from a pairing that answers a common question. Only characteristics that apply to every interface chosen are compared.

Add from the map

Useful comparisons

Comparison of Electrocorticography, Stereo-EEG, Intracortical arrays
Dimension
Function	ReadSense neural activity	ReadSense neural activity	ReadSense neural activity
Access	InvasiveA conventional grid under the dura is placed through a craniotomy. Epidural implants open the skull too, to the size of the implant, but leave the dura closed; films can be slid under the dura through a slit, and soft grids are being designed to unfold through a burr hole. Some of these systems are classed as minimally invasive in the company profiles.	Invasive	Invasive
Mechanism	Electrical	Electrical	Electrical
Target	Cortex	Deep brain, Cortex	Cortex
Information flow	Cortex → surface grid	Deep structures → depth electrodes	Cortex → penetrating array
Spatial resolutionCentimetres → Single neurons	Millimetre to sub-millimetre	Millimetre, sparse	Single neurons
Temporal resolutionSeconds → Sub-millisecond	Milliseconds	Milliseconds	Sub-millisecond
Signal richnessIndirect or summed → Single-unit detail	Population activity, including high gamma	Field potentials along each shaft	Spikes from individual neurons
CoverageOne small site → Whole brain	Several cm² per array	Sparse but deep and distributed	A few mm² per array
InvasivenessNone → Open, penetrating	Craniotomy; slit delivery for thin films	Stereotactic burr holes	Craniotomy; penetrating insertion
Long-term stabilityVariable → Stable for years	Stable over months to years in studies	Mostly temporary monitoring	Changes over months to years; varies by design
Maturity as a clinical modality	EstablishedRoutine in epilepsy surgery for mapping and monitoring.	EstablishedRoutine in planning epilepsy surgery.	Human feasibilityResearch implants in people since 2004. Not a routine clinical tool.
Maturity as a BCI application	Human feasibilityResearch implants in several countries. One epidural system is authorised in China for assisted grasp.	Early humanInterface research in monitored patients, during hospital stays of days to weeks.	Human feasibilityFeasibility studies across several groups and companies; the basis of the leading speech results.
Strengths	Stable signals that support long recalibration-free use Broad coverage of speech and motor areas Does not penetrate cortex	Reaches deep structures surface arrays cannot Established neurosurgical workflow	Highest information rate demonstrated in people Basis of the leading speech and handwriting results
Limitations	Cannot isolate single neurons Still requires opening the skull in most designs	Sparse sampling Most data come from short hospital stays, not chronic implants	Samples a small patch of cortex Long-term signal stability and tissue response remain open questions
Common applications	Speech and communication Computer and device control Mapping function before and during epilepsy and tumour surgery	Epilepsy assessment Speech and memory decoding research	Speech and communication Cursor and device control Robotic arm control
Human evidence	Speech decoded at a median 78 words per minute from a surface grid in one participant (2023), and a 128-channel implant that kept its control across days without recalibration (2021). A 32-participant trial supported the registration of an epidural system in China; its results are unpublished.	Many research studies in people who have the electrodes for clinical reasons. Little chronic interface data.	Two decades of peer-reviewed results, mostly from single participants: handwriting at 90 characters per minute (2021), speech at 62 words per minute (2023), 97.5% word accuracy held over 8.4 months (2024), two-handed typing at up to 110 characters per minute (2026), and more than 3,800 hours of use at home by one participant with no researcher present (2026).
What needs to happen next	Fully implanted wireless systems with high channel counts, and trials large enough to support wider authorisation.	Chronic depth-electrode systems designed as interfaces, not borrowed from monitoring.	Longevity data for the newer electrode designs, and a pivotal trial.
Related research	Brosler SC et al. 2026 Leinders S et al. 2026 Metzger SL et al. 2023 Lorach H et al. 2023 +10 more	None in our corpus yet	Card NS et al. 2026 Card NS et al. 2024 Wairagkar M et al. 2025 Willett FR et al. 2023 +13 more
Related companies	Precision Neuroscience CorTec ONWARD Medical INBRAIN Neuroelectronics Motif Neurotech +11 more	None in our corpus yet	Neuralink Paradromics Blackrock Neurotech BrainGate StairMed +2 more

Bars run from low to high and show a range, not a score. A wide bar means the characteristic varies across implementations. Qualitative editorial ranges; specific devices differ.

Descriptions, ranges and maturity levels are editorial analysis, kept qualitative where the literature does not support exact comparison across technologies. Specific devices differ.

