超越任何单一产品的更深层主张是方法论层面的。τ缩微是自登纳德以来第一个为整个堆栈提供共享优化目标的缩微原则。它向工艺技术人员、电路设计师、架构师、系统工程师和软件团队发出信号:这些群体现在正在以相同的单位优化相同的量,任何单层的改进必须传导至系统τ才算有效。它也向行业战略家和资本配置者表明,下一笔投资应跟随τ而非节点——竞争性的性能不再要求常驻在光刻技术的最前沿,而封装、存储带宽和互连架构设计现在承载着此前仅由前沿逻辑节点所拥有的战略权重。对于在成长过程中将“摩尔定律”等同于“进步”的一代工程师而言,这是一个困难的转变。几何时代事实上已经结束;否认这一事实不是可行的策略。通过微缩实现加速的时代正在让位于通过多层电子系统的τ优化实现加速的时代——而在未来六到十年中以τ为首要目标的公司、研究团体和生态系统,将决定此后十年计算的面貌。未来十年的工作范围已经划定。许多开放问题仍然存在,没有任何单一组织可以独自解决——工具链、标准、基准、器件物理和经济模型都需要超越任何单一公司的贡献。因此,本文既是一份来自前线的报告,也是一份邀请。前方的路线图要求苛刻,但方向是明确的。作者何庭波领导华为半导体业务。她所带领的团队在2020年至2026年间设计并量产了381颗芯片,覆盖移动、AI、汽车和基础设施市场,并且是τ缩微方法论以及本文所述LogicFolding、Unified Bus和Hi-ONE技术的来源。致谢本文汲取了华为半导体及其晶圆代工、设备、EDA和系统合作伙伴生态系统中数千名工程师六年工作的成果。作者感谢那些以耐心使这项工作成为可能的客户。参考文献1. Moore G E. Cramming more components onto integrated circuits. Electronics, 1965, 38: 114-117.2. Dennard R H, Gaensslen F H, Yu H N, et al. Design of ion-implanted MOSFET's with very small physical dimensions. IEEE Journal of Solid-State Circuits, 1974, 9: 256-268.3. Wong H S P. Beyond the conventional transistor. IBM J Res Dev, 2002, 46: 133-168.4. Zhirnov V V, Cavin R K, Hutchby J A, et al. Limits to binary logic switch scaling - a gedanken model. Proceedings of the IEEE, 2003, 91: 1934-1939.5. Moore G E. Lithography and the future of Moore's law. Proc SPIE, 1995, 2437: 2-17.6. Taur Y, Ning T H. Fundamentals of Modern VLSI Devices. 3rd ed. Cambridge: Cambridge University Press, 2021.7. IEEE. International roadmap for devices and systems (IRDS) 2023/2024: Interconnect and More-than-Moore chapters[R]. Piscataway: IEEE, 2024.8. Flamm K. Measuring Moore's law: evidence from price, cost, and quality indexes[R]. Cambridge: National Bureau of Economic Research, 2018.9. Soulié J P, Sankaran K, Troeye B V, et al. Selecting alternative metals for advanced interconnects. J Appl Phys, 2024, 136: 171101.10. Batude P, Ernst T, Arcamone J, et al. 3D sequential integration: a key enabling technology for heterogeneous co-integration of new functions with CMOS. IEEE J Emerg Sel Top Circuits Syst, 2012, 2: 714-722.11. Hennessy J L, Patterson D A. A new golden age for computer architecture. Commun ACM, 2019, 62: 48-60.12. Zuo P, Lin H, Deng J, et al. Serving large language models on Huawei CloudMatrix384. arXiv preprint, 2025, arXiv:2506.12708.13. Loh G H, Xie Y, Black B. Processor design in 3-D die-stacking technologies. IEEE Micro, 2007, 27: 31-48.14. Chen M F, Chen F C, Chiou W C, et al. System on integrated chips (SoIC™) for 3D heterogeneous integration. In: Proceedings of 2019 IEEE 69th Electronic Components and Technology Conference (ECTC), 2019.15. Chen Y M, Ko T, Ting K C, et al. Next generation TSMC-SoIC® platform for ultra-high bandwidth HPC application. In: Proceedings of 2024 IEEE International Electron Devices Meeting (IEDM), 2024.16. Ingerly D B, Amin S, Aryasomayajula L, et al. Foveros: 3D integration and the use of face-to-face chip stacking for logic devices. In: Proceedings of 2019 IEEE International Electron Devices Meeting (IEDM), 2019.17. Gomes W, Morgan S, Phelps B, et al. Meteor Lake and Arrow Lake Intel next-gen 3D client architecture platform with Foveros. In: Proceedings of 2022 IEEE Hot Chips 34 Symposium (HCS), 2022.18. Wuu J, Agarwal R, Ciraula M, et al. 3D V-Cache: the implementation of a hybrid-bonded 64MB stacked cache for a 7nm x86-64 CPU. In: Proceedings of 2022 IEEE International Solid-State Circuits Conference (ISSCC), 2022: 428-429.19. Vinet M, Batude P, Fenouillet-Béranger C, et al. Opportunities brought by sequential 3D CoolCube™ integration. In: Proceedings of 2016 European Solid-State Device Research Conference (ESSDERC), 2016: 226-229.20. Santos C, Vivet P, Thuries S, et al. Thermal performance of CoolCube™ monolithic and TSV-based 3D integration processes. In: Proceedings of 2016 IEEE 3D Systems Integration Conference (3DIC), 2016.21. Mallik A, Vandooren A, Witters L, et al. The impact of Sequential-3D integration on semiconductor scaling roadmap. In: Proceedings of 2017 IEEE International Electron Devices Meeting (IEDM), 2017.22. Vandooren A, Witters L, Franco J, et al. Sequential 3D: Key integration challenges and opportunities for advanced semiconductor scaling. In: Proceedings of 2018 IEEE International Conference on IC Design & Technology (ICICDT), 2018.23. Jiang Z, Lin H, Zhong Y, et al. MegaScale: Design and deployment of a 10,000-GPU system for LLM training. In: Proceedings of 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI), 2024.24. Narayanan D, Shoeybi M, Casper J, et al. Efficient large-scale language model training on GPU clusters using Megatron-LM. In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC), 2021: 1-15.25. Horowitz M. Computing's energy problem (and what we can do about it). In: Proceedings of 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC), 2014: 10-14.26. Gholami A, Yao Z, Kim S, et al. AI and memory wall. IEEE Micro, 2024, 44: 33-39.27. Aminabadi R Y, Rajbhandari S, Zhang M, et al. DeepSpeed-inference: enabling efficient inference of transformer models at unprecedented scale. In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC), 2022.28. Liu Y, Jiang L, Li D, et al. ByteDance Jakiro: Enabling RDMA and TCP over Virtual Private Cloud. In: Proceedings of the ACM SIGCOMM 2025 Conference, 2025: 114-128.29. Firestone D, Putnam A, Mundkur S, et al. Azure accelerated networking: SmartNICs in the public cloud. In: Proceedings of 15th USENIX Symposium on Networked Systems Design and Implementation (NSDI), 2018: 51-64.30. Liao H, Liu B Y, Chen X P, et al. UB-Mesh: a hierarchically localized nD-FullMesh datacenter network architecture. IEEE Micro, 2025, 45: 20-29.31. Yuan Y, Peng Y, Cheung S, et al. The perspective of all-silicon photonics and systems. APL Photonics, 2025, 10: 060901.32. Shekhar S, Bogaerts W, Chrostowski L, et al. Roadmapping the next generation of silicon photonics. Nat Commun, 2024, 15: 751.