NeuroXess's 256-channel flexible cortical-surface array used for real-time motor and Mandarin speech decoding at Huashan Hospital. Reported in temporarily implanted epilepsy and tumor patients; some figures are company announcements.
Independent, source-linked catalog sheet. Not a manufacturer-issued datasheet, regulatory decision or instructions for clinical use. Human evidence does not establish approval. Source-specific restrictions, conflicts and missing specifications are retained below.
NeuroXess 256-channel flexible ECoG BCI
NeuroXess has described a 256-channel flexible cortical array that decoded motor intent and Mandarin speech in patients who were implanted for clinical epilepsy or tumor work. The hardware details here come from the Science Advances paper; accuracy claims beyond the paper come from the company. A secondary source called the array penetrating; the paper describes it as a surface ECoG grid.
Identity
Field
Value and source scope
Device
NeuroXess 256-channel high-density flexible ECoG array with a skull-fixed headstage, used with the XessOS decoding software [1][2][3]
Manufacturer
Shanghai NeuroXess Technology Co., Ltd. (脑虎科技), founder Tiger H. Tao [1][4]
Interface class
Flexible cortical-surface (ECoG) array; the peer-reviewed paper calls it a 256-channel microelectrocorticographic BCI [2]
Origin
Company device with Huashan Hospital, Fudan University, supported by the Tianqiao and Chrissy Chen Institute and Shanghai and national grants [1][2]
First demonstrated
August 2024: real-time motor decoding in a 21-year-old epilepsy patient at Huashan Hospital (company announcement) [1]
First human implant
August 2024 per NeuroXess’s January 2025 announcement. Secondary coverage dates a company disclosure to April 17, 2025; the company’s own announcement dates the implant earlier, so that date is used [1]
Species studied
Human; NeuroXess also sells flexible depth electrodes for animal research, with single-unit recording in mice reported up to 10 months (company claim) [1][5]
Regulatory status
Investigational; implanted in epilepsy patients undergoing clinical seizure or lesion monitoring under hospital IRB approval. No market approval found [1][2]
Function
Record high-gamma cortical activity (70 to 150 Hz) and decode motor intent and Mandarin syllables in real time [1][2]
Target tissue
Cortical surface; in the Mandarin study the array covered the middle and superior temporal gyri, ventral sensorimotor cortex and part of the pars opercularis [2]
Geometry and architecture
Field
Value and source scope
Interface type
Flexible high-density ECoG grid on the cortical surface, headstage fixed to the skull [2]
Array layout
256 electrodes; layout beyond count and pitch unreported [2]
Electrode count
256 channels [1][2]
Pitch
3 mm center to center [2]
Electrode lengths
Unreported
Shank width and thickness
Unreported
Tip and exposed site geometry
Each recording contact 1.3 mm in diameter [2]
Contact coating
Unreported
Insulation
Unreported
Insertion method
Placed on the cortical surface during surgery, in the cases reported as part of epilepsy localization [2]
Anchoring and fixation
Unreported
Electrode and channel physics
Field
Value and source scope
Exposed site area
Unreported
Electrode material
Unreported
Impedance (with measurement frequency)
Unreported
Noise floor or SNR
The paper reports generally high SNR with minimal drift across days and no significant new bad channels over 11 days; no numeric SNR extracted [2]
Recording modality
Cortical surface potentials; features taken from the high-gamma band (70 to 150 Hz) [1][2]
Sampling rate
15 kHz raw, downsampled to 400 Hz for offline processing [2]
Stimulation capability
Unreported
Charge injection limit
Unreported
Reference and ground
Unreported
Tissue interface and bioresponse
Field
Value and source scope
Target tissue
Cortical surface, over temporal and ventral sensorimotor regions in the Mandarin study [2]
Insertion trauma and BBB disruption
Unreported
Vascular disruption risk
Unreported
Micromotion sensitivity
Unreported
Gliosis and encapsulation
Unreported
Neuron loss near sites
Unreported
Foreign-body response mitigation
Unreported
Typical failure modes
Not reported in the sources read; the array performed reliably over 11 days of monitoring [2]
System architecture
Field
Value and source scope
Onboard electronics
Unreported
Data path
Headstage fixed to the skull; the downstream link to the acquisition system is not described in the sections read, and wireless fully implanted operation is not described for this array [2]
Telemetry bandwidth
Unreported
Sampling rate
Unreported
Power
Unreported
Thermal management
Unreported
Packaging and hermeticity
Unreported
MRI compatibility
Unreported
Surgical complexity
Craniotomy placement of a surface grid as part of epilepsy monitoring; electrode placement guided by clinical need [2]
Output connectors
Unreported
Performance envelope
Field
Value and source scope
Acute yield
Over 11 days of monitoring about 9 hours of data were collected; no new bad channels emerged [2]
Chronic yield
Unreported
Stability over time
Unreported
Longevity
11 days of intracranial monitoring in the Mandarin paper; chronic duration unreported [2]
Revision and explant experience
Unreported
Adverse events
Unreported
Notable demonstrations
Peer-reviewed: median offline accuracy of 71.2% over 394 Mandarin syllables in a single-character reading task (Science Advances 2025). Company announcement: 71.2% accuracy across 142 common syllables within five days, decoding latency under 100 ms per character, and motor decoding with system latency under 60 ms [1][2]
Clinical and preclinical evidence
Field
Value and source scope
Human subjects
One 21-year-old epilepsy patient with a motor-cortex lesion (motor decoding) and one epilepsy patient with a language-cortex tumor (December 2024), per the company; the Science Advances paper reports one 43-year-old woman [1][2]
Preclinical cohort
Unreported
Follow-up duration
Days to about two weeks of in-hospital monitoring in the reported cases [1][2]
Indications
Motor, language and visual function restoration are listed as company goals; reported cases were epilepsy and tumor patients [1][5]
Trials and registries
Huashan Hospital IRB approval KY2024-842 for the paper; no registry identifier found [2]
Primary outcomes
Syllable and sentence decoding accuracy and latency; see Notable demonstrations [1][2]
Key limitations
Reported patients were temporarily implanted for clinical epilepsy or tumor care, not paralyzed users with a chronic implant. Few subjects. Company announcements are not peer reviewed [1][2]
Engineering tradeoffs
Field
Value and source scope
Strengths
High channel count of 256 over a 3 mm grid allowed fast functional mapping and decoding of both motor and Mandarin speech in days [1][2]
Limitations
Surface potentials have lower spatial resolution than penetrating arrays; reported use was short-term [2]
Scaling constraints
Unreported
Version boundary
The 256-channel array in the January 2025 announcement and the Science Advances paper. NeuroXess’s separate animal-research depth electrodes (4 um thick) are not this device.