A machined titanium case sits on a bench, roughly the width of a coin. A row of platinum pins passes through its wall, each one sealed into ceramic, each one a pinhole that salt water spends years trying to enter.
No press release ever shows this part. Photographs show electrodes, patients, and the demo where a cursor moves.
The feedthrough is the reason the demo can be implanted at all.
Key takeaways
- China's brain science mega-programme launched three brain-computer interface (脑机接口) projects in late October 2026; one exists solely to develop a high-density hermetic feedthrough.
- A feedthrough carries every signal and every watt of power through the wall of an implanted case without letting body fluid in. It is a seal and a connector.
- Density is the difficulty: published work describes 100-plus feedthroughs inside a 10 mm package, sealed to helium leak rates near 10⁻¹⁰ Pa·m³/s.
- The implant that opened the commercial phase — Neuracle's NEO, approved in March 2026 — reset the bar for every supplier upstream of it.
- Funding a part is not manufacturing it. The projects have just entered implementation, and no output has been announced.
What the component actually does
An implanted brain device has two halves that must never meet. Inside the case: an acquisition chip, a radio, sometimes a battery. Outside: electrodes sitting in tissue, bathing in warm saline that will destroy the chip in weeks if it gets in.
The feedthrough is the wall's only opening. It must pass tens or hundreds of conductors through a surface measured in millimetres, keep each one electrically isolated from its neighbours, survive repeated sterilisation, and hold a vacuum-grade seal for a decade.
Beijing's own description is compact: it connects the implanted acquisition chip to the external signal-processing equipment while isolating the electronics from body fluid. Everything about long-term survival runs through that sentence.
Why a national programme was created for a connector
On 28 October 2026, the Chinese Institute for Brain Research, Beijing (北京脑科学与类脑研究所) held the kickoff for three projects under the national "Brain Science and Brain-Like Intelligence" mega-programme. Seventeen institutions are involved, including device makers, test institutes and hospitals.
The three projects map onto three different failure modes:
- A flexible, high-throughput semi-invasive system for motor reconstruction, led by Luo Minmin (罗敏敏).
- The high-density medical feedthrough and system integration, led by Zhang Lei (张垒).
- Multi-region motor decoding that stays stable over months, led by Cui He (崔翯).
Two of the three are about software and electrodes, the parts that get written about. The middle one is about a seal.
A national programme carving out a whole project for one passive component is itself the argument. Governments do not fund connectors unless the connector blocks them.
The arithmetic that makes it hard
A 2018 IEEE EMBC paper from a Chinese team described a ceramic-and-titanium package carrying more than 100 feedthroughs through a 10 mm diameter, with helium leak rates on the order of 10⁻¹⁰ Pa·m³/s. The authors argued the joints were qualified for a century of implantation. That is a laboratory claim, not a shipped product.
American contract manufacturers already advertise multi-pin arrays past 100 channels, with helium leak detection to 1×10⁻¹⁰ atm·cc/sec, under ISO 13485. The capability exists. It is not cheap.
Each added channel is another hole in a pressure boundary, with its own seal, its own insulation path, and its own chance to end a patient's implant.
Demand is real, and smaller than the pitch
On 13 March 2026, the National Medical Products Administration (国家药品监督管理局) approved Neuracle's implantable BCI hand-function compensation system — registration number 国械注准20263120537 — for adults aged 18 to 60 with C2–C6 spinal cord injury graded A to C. The device sits outside the dura, is powered wirelessly, and drives a pneumatic glove.
Approval converted a research question into a procurement question. In its filing for a Shanghai listing, the company put a list price of RMB 475,000 (≈ US$67,000) per system, said it was listed in 27 provinces, and reported 13 units sold and eight implanted since sales began in July, with more than 70 orders and expressions of interest.
That is the honest scale of the pull: real, priced, and still counted in tens. Anyone tooling up for thousands of feedthroughs would be building ahead of the order book.
What a domestic feedthrough would not fix
A feedthrough is necessary and nowhere near sufficient. Signal drift, electrode degradation, surgical precision and decoder generalisation all sit downstream of it, and none of them improve because a seal got better.
There is also a trap in the phrase "domestic substitution". A part that passes a bench test is not a part that passes a decade inside a person. Qualification for implantable use runs through biocompatibility, accelerated ageing, and design-transfer reviews that take years, not quarters.
Honest limitations
The project's existence and structure come from the Beijing institute's own announcement and from official and financial media coverage of the same kickoff; its technical targets are not public, and no funding figure has been disclosed. The density and leak-rate figures come from a 2018 conference paper describing a research package, not from any Chinese product cleared for sale, and the "100-year" framing is the authors' extrapolation. The sales figures are from the company's reply to IPO reviewers as reported by two financial outlets; I did not read the filing itself. The NEO approval was checked against the NMPA registration database. I could not verify that any Chinese implant on sale uses an imported feedthrough — that claim circulates widely but traces back to commentary, not to a manufacturer or regulator. Read the supply-chain argument as a structural inference from where public money went, not as an audited bill of materials.
What readers can do now
- If you track neurotech, put feedthrough, packaging and test capacity on the same watchlist as electrode channel counts. The first two decide whether the third ever reaches a patient.
- If you invest, ask implant makers a blunt question: who seals your case, and have they qualified at implant grade? A blank look is an answer.
- If you are an engineer, read the 2018 EMBC packaging literature before assuming the hard part is the electrode. The seal is where multi-year failures concentrate.