Disclaimer: this is a modified cross-post of my post on the Space Exploration SE site, adapted to better fit the Worldbuilding community.

A Von Neumann Probe (VNP) is a sci-fi probe which explores the universe in a self-replicating fashion: it finds a location to recreate itself, does the necessary mining and processing to duplicate every part of itself, and then sends off the duplicate to start the cycle anew.

Suppose a society wants to create a VNP with our current level of technology. There have been real-life attempts to approach this idea (see NASA's Advanced Automation for Space Missions, and two 3D printed 3D printers named RepRap and Snappy as examples), but they all seem to have the fundamental barrier of silicon chip production. To my knowledge, producing silicon chips with a VNP is far from being achievable currently, since producing a full-scale clean room and electronics factory would take a VNP the size of cities or larger. I will assume graphene electronics technology is impractical as well, since that's still in its infancy and would probably need clean room conditions as well. Without a means of duplicating its own electronics, our probe cannot meet the requirements to be a VNP: 100% self-replication.

With this in mind, are there any somewhat practical alternates to silicon electronics that could be used for our fictional society's VNP? Vacuum tube technology, pneumatic instead of electric transistors, mechanical computers, something else? Emphasis on the practicality of the designs.

  • $\begingroup$ I'm going to present a frame challenge here: What makes you think Von Neumann probes = silicon chips? That'd be like me saying robots = silicon chips, when robots clearly utilize a myriad of technologies in tandem including motors, batteries, but also non-electrical technologies such as pneumatics and engines. If you are asking about computation specifically, you have not made that clear. $\endgroup$
    – DKNguyen
    Commented Aug 22, 2020 at 3:36
  • 1
    $\begingroup$ @DKNguyen our current electronics infrastructure is built on the backbone of silicon chips. Our VNP may be a 3D printer that's 3D printed or a CNC that's CNC'd, but currently the barrier in 100% replication is electronics production. $\endgroup$
    – Drake P
    Commented Aug 22, 2020 at 3:41
  • $\begingroup$ Again, I don't see why Von Neumann probe = electronics any more than I think robots = electronics. Are you asking about computation, specifically? Because if you are you haven't really said so. It just sounds like you are asking about Von Neumann probes as a whole. Or are you saying that 3D printers can make anything except for electronics? $\endgroup$
    – DKNguyen
    Commented Aug 22, 2020 at 3:42
  • 1
    $\begingroup$ @DKNguyen see my bolded second sentence I just edited in. It's not so much that robots = electronics = VNP. The concern is that without a means of replicating its own electronics, the probe isn't a VNP by definition. Since I see silicon production as untenable for a probe at current technology, I'm asking for feasible alternates to silicon electronics as a whole, so our probe can achieve full self-replication. $\endgroup$
    – Drake P
    Commented Aug 22, 2020 at 3:47
  • 1
    $\begingroup$ Yes, I believe so. Currently our technology can enable a 3D printer that 3D prints its own parts, for example, but without a means of producing the electronics to control it, full self-replication is unattainable. Closing that gap by using alternate means of computation is the goal of this question. $\endgroup$
    – Drake P
    Commented Aug 22, 2020 at 3:51

6 Answers 6


What will replace integrated circuits?

The replacement for the small scale integration of the early 1960s was medium-scale integrated circuits; medium-scale integrated circuits very replaced in the 1970s by large-scale integrated circuits; and large-scale integrated circuits were replaced by the very large scale integrated circuits in the mid-1980s. By the time we will be ready to make von Neumann probes, we will surely have unimaginably large scale integrated circuits.

Why not use discrete components, be they electronic transistors or valves, electromechanical relays, pneumatic valves, or mechanical gears? What is so special about integrated circuits?

The answer is threefold: speed, reliability, and power consumption.

  • Speed:

    Electronic valves are painfully slow; their problem is that by their basic principle of operation, electrons have to travel appreciable distances (on the order of millimeters in the smallest valves) from the cathode to the anode; this takes time and limits the speed at which the valves can operate.

    Yes, there are applications where special vacuum tubes are used in the gigahertz range. Those applications do not involve switching, and are not useful for computation.

