其他2018arxiv.org
仅使用连杆与旋转关节的机械计算系统
Mechanical Computing Systems Using Only Links and Rotary Joints
进阶 · 约 55 分钟读完前置知识：数字逻辑基础、机械连杆与关节运动学、摩擦与能量耗散
导读
这篇论文（作者为知名计算机科学家 Ralph Merkle 等）提出一种只用刚性连杆和旋转关节（“锁”）搭建机械逻辑门与机械计算机的方案，并采用四相时钟避免力在长信号线上累积。作者的模型估算认为，关节摩擦很小，在 100 MHz 下每个旋转关节每次操作只耗散约 2.4×10⁻²⁷ J，因此这类机械计算机的功耗在理论上可比半导体低几个数量级，同时速度仍较高。
HN 关注点：焦点一是这种极低功耗估算是否可信（噪声与量子极限），二是作者 Ralph Merkle 的身份，以及机械逻辑门的物理可行性。
HN 评论精选
这几条评论里：质疑 1 补充 3
质疑质疑论文的极低功耗估算，认为未考虑噪声与量子力学限制，缺乏说服力。
计算机的一个关键性能指标是能量耗散，其中一个来源就是每个逻辑门中旋转关节的摩擦。由于关节摩擦阻力小，用它们搭建的机械计算机原则上可比传统半导体计算机少耗散几个数量级的功率，同时仍能保持较高速度。这个说法看起来可疑，也许有更专业的人能评论。论文给的理由是：操作这个锁需要关节旋转最多 Δθ≈1 rad，在 f=100 MHz 时每个旋转关节耗散约 2.4×10⁻²⁷ J。这有点像说“我们让芯片在 1 微伏、1 皮安、100 MHz 下工作，每次操作 10⁻²⁶ J”，如果不仔细分析噪声和量子力学限制，就只是空想；否则几乎任何计算设备都能以任意低的功率运行。
darkmighty查看英文原文
> A key performance metric of computers is their energy dissipation. One contribution to dissipation is friction at the rotary joints in each logic gate. Due to the joint’s small frictional drag, mechanical computers constructed from them can, in principle, dissipate orders of magnitude less power than conventional semiconductor computers, while still operating at relatively high speeds. This claim seems dubious, perhaps someone has more expertise to comment. The justification > Operating this lock involves rotation at the joints by up to ∆θ≈1 rad. The model system analyzed in [11] is an excerpt of the links and joints shown in the closeup on the right of Figure 24. From [11, Eq. 2], this rotation dissipates bout 2.4×10−27J per rotary joint when operating at f = 100 MHz. Seems akin to saying "We operate our microchip at 1 microvolt / 1 pico ampere at f=100MHz, giving 10-26J per operation." -- which seems like a silly aspiration without carefully analyzing noise and quantum mechanical constraints. (without which it would seem almost any computing device could operate at arbitrarily low power).
补充指出作者是著名计算机科学家 Ralph Merkle，这解释了关注度。
难道没别人注意到第一作者的名字吗？Ralph C. Merkle 是相当知名的计算机科学家，参与了公钥密码学和哈希算法，也是区块链核心的 Merkle 树的发明者。
fernly查看英文原文
Did no one else notice the name of the lead author? Ralph C. Merkle [1] is a rather well known computer scientist, involved in public-key cryptography, hashing algorithms, and inventor of the Merkle Tree [2] that is at the core of the blockchain. [1] https://en.wikipedia.org/wiki/Ralph_Merkle [2] https://en.wikipedia.org/wiki/Merkle_tree
补充类比 Z1 等早期机械计算机，认为机械计算仍有优化空间，并吐槽 PDF 图片臃肿。
这让我有点想起 Z1 计算机（维基链接），不过这个机构在摩擦方面肯定更优。我猜机械计算机的大部分研究在电子或继电器计算机变得更有意思之后就基本停止了，所以肯定还有更好的机械设计。事实上，人们日常用的很多机械装置本身就能充当逻辑门，只是没意识到；尽管所有人都关注电子数字计算机，能用来构建计算机的机构其实多得惊人且简单。另外，这篇 PDF 对内容来说相当臃肿，因为所有图都是极高分辨率位图而非矢量图，尽管看起来像是用矢量图软件画的。
userbinator查看英文原文
It somewhat reminds me of https://en.wikipedia.org/wiki/Z1_(computer) although this mechanism is definitely superior in terms of friction. I guess the majority of research into mechanical computers stopped shortly after electrical/relay ones started becoming more interesting, which is why there's definitely better mechanical designs possible. In fact, people probably use a lot of mechanical devices which can serve as logic gates without knowing it --- although everyone focuses on electronic digital computers, the mechanisms on which computers can be built are surprisingly vast and simple. Also, the PDF is rather bloaty for the content, because all the figures are extremely high-resolution bitmap images instead of vector graphics, despite looking like they were created with a vector graphics program.
