【Huawei’s Tau (τ) Law: Another DeepSeek Moment for China Semis 伯恩斯坦分析师 Qingyuan Lin: 华为韬(τ)定律:中国半导体的又一个 DeepSeek时刻
Date: May 25, 2026
日期: 2026年5月25日】
We view Huawei’s Tau (τ) Scaling Law as another DeepSeek moment for China semis.
我们将华为韬(τ)缩放定律视为中国半导体的又一个DeepSeek时刻。
Against a backdrop of EUV lithography restrictions, the τ-law framework could emerge as a viable new roadmap for China’s semiconductor industry to continue scaling chip performance without EUV.
在极紫外(EUV)光刻机受限的背景下,韬定律框架有望成为中国半导体产业不依赖EUV、持续提升芯片性能的可行新路线图。
The core insight is not a single technology, but a new performance-scaling logic: shifting from geometric scaling (Moore’s Law) to time scaling (τ scaling) by systematically reducing signal delay across device, circuit, chip, and system levels.
其核心洞见并非单一技术,而是一种全新的性能提升逻辑:从几何缩放(摩尔定律)转向时间缩放(τ缩放),在器件、电路、芯片、系统四层系统性降低信号延迟。
Huawei’s LogicFolding—a 3D stacking approach—shortens critical signal paths, raising transistor density and efficiency on mature nodes.
华为逻辑折叠(LogicFolding)是一种3D堆叠技术,缩短关键信号路径,在成熟制程上提升晶体管密度与能效。
We see this as a sustainable, extensible playbook that can narrow the gap with global leaders over time.
我们认为这是一套可持续、可扩展的方法论,有望逐步缩小与全球领先者的差距。
Similar to DeepSeek’s algorithm-driven AI breakthrough amid GPU restrictions, Tau Law represents a constraint-driven architectural breakthrough for Chinese semis.
与DeepSeek在GPU受限下凭借算法实现AI突破类似,韬定律代表中国半导体在约束条件下的架构级突破。
关键预测与影响
❶No EUV required: Performance gains via delay reduction, not geometry shrinkage.
无需EUV: 性能提升来自延迟压缩,而非尺寸缩小。
❷ Four-layer co-optimization: Device → circuit → chip → system.
四层协同优化: 器件→电路→芯片→系统。
❸ LogicFolding metrics: Density +53.5%, efficiency +41% (Huawei data).
逻辑折叠指标: 密度+53.5%、能效+41%(华为数据)。
❹2031 target: Equivalent 1.4nm transistor density on mature nodes.
2031目标: 成熟制程实现等效1.4nm晶体管密度。
❺Strategic impact: Reactivates China’s existing mature-node fabs (hundreds of billions USD invested).
战略影响: 盘活中国现有成熟制程产线(数千亿美元投资)。
中美AI战略 AI兴衰定律 华为速记