    Nevertheless, electronic valves are the nearest thing to a practicable replacement for solid-state electronics; and, historically, they were indeed used for building practicable digital computers, the most powerful of which was the famous AN/FSQ-7 Combat Direction Central of the American Air Force Semi-Automatic Ground Environment, which directed and controlled the NORAD response to a potential Soviet air attack. The Q7 used about 50,000 miniature vacuum tubes, consumed 3 megawatts of power, weighed 250 tons and operated at the blazing speed of 75,000 instructions per second.

    That's slow.

    But why are we so obsessed with speed? Doesn't a von Neumann probe have all the time in the world?

    Oh no it doesn't, not if wants to do anything useful. Fabrication processes, for example, happen at the speed they happen, and in order to control them, the computer must operate fast enough to satisfy hard real-time requirements. Computerized machine tools and automated fabrication processes only became feasible when computers became fast enough to be able to keep pace with the outside world.

    Pneumatic valves, eletromechanical relays and mechanical gears are very much slower, and cannot really be considered. Fun historical factoid: the first automated telephone exchanges were built with electromechanical relays, obviously. By the 1950s they proved too slow, and were replaced with electronic exchanges. A technology which was proven too slow to operate telephone exchanges is not suitable for building artificially intelligent beings.

  • Reliability:

    Electronic valves are not reliable. Discrete electromechanical or pneumatic components are worse, and mechanical gears are the worst.

    The best and most reliable low-power long-life vacuum tubes, designed and built specifically for use in the mammoth computers of the late 1950s and early 1960s, reached lifespans of hundreds of thousands of hours. For the ludicrously slow and SAGE computer mentioned above, this meant that a failure would occur only every couple of hours or so, which was a tremendous achievement for the time, but it is of course unacceptable for the proposed application.

    This is the bane of any system built of many separate parts. The reliability of the system decreases exponentially with the number of separate parts. The usual mitigation is to divide the complicated system into separate modules containing only a reasonable number of parts, and duplicate or triplicate each module; this is what we do for airliners, for example, which, without this redudancy, would be much too unreliable to be useful. However, the best solution is to do away with the complex system completely, and replace it with one integrated solid-state part.

  • Power consumption:

    Electronic valves are voracious power consumers. Pneumatic valves are worse, and electromechanical relays and mechanical gears are the worst.

    Remember that SAGE computer above, and its stupendous 3 megawatt power consumption? That is half a million times more power than that consumed by a modern general-purpose low-power CPU such as the Intel Celeron N3000, which runs about two hundred thousand times faster... A lowly Celeron N3000 is about one hundred billion times more power efficient.

    But doesn't a von Neumann probe have all the power available it needs? Nope, it doesn't. It's the problem of cooling. That power needs to be dissipated as waste heat. Even in good conditions, such as in the friendly atmosphere of Earth, getting rid of 3 megawatt of waste heat requires the use of very large liquid cooled radiator or maybe a small cooling tower. I don't want to even think about how to reject that amount of heat in the vacuum of space. (And remember that 3 megawatts only buys you less than 100,000 instructions per second. Pitiful.)

The point being that there is no reasonable replacement for solid-state electronics using big, visible, separate components. On the contrary, the race is downwards towards smaller and smaller and even more tightly packed integrated components.

Then how would a von Neumann probe manufacture integrated circuits?

It is not the material out of which integrated circuits are made, it is the required size and precision of the parts. Modern integrated circuits are made in complex and extremely expensive fabrication plants because they have very very small very very precise features, and the only way we know how to make such very very small very very precise features requires expensive photolithography and clean rooms and vapor deposition machines and so on.

  • Not all integrated circuits are made of silicon. Just for example, gallium arsenide is also used in special applications.

  • And guess what, integrated circuits are not the only components in common use which require clean rooms and expensive equipment to make... The tiny nozzles of ink-jet printers, the minute light-emitting diodes of OLED displays, the exquisitely precise lenses of a modern superzoom camera, the microscopic mirrors (and their actuators!) of the digital light processing imaging devices used in modern cinema projectors, also require extreme fabrication technology. And the list can be greatly expanded...

