Skip to content

"Moore's Law Is Ending" Has Been Said for Years, but What Ended Was the Other Law ── Under Dennard scaling, halving the dimensions leaves power per area at exactly 1.0000 ── Once the voltage stops falling it becomes 4.0000 ── The count still doubles every two years, reaching 1024 in twenty, and now the power does too ── [Paper 374]

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)

Abstract

“Moore’s law is ending” has been said for a long time. This paper asks what ended──the answer is that what ended was not the law of counting but the law of power. No new mathematical theorem and no new law is claimed. Scope of this paper (scope note): No new mathematical theorem and no new law is claimed──Moore’s law, Dennard scaling, and the power-density wall are all standard. We do not build semiconductor engineering──short-channel effects, leakage current, and threshold-voltage physics are not entered. They are named and no more. We claim no date──exactly when Dennard scaling stopped is not something this paper measured. The mid-2000s is cited as the accepted figure and no more. We make no forecast──what happens next is not the subject. Implementation is not treated──multicore, dark silicon, and specialised accelerators are named and no more. We measure nothing──the power of particular products is not used. Only the consequences of the scaling relations are computed. Relation to earlier papers: Paper 363 showed a bat to have three sweet spots──there one word named several things; here one name covers two laws. Paper 346 showed half-life ceasing to be a constant for some reaction orders──there too the question was whether one statement suffices. Paper 300 showed whether two things share a root is decidable──by that test, the law of counts and the law of power have distinct roots. Paper 362 showed the bottleneck lying in dragging rather than lifting──here too, what stopped was not the thing being watched. What is added is setting out five scaling factors under Dennard scaling to show the cancellation, showing the ratio to the non-scaling case to be exactly 4, giving Moore’s factor at six points, placing an intermediate case and confirming it falls between the extremes, and putting the separator on whether the voltage falls. First, there are two laws: Moore’s speaks of counts, Dennard’s of power (Section 2). Second, this is the core of the paper. Under Dennard scaling, halving the dimensions leaves power per area at exactly 1.0000 (Section 2). Third, once the voltage stops falling, the same shrink gives 4.0000 (Section 3). Fourth, the two answers differ by exactly 4, with the shrink identical (Section 3). Fifth, the count still doubles every two years, so twenty years give 1024 ── and now the power demands 1024 too (Section 4). Sixth, the separator is whether the voltage falls, not the size of the devices (Section 5). “Moore’s law is ending” has been said for a long time──but two laws were being called by one name. Moore’s law speaks of a count and Dennard scaling of power──the first says only that devices on a given area double every two years and touches nothing about power; the second says only that shrinking leaves power per area unchanged and touches nothing about counts. They held together, so they looked like one. Under Dennard scaling the power cancels exactly──halving the dimensions gives 4.0000 times the devices and 0.2500 times the power per device, so power per area is exactly 1.0000. Not approximately: the growth in count and the fall in power per device are both k^2. Faster, more numerous, and no hotter ── that cancellation was what made progress look free. But the cancellation depends on the voltage falling──power per device becomes 1/k^2 only because the voltage becomes 1/k, and once the voltage is held, the same shrink gives 4.0000. The