Huawei wants to build chips that perform like 1.4-nanometre parts by 2031 without ever fabricating a 1.4nm transistor. The company laid out the plan Monday at the 2026 IEEE International Symposium on Circuits and Systems in Shanghai, where semiconductor chief He Tingbo introduced what Huawei calls the Tau Scaling Law. The pitch is blunt: stop shrinking transistors, start shrinking the time a signal spends traveling through the chip.
What Huawei actually said
The idea, spelled out in Huawei's keynote speech, is that Moore's Law has run out of cheap room. Transistors keep getting harder and pricier to shrink. So rather than chase smaller geometry, He's team proposes optimizing the time constant, the Greek letter τ, that governs how fast data moves between components. Compress that, the argument goes, and you recover the performance and density a smaller node would have given you.
The practical piece is an architecture Huawei calls LogicFolding, which it says reorganizes circuit layouts to cut wiring distance and signal load. Kirin chips due in autumn will be the first to ship with it.
"Equivalent" is doing a lot of work here
Read the fine print and the 1.4nm figure gets slippery. Huawei is not promising to manufacture a 1.4nm chip the way TSMC or Samsung will. It is promising a transistor density equivalent to 1.4nm, reached through architecture and packaging instead of lithography. The chips stay physically bigger and just behave as if they were tiny. That distinction is the entire story, because it is how Huawei plans to route around the US export controls that cut China off from the most advanced chipmaking gear.
Whether the equivalence holds is a separate question. Huawei has published no third-party benchmarks or yield data, so the only numbers on the table are its own. One industry breakdown notes that system-level approaches, advanced packaging and 3D chiplet stacking and signal-path compression, are a credible way to squeeze frontier performance from older, legally available tools. Credible is not the same as proven.
The 381-chip flex
He said Huawei has already designed and mass-produced 381 chips on this principle over six years, across phones and AI computing. It is a large number handed over without much breakdown, and six years of "381 chips" sweeps in everything from flagship Kirins to minor variants. It tells you Huawei has been busy more than it tells you the method scales to the frontier.
There is context behind the urgency. After 2020 sanctions knocked Huawei's phone silicon off the market, the company clawed back with domestically made Kirins, and He acknowledged its mobile chips had hit a performance "saturation zone" after last year's Kirin 9030 Pro. Tau is the exit she is selling.
She closed with a call for collaboration: "No single company can independently find all the answers along the path of semiconductor evolution." A generous sentiment from a company whose entire roadmap exists because it got locked out of everyone else's tools.
The first real test lands this autumn, when the LogicFolding Kirin ships and reviewers can measure whether the architecture delivers. The 1.4nm-equivalent claim stays untestable until closer to 2031.