    Yes, we do make very very small electromechanical devices -- the DLP devices mentioned above are an example, piezoelectric accelerometers are another; and they are made with the same technology as integrated circuits...

  • Moreover, we know that our way of making very very small very very precise parts is not the only way. In fact, there is a common, mundane natural process which also makes very very small very very precise parts, and it does not require eye-wateringly expensive fabrication plants, and advanced vacuum, and clean rooms, and extreme photolithography and so on: and that process is life.

    Living cells assemble minute components, and vehiculate them, and use them, by means a complex molecular machinery which operates in a warm, icky, aquatic medium. It is just that at our present level of knowledge we can make expensive fabrication plants, but we cannot, yet, design and make the kind molecular machinery used by living cells.

The point is that silicon integrated circuits are nothing special; they are not the only products in modern technology which require very very small very very precise features, and anything which requires very very small very very precise features can only be made, at our present level of technology, with complex machines, and clean rooms, and advanced vacuum, and so on. It doesn't matter whether the device to be manufactured is electronic or mechanical -- it's the size and precision of the parts.

But this does not need to be the case forever, and we know it won't be the case forever. People are working on developing technology which will allow direct manipulation of materials at the atomic and molecular level, maybe inspired by the processes used by living cells, maybe wholly new: and when such technology will become available, integrated circuits could be grown in a portable fabrication unit.

In the end, the von Neumann probes of the far future will have two avenues of replication:

  • Either they duplicate the development of human tehnology, that is, make the machines to make the machines to make the machines to make the machines which make integrated circuits, and microsocopic mirrors, and and accelerometers, and minute actuators and so on.

  • Or else, use some as yet unknown future technology which allows direct manipulation of materials at the atomic and molecular level, and directly grow the complex parts they need.

By the way, "chip" is a colloquial name for "integrated circuit". In a serious discussion, integrated circuits are called integrated circuits, not "chips", the same way that people are called people or persons and not "guys", applications are not called "apps", and fabrication plants are not called "fabs".

  • 1
    $\begingroup$ @DrakeP: Edited to illustrate why the race is towards more and more integration, and why big discrete components won't ever cut it. $\endgroup$
    – AlexP
    Commented Aug 22, 2020 at 17:54
  • $\begingroup$ thank you for the revisions. +1, even if the conclusions of the answer are a bit of a bummer ;) $\endgroup$
    – Drake P
    Commented Aug 22, 2020 at 20:28
  • 1
    $\begingroup$ @DrakeP: I still don't get why it is a bummer. Building a functional von Neumann probe requires anyway massive theoretical and technological breakthroughs is many areas. Building microchips one atom at a time is a problem which is at least theoretically solvable, and which already has prototype demonstrations. Other essential aspects, such as full autonomous general purpose AI, fully functional computer vision, or general purpose fine manipulators, are very much more vaporous vaporware at present. $\endgroup$
    – AlexP
    Commented Aug 22, 2020 at 21:20
  • $\begingroup$ Maybe the inability to manufacture these tiny chips is a good thing. It means your VNP has to take all of the chips it will ever use with it on the journey to a new star system and can only replicate as many times as it has chips for. The alternative is that VNPs could run out of control and turn the entire galaxy into paperclips! $\endgroup$ Commented Aug 24, 2020 at 13:36

What you are making these chip sets out of is not the issue. It is how small you are making them. Full-scale production facilities are unavoidable under current or near future tech as long you are trying to make something that small and precise, but there is a simple solution that does not require reinventing the computer chip.

The trick is that your Von Neumann probe does not need to be able to make computer parts at all. All your Von Neumann probe needs to be able to make is the factory where more Von Neumann probes are made. Until the factory can start churning out its own electronics, the factory can run entirely off computer systems already on the probe. By loading up with a bunch of spare single-board computers similar to Rasbery Pis, it could carry the programmable control systems for over 200 independently operating systems in a storage compartment no bigger than a bread box. This way, when it lands it can create computer controlled mining bots, refineries, production lines etc. all the way up until you have a fully functional factory.

The same could be true of other hi tech parts that you may need early on like optical systems or wireless communications components.

