[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-huawei-tau-law-v2":3,"news-related-d61acb60-34a9-44d0-b72e-fd3fac6e7697":33},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":20,"news_slug":26,"published_at":27,"created_at":28,"modified_at":29,"is_published":30,"publish_type":31,"image_url":13,"view_count":32},"d61acb60-34a9-44d0-b72e-fd3fac6e7697","华为韬(τ)定律 6 周内 V1 → V2:从 keynote 叙事升级到 ChinaXiv 同行评议","2026 年 7 月 3 日,中国科学院科技论文预发布平台(ChinaXiv)公示了一份署名何庭波的论文:《面向多层级电子系统的时间缩微理论》V2 版本——这正是 5 月 25 日 ICAS(上海)高调发布的\"韬(τ)定律\"的工程实现版(论文简称\"Teresa's Law\")。\n\nV1 是 keynote,只有理论原则与产品 roadmap;V2 是论文全文,**第一次把\"做出来\"换成了\"跑出来\"**——补充了大量工程落地细节、实测量化数据与 Kirin 2026 量产实测参数。这条 5\u002F25 → 7\u002F3 的六周时间线,正是华为把\"韬定律\"从 PR 叙事推到可被同行评议状态的关键一步。\n\n**韬(τ)定律核心一句话**:用\"信号传播时间常数 τ 的缩减\"替代\"晶体管几何尺寸的缩减\",作为半导体继续提升性能的新原则。传统摩尔定律靠 EUV 把晶体管做小,这条路被卡(对华为而言是缺光刻机);τ 缩微靠**架构创新**让信号跑得更快,而不是靠工艺让晶体管更小。实践路径是 LogicFolding——把 2D 平面电路折叠为 3D 垂直堆叠,缩短 critical path 的物理长度。\n\n**V1 给的是命题,V2 填上三件事**:一是 LogicFolding 物理实现的深度阐释,二是 Kirin 2026 量产实测参数(决定 V1 的\"等效 1.4nm\"是不是 PR 修辞),三是明确的产品演进路线图。论文还首次披露 2020-2026 年间,基于 τ 原理已量产 381 款芯片的事实。商业可信度层面,同篇新闻引用 HiSilicon 每年约 ¥100 亿 \u002F $13.77 亿美元的 τ 路线 R&D 投入——量级已超过大多数国际顶尖芯片实验室。\n\n**市场反应保持冷感**。从 5 月 25 日至今,相关概念股未现持续单边行情,华尔街仍未把韬定律计入估值溢价;同期三星、TSMC 也在做 3D stacking 但目标不在替代 EUV;Counterpoint Research 维持\"成本、功耗、热管理、系统集成\"四大工程挑战的保留意见。换句话说:**同行在走相似路,但华为赌的是\"无 EUV 仍可达 1.4nm 等效密度\"的更激进版本**。\n\n**对今天的意义**:V2 之后,半导体圈接下来几个月要做的事很具体——验证 2026 Fall Kirin 芯片的实测 PPA(性能\u002F功耗\u002F面积)与良率,看能否对上 V2 论文给出的数字。对得上,韬定律从\"理论野心\"升级为\"工程事实\",华为在中国 AI 芯片市场对 Nvidia 的挤压将从品牌层面下沉到价格-性能层面;对不上,韬定律 V3 必然需要再补大量现象级工程数据,否则 2027-2028 这两年的产业话语权仍会留在传统的\"几何缩微\"派。","https:\u002F\u002Fwww.huawei.com\u002Fcn\u002Fnews\u002F2026\u002F5\u002Fieee-iscas-tau-scaling","da74007a-2a9c-432d-9d3b-95dc9b782378",[10,14,17],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":18,"name":19,"slug":19,"description":13,"color":13},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",[21],{"id":22,"lang":23,"title":24,"summary":25,"content":13},"0e70bebe-85d1-44bd-8c71-ebd7fbcd4cff","en","Huawei's Tao-law V2 lands peer review within six weeks","On July 3, 2026, the ChinaXiv preprint platform of the Chinese Academy of Sciences published a paper by He Tingbo: \"Time-Scaling Theory for Multi-Level Electronic Systems\" V2 — this is exactly the engineering implementation version (the paper is \"Teresa's Law\") of the \"Tau (τ) Law\" that was high-profile released at ICAS (Shanghai) on May 25. V1 was a keynote, with only theoretical principles and a product roadmap; V2 is the full paper, **for the first time swapping \"make it\" for \"run it\"** — supplementing extensive engineering details, real measurement data, and Kirin 2026 mass-production measurement parameters. This 5\u002F25 → 7\u002F3 six-week timeline is exactly Huawei's key step in pushing the \"Tau Law\" from PR narrative to a peer-reviewable state. **Tau (τ) Law in one sentence**: use the \"shrinking of signal propagation time constant τ\" to replace \"shrinking of transistor geometry\" as the new principle for continued semiconductor performance improvement. Traditional Moore's Law relies on EUV to shrink transistors, but this path is blocked (for Huawei, no lithography machine); τ-scaling relies on **architectural innovation** to make signals travel faster, rather than on process to make transistors smaller. The practical path is LogicFolding — folding 2D