two answers differ by exactly 4, with only the voltage changed. A voltage falling by 1/sqrt(k) gives 2.0000, between the extremes. And the law of counts has not stopped──doubling every two years gives 1024 in twenty, and if Dennard scaling had held, power would have stayed 1.0 throughout; once it stopped, power demands 1024 as well. The devices fit but cannot all run at once. So “Moore’s law is ending” has the name wrong: what stopped is the law of power. One thing separates them──whether the voltage falls along with the shrink; not the size of the devices. The count grows equally in all three rows, so Moore’s law does not distinguish them. By the criterion of Paper 300 they have distinct roots: the count comes from geometry, the power from voltage squared ── what looked like one “progress” was two things happening to move the same way. To be said honestly──neither the date nor the mechanism (lowering the threshold makes leakage grow exponentially) was computed here. The 1.0000 and 4.0000 are values under idealised scaling with k=2. One last thing──while two laws move the same way they look like one law. Only when one stops does it become clear there were two. And then “it ended” is said in the name of the famous one, not the one that stopped. On the making of this work: The ideas and content of this work stem from the author's own considerations. Assistance from an AI (a large language model) was used for structuring, English translation, and checking the algebra. Any remaining errors or misinterpretations are solely the author's. Feedback and corrections are sincerely appreciated. ----- 「ムーアの法則が終わる」と長く言われてきた。本稿が問うのは何が終わったのかである──答えは終わったのは素子数の法則ではなく、電力の法則である。新しい定理も法則も主張しない。 本稿の射程(射程注記):新しい定理も法則も主張しない──ムーアの法則・デナード則・電力密度の壁はすべて既知である。半導体工学を作らない──短チャネル効果・漏れ電流・しきい値電圧の物理には立ち入らない。名前を挙げるにとどめる。年代を主張しない──デナード則がいつ止まったかの厳密な時期は本稿が測ったものではない。通説として二〇〇〇年代半ばを挙げるにとどめる。予測をしない──今後どうなるかは主題ではない。実装を扱わない──多コア化・ダークシリコン・専用回路といった対処は名前を挙げるにとどめる。数値を実測しない──特定の製品の消費電力は用いない。比例則の帰結だけを計算する。既刊との関係:論文363 は「芯」が三つあることを示した──そこでは一語が複数を指した。ここでは一つの名前で二つの法則が呼ばれている。論文346 は次数によって半減期が定数でなくなることを示した──そこでも「一つの言い方で足りるか」が問題だった。論文300 は同根か別根かが判定できることを示した──素子数の法則と電力の法則は、その基準で別根である。論文362 は律速が持ち上げではなく引きずりにあることを示した──ここでも、止まったのは注目されていたほうではない。加えたのは、デナード則の五つの量の倍率を並べて相殺を示したこと、電圧が下がらない場合との比が厳密に 4 になることを示したこと、ムーアの倍率を六点で出したこと、中間の場合を置いて両極の間に入ることを確かめたこと、分離子を「電圧が下がるか」に置いたことである。 第一に、二つの法則がある。ムーアの法則は素子数を、デナード則は電力を言う(第2節)。 第二に、これが本稿の芯である。デナード則の下では、寸法を半分にしても面積あたりの電力は厳密に 1.0000 倍のままである(第2節)。 第三に、電圧が下がらなくなると、同じ縮小で 4.0000 倍になる(第3節)。 第四に、二つの答は 4 倍違う。縮小の仕方は同じである(第3節)。 第五に、素子数は 2 年で 2 倍のままなので、20 年で 1024 倍になり、電力もそのまま 1024 倍を要求する(第4節)。 第六に、分離子は「電圧が下がるか」であって、素子の大きさではない(第5節)。 「ムーアの法則が終わる」と長く言われてきた──だが同じ名前で呼ばれていたのは、二つの法則である。ムーアの法則は数を言い、デナード則は電力を言う──前者は同じ面積に載る素子が 2 年で 2 倍になると言うだけで電力には触れず、後者は縮めても面積あたりの電力が変わらないと言うだけで素子数には触れない。二つが同時に成り立っていたので、一つに見えていた。デナード則の下では、電力が厳密に相殺する──寸法を 2 分の 1 にすると素子数は 4.0000 倍、素子あたりの電力は 0.2500 倍、面積あたりの電力は厳密に 1.0000 倍になる。近似ではなく、増え方と減り方が同じ k^2 だからである。速くなり、多くなり、それでいて熱くならない──この相殺が「無料の進歩」の正体だった。ところが相殺は、電圧が下がることに依存している──素子あたりの電力が 1/k^2 になるのは電圧が 1/k になるからで、電圧が下がらなくなると同じ縮小で面積あたりの電力は 4.0000 倍になる。二つの答は厳密に 4 倍違い、変わったのは電圧が下がるかどうかだけである。電圧が 1/sqrt(k) だけ下がる中間なら 2.0000 倍で、1 と 4 の間に入る。そして素子数の法則は止まっていない──2 年で 2 倍のままなら 20 年で 1024 倍になり、デナードが続いていれば電力は 20 年ずっと 1.0 のままだが、止まった後は電力も 1024 倍を要求する。素子は載るが、同時には動かせない。だから「ムーアの法則が終わる」は名前を取り違えていて、止まったのは電力の法則である。分けているものは一つ──縮小と一緒に電圧が下がるかどうかであって、素子の大きさではない。三行とも素子数は同じだけ増えるので、ムーアの法則は三行を区別しない。論文300 の基準で別根であって、素子数は幾何から、電力は電圧の二乗から出る──一つの「進歩」に見えていたのは、二つが偶然同じ向きに進んでいたからである。正直に言えば──止まった年代も、電圧が下がらなくなった機構(しきい値を下げると漏れ電流が指数的に増える)も、本稿が計算したものではない。1.0000 も 4.0000 も k=2 という理想的な比例則の上での値である。最後に一つ──二つの法則が同じ方向に進んでいるあいだ、それは一つの法則に見える。片方が止まって初めて、二つだったことが分かる。そして止まったほうではなく、有名なほうの名前で「終わった」と言われる。 作成にあたって:本稿の着想と内容は、著者自身の考察に基づくものです。文章の構成整理や英訳、数式の確認には AI(大規模言語モデル)の助力を得ました。最終的な内容の解釈や誤りがあれば、それらはすべて著者の責に帰します。お気づきの点があれば、ご教示いただければ幸いです。