补充指出长信号线上的力累积与线性应力问题，并补充论文用四相时钟和机械放大来缓解。
文章视频下的一条评论提出两个有意思的点：“存在线性应力问题：一条单一的‘激活’线（时钟信号穿过一串锁）最终可能到达某个点，机构完全无法移动。”你可以在 gif（merkle.com 链接）里看到长激活线的例子，视频对应片段也有更详细说明。我惊讶演讲没明确讨论这一点。晶体管之所以强大，不仅因为能开能关（开关），还因为能放大信号（杠杆）。补充：论文第 4.5 节确实谈到了这点，说处理缩放的方法是用四相时钟把移位寄存器中的单个单元（或更一般地，单个时钟逻辑元件如 Fredkin 门）隔离开，信号线上的力在遇到锁住的锁之前无法越过两个单元，防止力长距离累积；并且每个锁都充当机械放大器，小力可控制大得多的力。
mLuby查看英文原文
A comment on the article's video had two interesting points: > There is linear stress problems that can occur, a single "activation" line (clock signal through a train of locks) can eventually reach a point that the mechanism will be unable to move at all. You can see an example of that long single activation line in this gif http://www.merkle.com/images/4_phase_shift_register_v4.gif and described in more detail in this part of the video: https://youtu.be/yVX9Ob4SjGA?t=769. I was surprised this wasn't explicitly addressed in the talk. Transistors aren't just powerful because they can be off or on (switch) but also because they can amplify signals (a lever). Edit: The paper does address this in section 4.5, saying that "The method for dealing with scaling is to isolate individual cells in a shift register (or more generally, individual clocked logic elements, such as Fredkin gates) by using the four-phase clock. That is, forces along signal lines cannot move beyond two cells before encountering a locked lock, preventing any accumulation of forces over long distances. At any point in time, a cell is either blocked from transmitting forces to adjacent cells, or can transmit forces only to one other cell (either the predecessor or the successor)" and "Force accumulation in link logic systems can be mitigated because each lock acts as a mechanical amplifier, with a small force controlling a much larger one." > I'm really loving the idea of mechanical logic gates and computers... it's far easier to produce in 3D print, which will be vital in space exploration and colonization. Question is, what additional primitives are needed to construct all the primitives (locks and balances) from a bin of links and rotors? Maybe a "lifter" that raises a hinge pin out of 2D plane so it can be placed in a rotor.
相关资源
从这篇论文所有 HN 讨论帖的全部评论里整理出来的链接。
历次讨论
- 2018Hacker News 讨论帖112 分 · 26 评
- 2022Hacker News 讨论帖100 分 · 34 评
同领域还值得读
每个好的系统调节器都必须是该系统的一个模型 (1970)
Every Good Regulator of a System Must Be a Model of That System (1970)
这是控制论经典论文（Conant 与 Ashby，1970）：它试图论证，任何“好的”调节器都必须是其所调节系统的模型——即调节器与被控系统之间必须存在同构或模型关系。
其他2018 · 202 分 · 59 评进阶 · 约 45 分钟编辑精选复杂系统如何失效
How complex systems fail (2002)
总结了复杂系统如何失败的若干条经验法则：系统总是在多重故障的边缘运行，事故是多重因素叠加的结果，操作者实际上同时承担着维持生产与防范故障的双重角色，而事后归因往往把事故简化为某个…
其他2017 · 154 分 · 19 评入门 · 约 30 分钟编辑精选如何阅读论文
How to Read a Paper (2016)
介绍如何高效阅读学术论文，提出“三遍阅读法”：先快速浏览摘要、结论和图表把握概览，再理解主要内容，最后深入细节。
其他2017 · 715 分 · 80 评入门 · 约 10 分钟公元 1021 年欧洲人已在美洲存在的证据
Evidence for European presence in the Americas in AD 1021
这篇 Nature 论文用树木年轮定年，并结合公元 993 年宇宙射线事件标记，把纽芬兰 L'Anse aux Meadows 的 Norse 活动精确到公元 1021 年，为欧…
其他2021 · 655 分 · 391 评进阶 · 约 45 分钟一次通古斯级空爆摧毁了约旦河谷青铜时代城市 Tall el-Hammam
A Tunguska size burst destroyed Tall el-Hammam, Bronze Age city in Jordan Valley
研究约旦河谷青铜时代城市 Tall el-Hammam 被毁的原因。
其他2021 · 653 分 · 342 评入门 · 约 45 分钟