News Summaries

Chinese Researchers Turn Chip Heat‑Blocking ‘Islands’ into Smooth Superhighways for Cooling

For years, chip makers have struggled with a hidden enemy: the tiny “islands” that form when aluminum nitride (AlN) is used as a bonding layer between semiconductor materials. These uneven islands act like roadblocks, trapping heat inside the chip and causing performance loss or even burnout. A team led by Professor Zhou Hong at Xi’an University of Electronic Science and Technology has now cracked the problem. By applying a technique called ion‑implantation‑induced nucleation, they force the AlN to grow as a uniform, single‑crystal thin film instead of a rough polycrystalline mass. The result is a perfectly flat atomic lattice that lets heat flow three times faster than the old island‑style structure, slashing interfacial thermal resistance to just one‑third of its former value. Tests on third‑generation gallium‑nitride and fourth‑generation gallium‑oxide devices show dramatically improved cooling, paving the way for more powerful, reliable chips in everything from smartphones to defense systems. The breakthrough, published in Nature Communications and Science Advances, also offers a scalable, China‑origin platform that other manufacturers can adopt to overcome heat‑dissipation bottlenecks across the semiconductor industry.

Can Math Make AI Agents Trustworthy? A Startup Says Yes

A new startup called Harmonic, founded by Robinhood CEO Vlad Tenev and Stanford mathematician Tudor Achim, claims it has found a way to make artificial‑intelligence agents more reliable. Their secret weapon is a formal‑methods language called Lean, which lets developers write AI outputs as mathematical proofs that can be automatically checked. Harmonic’s flagship product, Aristotle, uses this approach to verify the code it generates, aiming for what the founders call “mathematical superintelligence.” The idea sounds promising, especially as the AI world has been wrestling with “hallucinations” – bogus or misleading answers that large language models sometimes produce. At a recent Davos summit, Google’s AI chief Demis Hassabis highlighted progress in cutting down these errors, and many companies are racing to build smarter, self‑checking agents. Harmonic’s early tests show its system topping reliability benchmarks, but critics point out that the technology is still narrow, focused mainly on coding tasks. Still, if formal verification can be scaled beyond math and code, it could give users a way to trust AI decisions in everyday applications. The debate is now whether this mathematical safety net will become a standard part of the AI toolbox or remain a niche experiment.

China Launches Second Phase of 6G Trials, Eyes 2030 Commercial Roll‑out

China’s Ministry of Industry and Information Technology has moved into the second stage of its 6G experiments, promising a faster, smarter network that could be on the market by 2030 and widely deployed by 2035. The ministry says it will juggle three pillars – building the infrastructure, putting it to use, and deep‑dive research – with a focus on “upgrading, iteration and deepening.” In practice, this means speeding up technology upgrades, laying out a forward‑looking industrial ecosystem, and pushing breakthroughs in next‑generation optics and quantum information. According to the China Academy of Information and Communications Technology, the country already has clear vision goals, early technical wins, and a full‑speed standards effort under way. Officials estimate the 6G market could be worth trillions of yuan once commercial services launch. The next five‑year plan highlights four key tasks: tighten technology and standards work, expand trial deployments and product development, boost international cooperation, and fuse 6G with artificial intelligence. By leveraging China’s massive market and rich AI‑driven use cases, the government hopes to spark new mobile‑intelligent services and create fresh engines of economic growth.

Yann LeCun’s New AI Startup Aims to Build Safer, Real‑World‑Smart Machines

Yann LeCun has long warned that today’s giant language models can hallucinate facts—a risk that becomes deadly in fields like medicine. To tackle that, he’s backing AMI Labs, a San‑Francisco‑based startup that’s betting on “world models,” AI that learns by interacting with the physical world instead of just text. The company’s mission is crystal clear: develop AI that’s reliable, controllable, and safe for high‑stakes applications such as industrial process control, automation, wearables, robotics, and especially healthcare. Its CEO, former Meta AI engineer and Wit.ai founder Thomas LeBrun, says the goal is to bring these world‑model systems into hospitals and other critical settings where errors aren’t an option. LeBrun isn’t going it alone. He’s joined by longtime Meta veteran Laurent Solly, who recently stepped down as the company’s European VP. The team’s deep ties to Meta suggest the social‑media giant could become AMI Labs’ first customer, even as LeCun publicly critiques some of Meta’s strategic moves. In short, AMI Labs represents a bold, contrarian push against the dominance of large language models, aiming to prove that true intelligence starts in the world, not just in words.

China’s T800 Humanoid Robot Stuns CES 2026, Showcasing Home‑Made Power and All‑Day Stamina

At CES 2026 the Chinese startup Zhongqing unveiled its T800 humanoid robot, turning heads with a performance that feels more like a human athlete than a lab prototype. Unlike many robots that rely on off‑the‑shelf parts, the T800’s joints, sensors, software and even its solid‑state battery were developed in‑house, giving the team full control over torque and power. The result is a knee joint that can deliver a staggering 450 N·m of torque—enough to let the 1.73‑metre, 70‑kg robot run, jump, spin‑kick and even execute backflips. Range anxiety, a common complaint in the industry, was tackled with a new solid‑state battery and smarter gait algorithms, extending continuous walking time to four‑to‑five hours and allowing all‑day operation in typical use. The robot’s sleek, “hexagonal warrior” look reflects a philosophy that beauty must stem from logical design, marrying industrial efficiency with artistic flair. Experts at the show praised the T800’s balance and dynamic control, noting that Chinese teams are moving from merely functional demos to setting new experience standards. Backed by decades of national R&D programs, Zhongqing’s breakthrough signals that China is now a serious contender in the global race for advanced humanoid robots.

Chinese Robots Set Sail: How Home‑Made Machines Are Winning Global Factories

China has poured record R&D funds into industrial robots, turning them from simple “Made in China” tools into high‑tech, smart‑manufacturing workhorses. Companies such as Nanjing ESTUN Automation are leading the charge, building a worldwide network of 75 service centers after snapping up UK motion‑control brand Trio and German welding leader KUKA. ESTUN’s new international management team is targeting Europe, the Americas and Southeast Asia, riding the wave of Chinese customers in the new‑energy vehicle and lithium‑battery sectors that are expanding abroad. Beyond ESTUN, firms like Guangdong Jonty are showcasing robots that can dance, cook and even run marathons, while Renjoy Smart Technology is deploying solar‑panel cleaning and installation robots across more than 260 plants in 25 countries, boosting efficiency three‑to‑four‑fold. EFORT Intelligent Equipment’s robots have found homes in Italian Maserati, German BMW and Volkswagen factories, with over 16,000 units sold in 2024 alone. Data from China’s customs office shows the nation’s share of global industrial‑robot exports rose to second place in 2024, and shipments jumped 61.5 % in the first half of 2025. As Chinese robot technology matures and AI integration deepens, these machines are no longer just exported goods—they’re becoming the backbone of modern factories worldwide.