Overview
On May 25, 2026, at the IEEE International Symposium on Circuits and Systems (ISCAS 2026) in Shanghai, He Tingbo—Huawei's director and president of its semiconductor business, famous for the 2019 "backup plans all become mainstream" letter—presented the Tao (τ) Law, a scaling framework the company positions as the first shared optimization principle across the entire computing stack since Dennard scaling. The post is a detailed Chinese analysis of the announcement; below is a structured English summary of its key claims.
Key points
Why a new law?
- Moore's Law's economic rationale is breaking down: at 3nm, fab investment exceeds $20B and per-chip design costs top $1B, with per-transistor costs rising. Quantum tunneling and leakage erode yields.
- The paper's framing: the core question is no longer "how small can transistors get?" but "to what extent, and toward what target?" Huawei's answer: time.
- τ is the RC time constant—how fast a circuit responds. The formula is textbook physics; Huawei's contribution is elevating it to an industry-wide optimization target.
- Instead of only shrinking transistors (geometric scaling), shorten signal paths (time scaling).
- Logic folding is front-end design: logic within a module is redistributed across two or more vertically stacked active layers, connected by ultra-short TSVs with 1.5μm pitch—unlike CoWoS/Foveros/HBM, which bond separately fabricated dies ("pseudo-3D").
- At a fixed node, Huawei reports 55% transistor density gain and 41% efficiency gain—equivalent to two full process-node generations (~3 years) under traditional scaling.
- Unified Bus (灵衢): unified protocol replacing layered stacks; latency from tens of microseconds to ~100ns (~500× reduction).
- Hi-ONE optical interconnect: 8Tb/s per module, reach extended from <1m to 100m.
- Atlas 960 SuperPod: 15,488 Ascend cards as one supernode.
- Claim: in AI systems, τ can shrink to 10% of the prior year's value annually.
- May 2020–May 2026: 381 chips designed and mass-produced under these principles, serving 1B+ users (Kirin, Ascend, Kunpeng, automotive, IoT).
- Kirin 9000S (Mate 60 Pro, SMIC N+2 ≈7nm) → Kirin 2026 (dual-layer logic folding, 238MTr/mm², 3.1GHz, ≈Intel 18A class); roadmap to 4GHz by 2029.
- Ascend 910B/910C (≈80% of H100 FP16 performance); logic folding enters the Ascend line with the 990; 100× AI hardware integration growth projected by 2035.
- Goal: density equivalent to a 1.4nm-class node by 2031 (155 → 400+ MTr/mm²), achieved via folding layers plus device/system optimization—not an actual 1.4nm process.
- Open problems the paper itself lists: physical limits on R and C, thermal management, crosstalk, hybrid-bonding yield (bond pitch 1.5μm, overlay <0.5μm), and an incomplete EDA toolchain.
- Huawei cannot access EUV lithography; Tao Law is the product of forced detour—but the whole industry (TSMC SoIC, Intel Foveros, backside power delivery) is pursuing post-Moore paths. As He put it: "Huawei hit this wall earlier than its peers."
- Advanced packaging (Yole: $46B in 2024 → $79.4B by 2030, ~23% CAGR for 2.5D/3D): beneficiaries include JCET, Tongfu Microelectronics, Hua Tian, and SJ Semiconductor.
- EDA: Peking University has already announced a "true-3D" EDA prototype targeting logic folding; domestic EDA vendors see a historic window.
- Optics/photonics: demand for silicon photonic devices grows with AI cluster scale.
- He's closing message is an open industry invitation: "Toolchains, standards, benchmarks, device physics, and economic models all need contributions beyond any single company."
The τ = R × C framework
Logic folding vs. 3D stacking
A four-layer optimization stack
1. Device: high-κ metal gates, GAA, strain engineering; parasitic R/C now dominate, so interconnect optimization is urgent. 2. Circuit: logic folding shortens critical-path wiring by 50–80%. 3. Chip: full software-hardware co-design around τ. 4. System:Six years of execution: 381 chips
The 2031 target and its caveats
Context: necessity as well as vision
Industry implications
Conclusion
Whether Tao Law becomes a true "law" depends on decades of validation, as Moore's Law had. But the post's verdict is notable: in the sixth year of sanctions, Huawei distilled its engineering practice into a systematic, publicly presented theoretical framework—the first such industry-guiding principle proposed globally by a Chinese company. The Kirin 2026's real-world performance this autumn will be the first key test.