16 x 16 gold-on-polyimide micro-ECoG array with 850 micrometer sites at 1250 micrometer pitch, in a titanium-housed implant with four Intan chips. 203 days in a Labrador, then intraoperative and short-term human motor-imagery decoding at Huashan Hospital. SIMIT with NeuroXess.
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.
A cortical-surface array from Zhou Zhitao’s group at the Shanghai Institute of Microsystem and Information Technology (SIMIT, CAS) and Tao Hu’s team at NeuroXess, tested clinically at Huashan Hospital, Fudan University (Wu Zehan is a corresponding author). It is a different array from the NeuroXess 256-channel flexible ECoG, which the sources there describe with 3 mm pitch. Details are from the Advanced Science paper (2025, full text via Europe PMC) and SIMIT’s Chinese release.
Identity
Field
Value and source scope
Device
µECoG array, flexible printed circuit, signal processing unit and titanium enclosure, with a decoding and device-control software stack [1][2]
Origin
SIMIT with NeuroXess and Huashan Hospital [2]
Interface class
Cortical-surface array, mesh recording area [1]
Species studied
One Labrador (male, about 18 months, 30 kg) for 203 days; human intraoperative and short-term trials [1]
Regulatory status
Research; Huashan Hospital IRB approval KY2021-918; participants were tumor patients needing awake surgery [1]
Geometry and architecture
Field
Value and source scope
Electrode count
256 channels, 16 x 16 [1]
Site diameter and pitch
850 µm sites, 1250 µm apart; 64 electrodes/cm2, which the SIMIT release calls 64 times the density of clinical ECoG [1][2]
Layers
Ultrathin mesh recording area with thickened lead zones for robustness [1]
Implant electronics
Four Intan RHD2164 chips; flexible printed circuit; customized titanium enclosure, watertight and airtight [1]
Materials and fabrication
Field
Value and source scope
Process
Sandwich of gold in polyimide, adapted from the group’s earlier silk-enabled bioelectronics work [1]
Layers
7 µm PI-2610 on a silicon wafer; Cr 50 angstrom and Au 150 nm by e-beam evaporation with lift-off; 13 µm PI encapsulation [1]
Openings
Aluminum etch mask, aluminum etched, oxygen plasma dry etch for site openings and perfusion holes [1]
Release
Hydrofluoric acid etch from the wafer [1]
Pre-implant test
More than 95% of electrodes under 1 MΩ at 1 kHz; RMS noise under 2 µV in PBS [1]
Performance envelope
Field
Value and source scope
Canine stability
203 days; electrode yield fell 5.49%; SNR stayed above 20 dB; real-time 3D motor decoding mean accuracy 0.84 [1][2]
Decoders
LSTM and position-velocity Kalman filter on high-gamma power spectral density features; recalibration 5 to 7 minutes between sessions [1]
Human intraoperative
After 7 minutes of training the participant played ping-pong (1D, mean accuracy 0.90) and snake (2D) by brain signals [1]
Human short-term
Implanted up to about 12 days; 19.87 h of training; center-out up to 1.13 bits/s, WebGrid up to 4.15 bits/s after interface changes [1][2]. The SIMIT release compares this with a 4.60 bits/s figure for a Neuralink participant; that comparison is the institute’s, not the paper’s
Density effect
Within a 2 x 2 cm area, higher electrode density improved decoding without more coverage [1]
Limits
Human follow-up is days, not months. Participant numbers beyond the quoted sessions and supplementary figures were not read. The “comparable to intracortical” language is the authors’ framing.