View source

Similar papers

#computer vision Open access Jun 2016

Software Development in Startup Companies: The Greenfield Startup Model

The results are packaged in the Greenfield Startup Model (GSM), which explains the priority of startups to release the product as quickly as possible, and the need to shorten time-to-market, by speeding up the development through low-precision engineering activities.

Carmine Giardino, Nicolò Paternoster, M. Unterkalmsteiner et al. · 178 citations · ⚡14
#computer vision Open access Oct 2016

Software Startups - A Research Agenda

Software startup companies develop innovative, software-intensive products within limited timeframes and with few resources, searching for sustainable and scalable business models.

M. Unterkalmsteiner, P. Abrahamsson, Xiaofeng Wang et al. · 157 citations · ⚡17
#machine learning Review Open access Oct 2016

“Failures” to be celebrated: an analysis of major pivots of software startups

This study conducts a case survey study based on the secondary data of the major pivots happened in 49 software startups, and demonstrates that customer need pivot is the most common among all pivot types.

Sohaib Shahid Bajwa, Xiaofeng Wang, Anh Nguyen-Duc et al. · 127 citations · ⚡15
#computer vision Review Open access May 2015

A survey study on major technical barriers affecting the decision to adopt cloud services

The comparison of adopter and non-adopter sample reveals three potential adoption inhibitor, security, data privacy, and portability, which underlines the importance of the technical and security perspectives for research investigating the adoption of technology.

Nattakarn Phaphoom, Xiaofeng Wang, S. Samuel et al. · 111 citations · ⚡8
#computer vision Conference Open access Dec 2013

Affordable and Energy-Efficient Cloud Computing Clusters: The Bolzano Raspberry Pi Cloud Cluster Experiment

The ongoing work building a Raspberry Pi cluster consisting of 300 nodes is presented, with potential use cases being an inexpensive and green test bed for cloud computing research and a robust and mobile data center for operating in adverse environments.

P. Abrahamsson, S. Helmer, Nattakarn Phaphoom et al. · 110 citations · ⚡7
#computer vision Book Open access Mar 2017

On the Unhappiness of Software Developers

The results indicate that software developers are a slightly happy population, but the need for limiting the unhappiness of developers remains, and 219 factors representing causes of unhappiness while developing software are identified.

D. Graziotin, Fabian Fagerholm, Xiaofeng Wang et al. · 84 citations · ⚡6

Related blog posts

MIT News · Artificial Intelligence Sep 14, 2026

New method enables AI for safety-critical situations

The “HardFlow” algorithm could help generative AI models produce high-quality outputs that obey strict requirements when “pretty close” doesn’t cut it.

GPT-Lab Sep 10, 2026

Responsible AI Must Consider Its Afterlife

AI may appear weightless, but every model depends on physical infrastructure. To understand responsible AI, we need to look beyond algorithms and consider the entire lifecycle of the hardware behind them. The post Responsible AI Must Consider Its Afterlife appeared first on GPT-Lab.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.