At a semiconductor symposium in Shanghai today, Huawei unveiled the Tau Scaling Law, a new chip design principle that focuses on data movement speed rather than transistor miniaturization. The accompanying architecture, called LogicFolding, reorganizes internal chip wiring to shorten signal paths. The first chip to adopt this architecture will be the Kirin series, expected by the end of 2025.
The Real Context
The US restricts Huawei's access to EUV lithography machines, preventing the company from following the standard "X nm → X-2 nm" node progression. A comparison of current stable mass production nodes shows the gap:
- TSMC: 2 nm (mass produced, 1.4 nm planned for 2028)
- Samsung: 2 nm
- Intel: 18A (equivalent to ~1.8 nm)
- Huawei/SMIC: 7 nm
Huawei is roughly three generations behind. Instead of trying to catch up conventionally, He Tingbo, president of Huawei's semiconductor business and head of the secret chip project, set a goal: achieve "equivalent 1.4 nm transistor density" within five years. The key word is "equivalent" — not actually using EUV to etch 1.4 nm features, but achieving similar performance density through other means.
What Is LogicFolding?
Omdia analyst He Hui summarized:
"Rather than depending solely on smaller transistors, the company is focusing on shortening interconnect, lowering latency and improving data movement inside the chip."This approach is not new in academia — 3D stacking, chiplet packaging, and near-memory computing explore similar ideas. But Huawei elevates it from point optimizations to a "law" — a paradigm for all SoC designs over the next five years. He Tingbo also revealed that over the past six years, Huawei has designed and mass-produced 381 chips using these principles, meaning LogicFolding is already a proven methodology, now being systematized and pushed across the ecosystem.
Kirin as the First Test
The Kirin series launching late this year will be the first product to carry the LogicFolding label. Mobile SoCs are the most constrained by the power, area, and performance triangle, making them ideal for validating the new architecture. If LogicFolding can shorten signal paths and reduce power consumption, performance could jump even on the same 7 nm process. If successful, the Ascend AI training/inference chip line could follow, broadening the impact.
Link to DeepSeek
Last week, DeepSeek permanently cut V4-Pro API prices by 75%, enabled by the V4 series running on Ascend 950 from the design stage, bypassing NVIDIA. If Tau Scaling Law achieves 1.4 nm equivalent density in five years, the next-generation Ascend could match NVIDIA Blackwell successors in compute density. That would allow domestic large models like DeepSeek, Qwen, and Zhipu to scale further and cut costs, alleviating the "hardware gap" anxiety that has plagued China's AI industry. NVIDIA CEO Jensen Huang recently acknowledged: "We've really largely conceded that market to them." Two years ago, such a statement would have been unthinkable.
Challenges Ahead
LogicFolding faces three major hurdles:
- EDA tools. Designing chips with non-traditional geometry may not be supported by existing Cadence/Synopsys tools. Huawei may need to develop its own or collaborate with domestic EDA firms — a challenge the EDA industry has discussed for years.
- Thermal management. Tighter wiring increases heat density. In high-load data centers, effective cooling is critical; without it, 1.4 nm equivalent density remains a slideware concept.
- Mass production economics. Huawei has not disclosed wafer cost, yield, or ramp-up curves for the equivalent density. A working demo is far from large-scale shipment.
Huawei has presented a five-year roadmap. But the roadmap itself signals that Chinese semiconductors have found a path that does not require EUV, and are pushing it downstream. The real test will be the Kirin launch later this year — no matter how elegant the architecture, ultimately the chip must deliver.
Sources: CocoLoop, China's Huawei reveals chip design breakthrough amid US sanctions (Rappler); Huawei reveals chip design breakthrough amid US sanctions (Free Malaysia Today); Huawei Unveils Major Chip Design Breakthrough as China Pushes Past US Sanctions (Modern Diplomacy); Nvidia says it has 'largely conceded' China's AI chip market to Huawei (CNBC)