Once you have a fully functional IC factory, then you can produce more and more of these simple computers to run things up until you have enough to manufacture all the things you need to start making more VNPs, including the hundreds of spare computers.

  • 1
    $\begingroup$ Space itself is already cleaner than some clean room levels at 5 particles per CC so you don't need to build a clean room, just the machines inside. $\endgroup$ Commented Aug 22, 2020 at 18:36
  • $\begingroup$ Adding a later of indirection, constructing a factory to make IC chips, does not solve the problem of making IC chips. $\endgroup$
    – CWallach
    Commented Aug 23, 2020 at 7:33
  • 1
    $\begingroup$ @CWallach How so? The issue here is that the whole factory is too big and complex to fit on a VNP, but the planets is hops are all going to have plenty of materials and space to work with. Humans technology is all based on using tools to make tools. A person thrown on a deserted island could with enough time and knowledge make a IC, the question is only how many intermediate stages he needs to account for. So, the VNP would be based on this concept, by only bringing the smallest and most time saving of tools in the process of going from point A to point Z you minimize steps and time wasted $\endgroup$
    – Nosajimiki
    Commented Aug 23, 2020 at 13:47
  • $\begingroup$ Equipment in the factory needs IC chips. How are those suppose to be manufactured? $\endgroup$
    – CWallach
    Commented Aug 23, 2020 at 16:06
  • 1
    $\begingroup$ @CWallach The IC chips you need to run a factory are very small; so, your VNP can carry all the processing power you need to get started on board. I was perhaps too vague in my wording originally; so, I reworded it to expressly explain that the extra processing power is in the form of parts. You don't need to build IC chips because those are very small and easy bring enough of with you to get up and running. $\endgroup$
    – Nosajimiki
    Commented Aug 24, 2020 at 2:50

Maybe you can circumvent the need for electronics by using nanobot networks. But it would be slow. But at least they could reproduce themselves.

For electronics:

The telescope does it

Surely your VNP has some kind of sophisticated very large telescope to spot things in the nothingness of Space. Turns out, in chip manufacturing, the projector with its optics is the most complicated part!

The hard vacuum is giving you a Fab space for free. Just blow up a tent with an inert gas and you have a very clean atmosphere to work in. No dust there.

Chip Fabs are built as vibration-free as possible. That's for free in space.

Then you need a good monocrystal of Silicon or any other future chip material. That's low tech, we were making them 70 years ago. Today those are bigger and cleaner, but still that's not the point which would stop a VNP.

Then you need to saw it into plates and polish those into perfectness. I still guess it wouldn't stop a VNP.

Then you have to apply chemical agents, layer after layer, and light them with a miniaturized plan of the chip. This is the complicated part, it is here where the battles are fought today.

You take a supersized chip plan, use some kind of optics to miniaturize it to the wished-for size and then you use the smallest photon (read: highest possible energy, today this is UV) that your optics can work with, to project the picture on the waver. Clean away the agent, apply the next layer of chemicals, repeat with the next layer's chip plan. You need to repeat that a dozen times over with different chemicals.

They are even building limited 3d structures today, but I don't know how... I left that area 20 years ago. 😬

The consecutive plans have to be projected to the exact same spot, to within few nanometers exact for today's electronics, as sharp as possible. There is always something going wrong with today's tech, so that's the reason why we have computers with three processor cores: the non-functioning ones are software-disabled, the others sold. You could use those little failures as a story-device to explain differing personalities among your probes.

Your VNP certainly has super-good optics for space observation and the capability to replace them in case of. I think it has matching production capabilities already, right?

So there is no reason not to give it the electronics production optics, too. Or plans how to make them and how to use them.

It all comes down to the little fine adjustments in the end, which is time-costly. But if a VNP has something then it is patience and time, so no problem here.


Possibly ludicrous answer:

Build it without any semiconductors.

A semiconductor is just a neat way of making conductors, insulators and switches in one easy package. There's nothing stopping you building a complex electrical logic circuit from relays. Of course, a relay-based computer is going to be really power hungry and have really slow clock speeds. Clock speeds aren't a problem because your VNP will spend thousands of years in transit, so a couple decades to make a decision about something shouldn't be much of a problem. Power is a bit more of a problem, but it's a problem even for silicon based VNP's. You only really need to run the computer when you're in a star system of interest, so you should have free solar energy at that time.

