0 consumption at idle. Have to put in some good old fashioned Work=Force*Distance when you perform an operation though.
Also, how many toggles can each gate go through before material fatigue causes the logic to get stuck? Some interesting questions which would need answering before it can be utilized by any industry.
Like as if transistors can't pull that out. Just active logic families all went extinct long long time ago because passive logic (powered at idle) was so much easier to design.
My understanding is because FRAM is still difficult to manufacture and is keeping the cost high. It's advantages will lead to it to be pulled in for those low power uses but it'll be a while to make it's way into the niche.
Flexures can last forever if you stay within the plastic deformation range for that material. These can fail from other things though, e.g. mechanical shock, thermal extremes, stuff like that.
I'm skeptical of this claim. Can you expand on your answer that there are materials that can be cycled infinitely in the plastic regime, or point to some sources that explain further?
I did a PhD in a materials science lab, and my understanding is that at least for metals, you still eventually run into fatigue & failure, even if you stay within the elastic regime (which nominally appears reversible). I'm not sure why plastic deformation would be any better, given that you'll have atom positions shifting and (if crystalline) dislocations propagating. I'm not a fatigue expert by any means, so I'd love to learn more.
On the Wikipedia page for fatigue limit[1], I see a link to a reference that says there are no metallic materials that can be infinitely cycled[2]. Is the bonding somehow different in this 3D printed material that causes fatigue to work differently?
One of the graduate students I worked with studied the TI micro-mirrors used in DLP projectors. They were designed in the first round of micromachined devices so by EEs rather than MEs, and as a result they used aluminim.
So the question to answer is, "how does a multi-million pixel display unit array of doubly supported tortional Al hinges going through >1% strain survive for 1000s of hours (much longer than the bulbs) oscillating at 100s of kHz". That's ~10^12 cycles each with ppm defect rates!
Since you know that this far exceeds any reasonable fatigue strain and the defect density /dislocation propagation should be huge! The key is that the aluminum is <1um thick a few um wide and <100um long. The majority if the strain is concentrated at the supported ends, but it's so thin that the whole high strain region is single crystal!
TI didn't originally know why it worked... just that it did.
I won't say mechanical gates are a great idea... and at 1MHz they might start failing after a few continuous months. The truly unfortunate part is that the manufacturing processes aren't really designed to control for the grain structure... they're designed for etch repeatability and conduction stability so yield could fall apart while tour processes seem nominal.
I don't pretend to understand this, but they could be saying that it's not infinite, just very, very large. According to the S/N curve on Wikipedia, it looks like staying in the plastic region for steel means you effectively get enough cycles that the part will "effectively" be immortal (just maybe not going to last until the death of our sun at MHz cycles).
An analogy might be with transistors - if you run them at high temps, you might get ten years out of them. If you're running them at room temps, you might not be able to see signs of anything other than sporadic failure even over long periods of time.
In the industry flexures are considered to have 'infinite' life if they are correctly designed and if you stay well within the working parameters. If you stay within about 2/3rds of the elastic range it lasts forever. Dan Gelbart: https://youtu.be/PaypcVFPs48?t=660
Sorry I meant elastic range. These are only a few hundred atoms across in the flexure areas so if you stay well within the range, it does last 'forever' (years)
A key benefit of the proposed approach is that such systems can be additively fabricated as embedded parts of microarchitected metamaterials that are capable of interacting mechanically with their surrounding environment while processing and storing digital data internally without requiring electric power.
Mechanical circuits could be a low performance replacement for electronics in environments where the latter fail, e.g. because of high temperatures or radiation.
I think I remember something about an idea for a mechanical communications system for a Venus lander that worked by wheels that increased or decreased radar reflections so data could be read from orbit by bouncing a radar signal off it.
Exactly why is it important that this particular structure can be additively manufactured? Typical additive processes can create essentially anything, exactly because there are no limitations steming from tooling geometry and workholding that are typical for traditional subtractive machining.
Sure, anything -- so long as it doesn't have geometric extent, fit, finish, accuracy, internal strain, material composition, or volume economic requirements within a couple orders of magnitude of the competing traditional processes.
In other words, it's nice to know that they were able to achieve these results on a 3D printer rather than a multimillion dollar MEMS facility.
Speaking as Stephenson/Doctorow fan as well as a member of ever-growing sector of the population that have had their technological paranoia continually proven right for the last fifteen years, improvements in additive manufacturing of miniaturised computation technology assists in keeping a small section of power with the larger population, rather than being forced to rely on hyperlarge gated conglomerates to parcel out materials to us.
I don't know why this is being down-voted. Anyone care to elaborate? If we don't want to rely on centralised manufacturing we need to be able to make computers in the privacy of our own homes. Otherwise, whoever controls the entities that manufacture our computers can put in backdoors. The government has straight-up said that they want backdoors into all encryption. Worrying about this, and making progress towards decentralised manufacture, doesn't seem crazy to me.