All 32 interfaces as a table

Every interface, with its function, access, mechanism, target and maturity
Function	Access	Mechanism	Target	Clinical modality	BCI application
Read	Non-invasive	Electrical	Cortex	Established	Human feasibility
Read	Non-invasive	Magnetic	Cortex	Established	Research frontier
Read	Non-invasive	Optical	Cortex	Human feasibility	Research frontier
Read	Non-invasive	Electrical	Muscle / neuromotor	Established	Established
Read	Minimally invasive	Electrical	Cortex	Human feasibility	Human feasibility
Read	Minimally invasive	Acoustic	Cortex, Deep brain	Early human	Preclinical
Read	Invasive	Electrical	Peripheral nerve, Muscle / neuromotor	Early human	Early human
Read	Invasive	Electrical	Cortex	Established	Human feasibility
Read	Invasive	Electrical	Cortex	Human feasibility	Human feasibility
Read	Invasive	Electrical	Deep brain, Cortex	Established	Early human
Write	Non-invasive	Magnetic	Cortex	Established	Research frontier
Write	Non-invasive	Electrical	Cortex	Human feasibility	Research frontier
Write	Non-invasive	Acoustic	Cortex, Deep brain	Early human	Research frontier
Write	Invasive	Electrical	Peripheral nerve	Established	Early human
Write	Invasive	Electrical	Retina	Human feasibility	Human feasibility
Write	Invasive	Electrical	Cochlea	Established	Established
Write	Invasive	Electrical	Brainstem	Established	Established
Write	Invasive	Electrical	Cortex	Established	Early human
Write	Invasive	Electrical	Cortex	Early human	Early human
Write	Invasive	Optical, Molecular	Cortex, Deep brain, Retina	Early human	Preclinical
Write	Invasive	Electrical	Deep brain	Established	Early human
Bidirectional	Invasive	Electrical	Peripheral nerve, Muscle / neuromotor	Early human	Early human
Bidirectional	Invasive	Electrical	Cortex	Established	Early human
Bidirectional	Invasive	Electrical	Cortex, Deep brain	Established	Human feasibility
Bidirectional	Invasive	Electrical	Cortex, Spinal cord	Early human	Early human
Bidirectional	Invasive	Electrical	Cortex	Early human	Early human
Bidirectional	Invasive	Electrical	Deep brain, Cortex	Established	Early human
Bidirectional	Invasive	Electrical	Deep brain	Established	Human feasibility
Biohybrid	Ex vivo	Electrical, Biological	Cultured neural tissue	Adjacent frontier	Adjacent frontier
Biohybrid	Ex vivo	Electrical, Optical, Biological	Cultured neural tissue	Adjacent frontier	Adjacent frontier
Biohybrid	Ex vivo	Biological, Electrical	Cultured neural tissue	Adjacent frontier	Adjacent frontier
Biohybrid	Ex vivo	Biological, Electrical	Cultured neural tissue	Adjacent frontier	Adjacent frontier

Use-case explorer

Start from what people want to do

Nine things people want an interface to do. Each links the technology, the companies, the research, the studies and the tools behind it, and says what is available today.

Decoding attempted speech or writing into text and voice.

Can someone who cannot speak hold a conversation again?

Research participants with paralysis have used implanted BCIs to produce text and synthesised voice from attempted speech. All such systems remain in early feasibility studies.

Latest milestoneA single surface implant decoded speech and upper-body gestures at the same time in three participants.

Technology: EEG MEG / OPM Endovascular ECoG Intracortical sEEG
Companies: Neuralink Synchron Precision Neuroscience Paradromics Blackrock Neurotech BrainGate StairMed NeuroXess NeuCyber NeuroTech ABILITY Neurotech JiMED g.tec medical engineering Araya Cognixion
Research: Simultaneous speech and gesture decoding for multimodal communication in paralysis Long-term independent use of an intracortical brain–computer interface for speech and cursor control Implanted brain-computer interface functionality during nighttime in late-stage amyotrophic lateral sclerosis An Accurate and Rapidly Calibrating Speech Neuroprosthesis An instantaneous voice-synthesis neuroprosthesis+13 more
Clinical: VOICE Connect-One BrainGate2 BRAVO UNP CortiCom+3 more
Tools: BRAND torch_brain img_pipe MNE-Python Braindecode
Datasets: Brain-to-Text '25 Brain-to-Text Benchmark '24 LibriBrain / LibriBrain100 Podcast ECoG
Frontier: Speech & Communication Cognitive Interfaces
On the calendar: BCI Meeting 2027

Why do such different devices exist for the same goal? Each trades signal quality against surgical access. Read the dossier

The interface map