planar circuits into 3D vertical stacks, shortening the physical length of the critical path. **V1 gives the proposition, V2 fills in three things**: first, an in-depth explanation of the physical implementation of LogicFolding; second, mass-production measurement parameters of Kirin 2026 (determining whether V1's \"equivalent 1.4nm\" is PR rhetoric); third, an explicit product evolution roadmap. The paper also discloses for the first time that between 2020-2026, 381 chips have been mass-produced based on the τ principle. At the commercial credibility level, the same news quotes HiSilicon's annual τ-route R&D investment of about ¥10 billion \u002F $1.377 billion — the order of magnitude already exceeds most international top chip labs. **Market reaction remains cool**. From May 25 to now, related concept stocks have not shown a sustained unilateral trend, Wall Street has not yet priced the Tau Law into valuation premium; at the same time, Samsung and TSMC are also doing 3D stacking but the goal isn't to replace EUV; Counterpoint Research maintains reservations on \"cost, power, thermal management, system integration\" four engineering challenges. In other words: **peers are walking a similar path, but Huawei is betting on a more aggressive version of \"no EUV still reaches 1.4nm equivalent density\"**. **Today's significance**: After V2, what the semiconductor community needs to do in the next few months is very specific — verify that the measured PPA (performance\u002Fpower\u002Farea) and yield of the 2026 Fall Kirin chip can match the numbers in the V2 paper. If they match, the Tau Law upgrades from \"theoretical ambition\" to \"engineering fact\", and Huawei's squeeze on Nvidia in the Chinese AI chip market will sink from the brand level to the price-performance level; if they don't match, Tau Law V3 will inevitably need to supplement a lot of phenomenal engineering data, otherwise the industry discourse power in 2027-2028 will remain with the traditional \"geometric scaling\" camp.","huawei-tau-law-v2","2026-07-05T15:55:00Z","2026-07-05T15:58:07.467277Z","2026-08-19T02:08:40.142862Z",true,"agent",141,{"items":34},[35,40,45,50,55,60],{"id":36,"title":37,"news_slug":38,"published_at":39},"fb1cbe25-8b85-41ec-b619-9a27b405ec34","AMD 收购 Taalas:把 AI 模型权重「刻进硅片」的推理新打法","amd-acquires-taalas-inference-chip","2026-08-19T01:00:00+00:00",{"id":41,"title":42,"news_slug":43,"published_at":44},"ae924988-53ef-4f46-b1c2-c43c4e9866fe","欧盟掏 100 亿欧元建 7 座 AI 超级工厂：每座堆 10 万颗顶尖芯片，瞄准美国算力代差","eu-ai-gigafactory-7-factories-100000-chips","2026-07-31T04:30:00+00:00",{"id":46,"title":47,"news_slug":48,"published_at":49},"5b5cd125-c167-4472-8dc5-2414699b73ce","壁仞科技 NPO 1024卡超节点:把国产 GPU 集群从'集中'拆成'解耦'","biren-npo-1024-supernode","2026-07-19T02:30:00+00:00",{"id":51,"title":52,"news_slug":53,"published_at":54},"66d66fa2-364e-4fa2-a9b2-e69e6f86dc8c","Vera Rubin 平台登陆 ISC 2026：144 张 GPU + 100% 液冷，把 TOP500 算力压进科研机柜","nvidia-vera-rubin-isc2026-144-gpu","2026-06-22T20:00:00+00:00",{"id":56,"title":57,"news_slug":58,"published_at":59},"774de6ac-98e1-4343-a67a-bfdc72d377bb","INFORMS 实证:AI 广告真实投放胜过设计师,18 个月后仍领先","informs-ai-ads-beat-human-designers-18-months","2026-08-22T14:00:00+00:00",{"id":61,"title":62,"news_slug":63,"published_at":64},"43eda321-b0b7-4df7-b20e-9758cbab42c9","记忆越完整,眼前题越做不对:MemTrapBench 把 LLM 长期记忆框架打回原形","memtrapbench-llm-memory-cognitive-traps","2026-08-22T04:00:00+00:00"]