For sixty years the chip business has chased one idea, which is that smaller is better. Huawei has now put forward a different one, and it arrives with a name, a roadmap, and an assertion that the company has quietly been building silicon on the principle for six years.
The setting was the IEEE International Symposium on Circuits and Systems in Shanghai. A keynote from He Tingbo, a Huawei board member and president of the company's semiconductor division HiSilicon, introduced what the company calls the Tau Scaling Law, pitched as a stand-in for Moore's Law. Tom's Hardware reports the framework is aimed at closing the technology gap with global semiconductor leaders such as TSMC and Nvidia.
Here is the plain-language version. Moore's Law is geometric, so you shrink the transistor, fit more of them in, and go faster. That road runs through extreme ultraviolet lithography, or EUV, which people say letter by letter, and which is the machinery that prints the finest features onto silicon. It is also precisely the machinery that US trade restrictions keep out of Chinese fabs. Tau scaling swaps the target from size to time, asking not how small a component can be made but how quickly data can cross the system.
Turning that theory into product required an architecture, and the one Huawei engineered is called LogicFolding. Picture a cramped flat where the tenant, rather than knocking through a wall, builds a mezzanine. Logic circuits fold and stack into a dual-layer arrangement, the internal wiring gets shorter, and the delay a signal picks up along the way shrinks with it.
The numbers attached to that design belong to Huawei, and no independent benchmark arrived alongside them. The company puts the gain at a 55 percent increase in transistor density and a 41 percent improvement in power efficiency. It also says 381 chips have already been designed and mass-produced on the methodology, which if borne out makes this less a laboratory curiosity than a production line nobody outside the company had heard about.
Customers meet the thing this autumn, when flagship Kirin smartphone processors become the first commercial silicon to carry the architecture. Huawei says it then intends to scale LogicFolding into its Ascend AI processors and into high-capacity data centre clusters by 2030, which would hand Chinese operators a domestic option where restricted Nvidia hardware currently sits. By 2031 the company projects it could design parts with transistor density equivalent to a 1.4 nanometre process.
That last figure deserves holding at arm's length. Density equivalent to a process is not the same thing as the process, and TSMC expects to be mass-producing genuine 1.4 nanometre chips by 2028. The question worth watching is not whether the label matches but whether folding logic upward buys enough performance for the workloads China cannot currently serve at home.
Markets did not wait for benchmarks. Shares in SMIC, the largest contract chipmaker in China, rose 7.6 percent once the announcement landed.





Leave a Reply