Does anyone on HN can explain this question to me: is there a real obstacle that prevents one to build a general-purpose digital computer using pure macroscopic mechanical components? Historically, we had several types of special-purpose mechanical computers, but a general-purpose mechanical computer was never built.
Was it because the mechanical parts are too prohibitively rigid/inflexible/heavy/expensive for a computer? Or, a real barrier for mechanical computers doesn't really exist, and we didn't have it simply because it was not economically worthwhile to build one after electronic computers were feasible?
First computers used relays, becase they were already faster and cheaper than purely mechanical computer. Then came lamps which were even faster. Nowadays mechanical computers are just uneconomical curiosity, but thanks to miniturisation they look like they might be economical for several VERY special purposes where silicon just doesn't work.
> Was it because the mechanical parts are too prohibitively rigid/inflexible/heavy/expensive for a computer?
Yes, and it was all those reasons at once. Electrical/electronical computers were better at every metric simultaneously.
You might want to read up on https://en.wikipedia.org/wiki/Analytical_Engine There is good reason to believe that it can actually be built, and a simpler design that is not general-purpose is exhibited in the Science Museum in London.
Thanks, I confused Differential Engine and Analytical Engine, and believed the latter was a more much powerful version, but still not general-purpose. But apparently Analytical Engine is Turing-complete.
Not an expert, but two obvious issues come to mind:
1) Energy loss due to friction & constantly working against inertia (regular and rotational).
2) I don't think you could get anywhere close to even a kilohertz without the whole contraption shaking itself apart. When we say a CPU has a clock of 1GHz, this literally means some components are being activated and deactivated a billion times a second.
Both problems seem to be correlated with size. That is, the smaller it gets, the faster a mechanism can run.
Not sure if possible, but I've found the idea of building a general-purpose fluidics machine fascinating. This might solve the problem in (1).
I do agree with (2) though - I think anything macroscale would simple be orders of magnitude slower than what we can archieve at microscale. I don't know if there is any way out of this.
> I think anything macroscale would simple be orders of magnitude slower than what we can archieve at microscale. I don't know if there is any way out of this.
This is what I was wondering about.
A general-purpose computer, abierto how slow, is still a computer that can process huge amount of computation. Back in the 19th century, the availability of many mathematical and engineering tables was still a problem. And in 1920s, to calculate the requirements of the Afsluitdijk dam project in the Netherlands, the famed physicist Lorentz helped deriving a model from basic fluid dynamics, and it took several years to run the "computer simulation" - to calculate the differential equations in that model numerically by a team of human computers.
Lorentz said,
> The numerical calculations were so lengthy, that we came close to the ultimate limit of what can be done in this way. I myself had no part in this. I did try once or twice to set up and work out such a calculation, but then it would turn out that I had made a mistake, so that it had to be done all over again by others.
As it has been pointed out, Analytical Engine was a real possibility and unlike the Differential Engine, it was genuinely Turing-complete (I thought it was just a numerical solver, and believed Bruce Sterling's Sci-Fi was a bit exaggeration). Just imagine how the course of history would change if some military or industrial funding in the 18th century went to build such a computer instead...
You should have a look at the old IBM Hollerith machines. Although I think they count as electro-mechanical.
> a real barrier for mechanical computers doesn't really exist, and we didn't have it simply because it was not economically worthwhile to build one after electronic computers were feasible
Bit of both: electronics is easier to miniturise, and as a result of that it's easier to manufacture, easier to maintain, lighter, and consumes less power.
Speed of operation of a mechanical computer is limited by the acceleration of the parts.
Like all logic chips today, with a standalone programmer. You make program on your desk computer then send it to chip. Electronic-mechanical converter would setup your device.
Without computer you could do it old-way with a panel of switchable bits and clock signal. You enter required bits with user-switchable levers, then switch clock lever up and down, then enter another machine-word. Or make your program on punch cards.
Apparently it runs fast enough to be programmed interactively. Keyboard input via mechanical linkages is a solved problem: consider the humble typewriter. The Selectric typewriter even had macroscopic digital mechanical "circuitry"; each keypress moved a lever with lugs that encoded a binary code specifying coordinates for where on the typeball the character was located, and special linkages called whiffletrees converted the digital code into analog rotations for the typeball mechanism in the manner of an R-2R ladder or similar in the electronic realm.
The idea is that the CPU and peripherals would communicate via mechanical linkage (say, levers moved, or not moved, according to whether the signal is "high" or "low") rather than electronic voltage levels on a wire. For some peripherals, electromechanical adapters could be built that use relays to convert electronic signals to mechanical ones and back, allowing you to theoretically even plug in standard PC keyboards and displays into such a mechanical computer.
I'd have thought a mask ROM would be incredibly easy to build. And the whole system would be nonvolatile anyway, so storage of mutable data would be straightforward.
Ultimately the code would compile down to binary; getting it into the machine is just like any other interface.
