A 29-year-old art teacher who could no longer move his fingers after a spinal-cord injury picked up a brush again — guided not by his hands, but by signals read from the surface of his brain. A few months earlier, a different patient ordered his own New Year's groceries with a thought. These are not prototypes behind glass. They are early human cases from a Chinese brain-computer interface (脑机接口) that just crossed a line most brain chips never reach: a regulated clinical trial.
The device behind those moments is the "三全" (fully implanted, fully wireless, fully functional) system from NeuroXess (脑虎科技), a Shanghai life-science company founded in 2021. What makes it worth a closer look is not the demo videos. It is the route it takes into the skull — and the regulatory door that just opened for it.
The implant that sits on the brain, not in it
Most people picture a brain chip as something pushed deep into tissue, the path Neuralink made famous. The "三全" system goes the other way. It is subdural (硬膜下), meaning a flexible electrode is laid on the surface of the cerebral cortex, just under the dura mater, without penetrating the brain itself. The company sums up the safety logic in three plain phrases: 不伤脑 (no brain damage), 不烧脑 (no brain burning), and 易推广 (easy to spread).
The "no brain burning" part is a real engineering choice, not a slogan. The system splits into two parts: the electrode on the brain, and the battery, wireless transmitter and charger placed under the chest skin. Keeping the heat-generating components away from the brain avoids the local temperature rise that worries clinicians about long-term implants. The surgery reuses the mature DBS (deep brain stimulation) routine, needs no dedicated robot, and — the company argues — could be performed at hundreds of Chinese tertiary hospitals.
On the numbers the system has published: full-link latency under 50 milliseconds, and a brain-controlled cursor decode rate of 5.2 bits per second. Those are modest next to a mouse, but they are enough to drive a pneumatic glove or a cursor for someone who has no other way to move.
From one patient to a regulated trial
The leap from a single implanted patient to a registrable product happened fast. On July 7, 2026, NeuroXess started its GCP (Good Clinical Practice) registration clinical trial at Huashan Hospital (复旦大学附属华山医院) in Shanghai, with neurosurgeon Mao Ying (毛颖) as principal investigator. This is the first domestic subdural implant to enter the Class III (第三类) medical-device registration pathway.
The trial is registered as NCT07720882: a prospective, multicenter, single-arm study across 15 tertiary hospitals in a dozen provinces, planning to enroll 32 patients with cervical spinal-cord injury and resulting quadriplegia. The primary endpoint is the responder rate on the Action Research Arm Test at three months — a standard, conservative measure of whether the hand function actually improves.
Then the regulators stepped in
The timing was not an accident. On June 30, 2026, China's National Medical Products Administration (NMPA, 国家药监局) published two guidance documents — the Brain-Computer Interface Medical Device Product Classification Guideline and the BCI Medical Device Generic Naming Guideline — the country's first formal rulebook for the field. Under them, any implantable or invasive BCI is managed as a Class III (第三类) device, the highest-risk category, with classification code 12-00.
NeuroXess entered NMPA's innovation "green channel" (绿色通道) — the special review program for novel devices — on July 21, 2026, the first subdural flexible BCI to qualify. That does not mean approval; it means the review is fast-tracked and the company is now collecting the data NMPA will actually judge.
Why the "safety-first" path keeps winning in China
Step back and a pattern shows up. China's leading implants are deliberately avoiding the penetrating-electrode route:
- NEO, from Neuracle (博睿康) with Tsinghua University's Hong Bo (洪波) team, was approved by NMPA on March 13, 2026 — the world's first approved implantable BCI, an extradural (硬膜外) 32-channel device that lets paralyzed patients control a pneumatic glove.
- BeiNao-1 (北脑一号), a 128-channel extradural flexible system from the Beijing institute and NeuCyber (芯智达), began multicenter GCP trials in 2026.
None of these spear into brain tissue. The bet is that doctors adopt the implant they can defend to a patient, not the one with the sharpest demo — and that a larger safety margin is what turns a research milestone into something hospitals will actually implant.
What readers can do now
- If you follow medtech: track NMPA registration status and the innovation green channel, not implant count. A 32-patient single-arm trial clearing its endpoints is the real signal; full approval is still a year or more out.
- If you work in rehab or neurology: today these systems target cervical spinal-cord injury and quadriplegia for hand-function restoration. They are not yet tools for stroke, ALS or speech broadly — watch the trial's secondary endpoints before generalizing.
- If you build hardware: the durable bottleneck remains the electrode–tissue interface and signal decoding, not compute. That is where the next decade of value sits.
Honest limitations
Figures in this article come from NeuroXess's official announcements, Huashan Hospital / People's Daily Shanghai reporting, China Securities Journal (上海证券报), and the clinical-trial record brainmed.com; the NMPA guideline dates and classification are from NMPA's official site. The trial is early-stage (32-patient, single-arm, primary endpoint expected in 2027) and not yet backed by large peer-reviewed efficacy data. Patient-level functional gains (the art teacher, the New Year's grocery order) are company and hospital disclosures, not independently audited outcomes. "Subdural" is still surgery under the skull — it is not non-invasive. Comparisons to NEO and BeiNao are directional, reflecting different implant sites and patient populations, not head-to-head studies.