Companies to lead China's applied state research | AMEC ships etch tool for 3D chips | iFlytek trains model on Chinese chips

Your morning brief on Chinese technology.

Bottom Line Up Front

Party and State Council offices call for company-led research in a joint directive saying state science projects with a clear industrial goal should in principle be led by companies. AMEC shipped the first unit of its gate-all-around (GAA) etch tool to a major domestic chip customer for production line validation after lab tests showed it removed silicon-germanium more than 700 times faster than adjacent silicon. iFlytek released its Spark X2.6 large language model, which has 293 billion parameters and which the company says was trained entirely on domestic chips. Market regulator official Li Changqing credited regulator probes with driving 71% of the 20.4 million smart cars recalled in China through September.

A joint directive from the general offices of the Communist Party Central Committee and the State Council, China's cabinet, says state science projects with a clear industrial goal should in principle be company-led. Companies running those projects would also lead the commercialization of the results, giving industry a direct say in what state-funded applied research pursues and how it reaches the market. The document seeks decisive breakthroughs in fields including integrated circuits, industrial machine tools and high-end instruments. It names quantum technology, brain-computer interfaces and 6G mobile networks among frontier fields to build into future industries. Private companies would also gain access to a pilot of the National Natural Science Foundation of China, the main state funder of basic research.

Read at State Councilgovt ↗

Chip equipment maker AMEC ships first etch tool built for next-generation 3D chips

AMEC, a Chinese maker of chipmaking equipment, shipped the first unit of its Primo Domingo etch tool to an unnamed major domestic chip research and manufacturing customer. AMEC says it fills a domestic gap in an etch step needed for gate-all-around (GAA) transistors, the design used at leading logic nodes, and for stacked 3D memory chips. Etching carves circuit features by removing material. In lab tests the tool removed silicon-germanium more than 700 times faster than the silicon beside it, the selectivity that layers nanometers apart require. Launched in March, the tool now enters validation on the customer's production line, where a chipmaker checks whether lab results hold in its own manufacturing.

Read at ITHome ↗

iFlytek releases large language model it says was trained entirely on Chinese chips

Chinese AI company iFlytek released Spark X2.6, a large language model with 293 billion parameters that it says was trained entirely on domestic chips. iFlytek says the limited memory bandwidth of domestic chips, how fast they move data to and from memory, drags down training, so the model tests how far Chinese hardware can carry a large build. Chairman Liu Qingfeng said other labs train on Nvidia cards and put domestic computing about two generations behind global benchmarks, a gap he said would soon narrow. Developers can use the model through iFlytek's developer interface for a time-limited 0.9 yuan (13 U.S. cents) per million input tokens, the word fragments models read.

Read at ITHome ↗·Read at ITHome ↗

China's market regulator says its probes drove 71% of smart car recalls this year

Regulators' investigations, interviews and on-site checks drove 71% of the 20.4 million smart cars recalled in China through September, said Li Changqing, an official of the State Administration for Market Regulation (SAMR). SAMR is China's product safety regulator. Li said vehicle defects are moving from hardware into software, which puts the code carmakers ship and update remotely inside a recall system the regulator mostly drives. He called over-the-air updates, software pushed to cars remotely, a new risk variable. A national standard SAMR approved in September sets specific rules for carrying out recalls that way.

Read at ITHome ↗

POLICY & REGULATION

China's industry ministry starts work on second batch of conditional self-driving approvals

Work on a second batch of Level 3 autonomous driving pilot approvals is under way, said Guo Shougang, a department head at the Ministry of Industry and Information Technology (MIIT), following the ministry's pledge in the October 21st edition to advance such pilots cautiously. Level 3 cars drive themselves under set conditions while the driver stays ready to take over, so a second batch would extend that permission to more models. Changan's Deepal SL03 and BAIC's Arcfox Alpha S won the first approvals in December 2025, each limited to single-lane driving on designated highways and expressways.

Read at National Business Dailystate media ↗·Read at CnEVPost ↗

III

ROBOTICS & AUTONOMOUS SYSTEMS

Tsinghua and AgiBot say robots learning from fleet experience outperformed human operators

Robots trained with a method from Tsinghua University, robot maker AgiBot and Infinigence AI beat people steering them remotely by an average 16.7 percentage points in success rate across five real-world tasks. Human correction of robots does not scale as fleets grow, so learning from the fleet's own successful runs lets deployed robots improve without a matching rise in expert labor. When a robot fails, the method retrieves similar successful moments from the fleet's pooled runs and trains on them. Humans only label each run's outcome. Across 120 trials per task on robots that contributed no training data, success rates ranged from 93.3% to 99.2%, the team reports in a preprint posted before peer review.

Read at 36Kr ↗

QUANTUM & FRONTIER SCIENCE

Zhejiang University team claims China's first purified silicon-28 gas and wafers for quantum chips

Last week a Zhejiang University team released silicon-28 silane, a gas used to grow crystal layers, and 4-inch wafers carrying such a layer, which ITHome reports are China's first of each. Natural silicon contains silicon-29, whose nuclear spin disturbs the quantum bits in silicon chips, so a domestic source of purified silicon-28 gives Chinese silicon quantum-chip research a home-built supply of its base material. State nuclear conglomerate China National Nuclear Corp. said in June it had begun mass-producing silicon-28 at above 99.99% isotopic abundance, a step ITHome says covered isotope enrichment but not conversion into wafer material. National Institute of Metrology tests put the gas's silicon-28 share at 99.99576%, which ITHome notes is material-preparation test data.

Read at ITHome ↗

Beijing quantum computer maker claims China's first continuous atom reloading, on a machine with a 4,000-qubit array

Beijing startup Nakai Quantum says its Qiyuan-1 neutral-atom quantum computer can continuously load fresh atoms, which 36Kr reports no Chinese team had shown publicly before, and that the same machine has assembled a fully occupied array of about 4,000 qubits (quantum bits). Neutral-atom machines hold atoms in laser beams as qubits and lose them as they run, so they normally pause to reload. Continuous loading keeps neutral-atom machines running for hours rather than fractions of a second, which long, useful quantum computations depend on. Harvard's Lukin group first showed continuous loading in 2025, keeping more than 3,000 atoms running for over two hours. Nakai says it carries fresh atoms in on one moving laser conveyor where Lukin's group used two. Qiyuan-1 is scheduled for formal release in the fourth quarter of 2026.

Read at 36Kr ↗

DIGITAL INFRASTRUCTURE

China's central bank launches new digital-yuan system and smart contract framework

The People's Bank of China's digital currency center released DCEPS, a central-bank-side system for the e-CNY, China's central bank digital currency. It also released an e-CNY smart contract service framework. Smart contracts are payments that execute automatically once preset conditions are met, so building them into the central bank's own currency would let a payment settle without either party triggering it by hand. The launch at the Financial Street Forum came three months after the launch of a cross-border e-CNY settlement platform.

Read at Caixin ↗