Of course, computation is only part of what silicon is used for. For example, solar panels. There are probably solutions here as well: bimetallic strips moving magnets through coils and set your spacecraft rolling. (Or if you have some sort of fluid system that can last the life of the VNP, have a solar-stirling engine).

Camera's are also made from silicon. Can we do this some other way? Sure, photomultiplier tubes don't need semiconductors. You'd need a fantastic array of them to get any resolution but because space is a vacuum maybe you could make them more compact?

Silicon is just a convenience material. Computation, actuation and detection can all be done other ways at the cost of additional power requirements and larger equipment. Remember that silicon is a recent invention, and analysis of stellar motion, particle physics and just about everything else predates the transistor by decades to millennia. So perhaps to design your VNP, have a look at how the scientists of yesteryear did things.....


Brains. Once you learn how brains actually work, you could potentially engineer special-purpose ones. These need not be human scale brains: something the size of a mouse or rat brain should be adequate for running a space probe.

The advantages is that if you provide a self-contained support system (AKA "body") for the brain, you can easily solve the problem of building new ones. On the down side, you'll need a rather larger life support system, since you can't just plug them into a wall socket.


Human computational machines have only ever used 5 basic technologies. They are, in rough chronological order:

  1. Purely mechanical (Babbage engine, brass gears and steam power, theoretical only never built).
  2. Electro-mechanical (relays, but digital logic is possible).
  3. Vacuum tubes (purely electronic, but difficult to scale, digital and analog both possible).
  4. Purely analog electronics (useful for solving some mathematics problems, but general-purpose logic is impossible).
  5. Semiconductors (primarily silicon, but occasionally indium/gallium/etc used, digital logic really took off with this stuff).

None of the preceding four are appropriate for a VNP. The first two are mechanical and prone to physical wearing. The third is so failure-prone it's amazing anyone ever bothered to attempt it, but when you're a government and throwing millions at problems you're desperate to solve, I suppose you'll do nearly any absurd thing. The fourth was mostly for research projects.

And of course, the fifth has given us (lately) a global computer network that lets you ask the question and me answer it. It might even be sufficient for your VNP, if not for the problems you describe.

Almost all of our impressive "software" has two primary qualities. The first is that it is digital, relying on discrete values. It need not be binary/boolean necessarily, but the analog systems are out. Furthermore, it is also needs to proceed on a Turing-machine-style logic. While we can theorize about how things like a Game-of-Life pattern can produce awe-inspiring complexity, we lack the science to create software that takes advantage of that.

We can also make some guesses about other things that the technology can't be. It probably can't be photonic (current conceptions of this assume that it'd be semiconductor as well). It can't be much humbler than what we're capable of now (1970s software and 1970s storage density just won't cut it, nor likely even 1990s).

I propose that out of all the concepts and ideas I've read about over the years, only one is reasonable appropriate for your question. Magnetic logic.

Instead of using a semiconductor for this, researchers have investigated in the past few decades using iron as a substrate. They'd magnetize small regions on this substrate, such that they were adjacent in a chain. Since magnets don't tolerate the same poles of another magnet next to them, one magnet flipping its poles would cause the adjacent magnet to flip, and so on down the line. When I read about this 15 or 20 years ago, the idea was that this would be ultra-low power computation (no need to feed electricity to it for signal propagation). Unfortunately, I can't find the original article (might have been Scientific American or some other pop science rag).

The big question with that though, was how to implement logic gates. For that, I present this article.


Again, it's using a ferrous substrate. Cobalt this time, but this is just the first attempts with the technology (it's possible that an advanced version might use something else).

Refining iron or cobalt is something simple machines could manage, even to within the tolerances this technology might require. There's no photolithography or difficult chemistry involved. Masking might be done with something as simple as laser-sintering. Though the solid-state lasers themselves again probably require semiconductors, these aren't typically integrated circuits and are far simpler to manufacture.

I don't know that there's a popular term for whatever this technology might be called.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .