Semiconductor Manufacturing in Space

[NGTOne]

I had an idea this morning as I was posting in another thread on this forum - as an early space industry, I feel like semiconductor manufacturers might be willing to migrate upwards, for a few different reasons.

First, and likely foremost, the natural cleanroom that exists in space. Semiconductor manufacturing, due to the extremely small sizes of the components (the latest Intel process, I believe, utilizes transistors just 22nm across), necessitates a cleanroom environment - any contamination during the production process will destroy a chip. The best cleanroom in existence, naturally, is the vacuum of space - far better than exists on Earth, and much easier to maintain. This effect is especially pronounced at higher orbits, where the Earth's atmosphere drops off, though even LEO might work.

Second, zero-grav - processor manufacturing is somewhat limited by the size of a silicon wafer that can be created, and (I believe) that one of the primary limiting factors for THAT is structural strength - silicon wafers are INCREDIBLY fragile. There's a certain minimum thickness required for a given diameter of wafer, to prevent it from cracking (and thus being ruined) from gravity acting on it during handling. In a zero-g environment, the only limitation to the size of a given wafer is what the machinery is designed for, rather than structural issues. Bigger wafers, naturally, lead to productivity increases - you can produce more chips out of a single run of the process.

Third, the nature of semiconductor manufacturing - a small amount of raw material can, under ideal circumstances (which don't exist on Earth) produce an incredibly large amount of product, and the margins are good on the product itself - most of the expenses in semiconductor manufacture occur from the construction of the manufacturing facilities (constructing a terrestrial semiconductor fab for modern chips can run well over a billion dollars, and a good chunk of that comes from building what amounts to a giant cleanroom). A hypothetical space-based semiconductor fab could use simple, (comparatively) low-cost rockets to deliver materials and retrieve finished product, due to the low-mass nature of a production run (a ton of silicon can produce, under ideal circumstances, thousands upon thousands of CPUs, and finished CPUs have a value/mass that is much greater than the per-kilogram launch cost of some cheaper rockets (Dnepr, Proton)).

This is my example of the margins on this:

An average CPU (no additional components like heatsink or fan, but I'm assuming it has a thermal cap) weighs about 45 grams. Under a reasonable assumption of 80% efficiency, you get 17 (rounded down to integer value, .7777~ of a CPU isn't particularly useful, so keep in mind that the actual profit margin may be a bit higher) units for every kilogram of raw material launched (assuming, naturally, that the materials are sent up in precisely the right proportions). Now, at a retail value of $300/unit (arbitrary figure, represents mid-high range CPUs), those 17 CPUs net you $5100. The launch cost per kilogram for a Dnepr rocket is about $3800/kg, which nets you about $1300/kg of profit, given 80% system efficiency. For a Proton rocket, your launch cost is about $4300/kg, which still leaves you about $800/kg in profit. Emergent space-access systems such as Skylon (or even Falcon Heavy) can be expected to drive the cost down further. And, since you're launching raw materials and not people or expensive equipment, the cost of a launch failure is comparatively low. Costs can be expected to decrease further once asteroid/lunar mining takes off, as well.

Naturally, this is a paper calculation, and may be affected one way or another by variances that can't be accurately predicted. Different models of CPUs can significantly affect the margins on this - the most expensive of Intel's i7s retails in the range of $1100/unit (and the most costly Xeon for 3 times that), which pushes the margin through the roof.

I don't know, what do you guys think?

Edited September 12, 2013 by NGTOne

[SargeRho]

I didn't consider semiconductor production in space yet, but I think if prices drop sufficiently, it might just become reality.

Similarly, you can make stupid high quality superalloys in space. Due to microgravity, the molten alloy will mix evenly. And the lack of Oxygen makes production more efficient, as you lose no material to oxidation. You can mine the materials you need from near earth objects - including the fuel to haul the materials back - and in the end, make your production fully independent of supply missions from Earth. If there are workers in those orbiting factories, they would obviously need to be sustained though, thanks to the proximity of LEO to earth's surface, you don't need any people there, you can just control it remotely.

[K^2]

I was thinking about that also. Moon might be better than LEO, but you do have a point on gravity. The main reason I liked the vacuum of space for the production environment is that you could switch all your doping to ion beams. Wavelength of the UV light they use to create masks is part of the limitation for size of the components. With ion beams, you could potentially go to a few nm. Of course, things get really weird at these scales. Semiconductors become very non-classical. That might actually lead to certain advantages, but designing the chips will get harder.

[nhnifong]

Raw silicon isn't the only thing that needs to be shipped up regularly. Semiconductor fabs use a large amount of distilled water, liquid nitrogen, and liquid oxygen. They are often built near air fractionation facilities for this reason.

[Nibb31]

I can't comment on the advantages of microgravity in superconductor production, because I'm not too familiar with that field. I wonder how desirable a larger wafer-size actually is or if the industry is really interested in making bigger chips when the trend is to make devices smaller and more integrated.

However, it is still much cheaper to build a clean room or a vacuum chamber on Earth (a cost in the thousands of dollars) than to launch a manufacturing satellite (millions of dollars) or a manned manufacturing orbital station (billions of dollars). The "vacuum of space" around something like the ISS or any spacecraft is far from being pristine anyway. The vicinity of the ISS is just about as dirty as it gets, with particles, paint flakes, residue from venting, outgassing and RCS... In fact, the ISS is surrounded by clouds of crap.

An industrial vacuum chamber is much better, cheaper, and easier to maintain and service, and in addition, gravity has the advantage of pulling any polluting particles to the bottom of the chamber, insuring a cleaner environment than in zero-g.

However, if the need ever arises for space manufacturing, I would envision something like the X-37B to do the job rather than a full blown space factory. Actually, it's hard to think of any use-case for the X-37B other than space manufacturing.

Edited September 13, 2013 by Nibb31

[NGTOne]

I can't comment on the advantages of microgravity in superconductor production, because I'm not too familiar with that field. I wonder how desirable a larger wafer-size actually is or if the industry is really interested in making bigger chips when the trend is to make devices smaller and more integrated.

However, it is still much cheaper to build a clean room or a vacuum chamber on Earth (a cost in the thousands of dollars) than to launch a manufacturing satellite (millions of dollars) or a manned manufacturing orbital station (billions of dollars). The "vacuum of space" around something like the ISS or any spacecraft is far from being pristine anyway. The vicinity of the ISS is just about as dirty as it gets, with particles, paint flakes, residue from venting, outgassing and RCS... In fact, the ISS is surrounded by clouds of crap.

An industrial vacuum chamber is much better, cheaper, and easier to maintain and service, and in addition, gravity has the advantage of pulling any polluting particles to the bottom of the chamber, insuring a cleaner environment than in zero-g.

However, if the need ever arises for space manufacturing, I would envision something like the X-37B to do the job rather than a full blown space factory. Actually, it's hard to think of any use-case for the X-37B other than space manufacturing.

First off, SEMIconductors, not superconductors

Also, wafer size has nothing to do with component size - the trend is towards larger wafers, each of which can produce more individual components (which increases plant productivity). Since semiconductor mass-production began in earnest, wafer sizes have increased from 25mm diameter to 300mm diameter, and the industry is looking at adopting 450mm as the next standard. You can cut many individual elements from one wafer, and having one big wafer rather than a bunch of smaller ones increases your process speed.

The reason semiconductor fabs are so expensive on the ground is because you have to create a GIANT cleanroom - imagine building a hermetically sealed building the size of a warehouse and you get the picture. And, no, gravity doesn't pull ALL contaminants downwards - the kind that are the most risk for contamination during semiconductor manufacturing are the kind that like to float around and have to be drawn out of the facility using extensive HVAC. If a single speck of dust lands on your CPU while it's being photolithographed, that unit is now ruined and has to be junked.

[PB666]

https://mipt.ru/en/news/motttransition

http://www.sciencemag.org/content/349/6253/1202 <------ Primary reference

The manuscript fot this paper, oddly, was submitted over 1 year ago?

[Elthy]

Afaik one of the current limits in shrinking semiconductor products is the wavelength of the light used. They would require even shorter wavelenghts, but those get absobed by air so they have to use a vacuum, which makes the machinery extremly expensive. This would be far better in space...

Im not sure if the microgravity is actualy good for production, while it reduces structural loads it makes most stuff much harder since EVERYTHING floats around if not properly hold.

But the point where this concept realy brakes is setup/maintainance for such a factory. You would have to send up thousands of tons of machinery, which would get obsolete in a few years, seting up such a factory is propably about as complicated as building the ISS. Also what do you do if something breaks? Send up a crew which would need at minimum weeks to prepare?

[Nuke]

intel had a really hard time rolling out its 14nm process. hence the haswell rehash (and why im still running a 22nm machine). so the traditional encarnation of moors law might be coming to an end. there are still things that can be done. bigger wafers do reduce cost per unit. another thing you can do is build vertically. this can further reduce latency within the chip by reducing the distance from the actual cpu to the end of the cache bus. 3d blocks of sram could increase capacity while decreasing latency within the cache. i think that 3d construction might be easier in a zero g environment.

perhaps we might see nanometer scale 3d printing or something like it. using a head to stimulate crystalline growth and place the contaminants needed to create the pn junctions. though i can imagine that being more expensive and harder than lithography and chemical processes.

[VirtualCLD]

Fortunately this right up my alley of expertise, but I will start by saying that I am on the chip design side, not the fabrication side. We do have a couple of research foundries at my place of work (using older process nodes) and I am knowledgeable in the fabrication process flow. Unfortunately I'm away on travel right now so I can only respond with my phone, but if I was in front of my computer, I would be able to give a more detailed response.

I will start off by saying that what makes a foundry (particularly commercial foundries) so large and expensive is not the clean room environment, but of the equipment within the clean room. We're talking giant furnaces, chemical vapor deposition machines, photo lithography stepper machines, photoresist spinners, plasma dry etch machines, acid wet etch baths, etc...

The other thing to remember is that all of this equipment is expensive so fabs make money by producing large volumes. So a foundry must be running 24/7 to make money (three shifts) and they do things in a massively parallel and pipelines flow. That means they have a large quantity of these very heavy machines.

What's limiting semiconductor technologies today are things like light wavelengths used in photo lithography (as Eltham mentioned), getting your crystalline structure defect density low enough, quantum tunnelling through the MOSFET channels and the fact that your device sizes can be measured in tens of atoms. The reason moving to 450mm size wafers is hard is due to the fact that the machinery is expensive and it is difficult to grow a "defect free" silicon boule that is big enough. On average, the boule need to be 50% wider than your wafer size. Other issues are the fact that the wafer must meet a certain flatness and roughness spec. I don't really see how a zero-g, neat vacuum environment would help with those issues. If anything, it could make things worse, since you tend to use gravity to help you grow silicon and make wafers. Admittedly I've now gone from device fabrication to material growth, so I have gone a little off track.

In terms of fabrication, you would need to contain your wet acid baths, you would need to hold your wafers with nanometer alignment accuracy without the aid of gravity (so no traditional flat wafer chucks) and you would need a method of vapor deposition that doesn't require gravity. Not to mention all of the heavy, expensive machinery. Also, we actually can make 450mm wafers that are structurally sound and they can meet the specs, but making the device fabrication machinery accurate enough and affordable enough is what's preventing industry from jumping to 450mm wafer runs.

I will add that it might be interesting to see what effect zero-g has on material growth (aka the silicon crystalline structure) as well as epitaxial growth at the research level. I am not a materials engineer so I don't know.

[J.Random]

^ what he said. Hydrofluoric acid and other toxic or volatile stuff in a pressurized zero-g environment - what could go wrong, right? That's also why we can't have a self-sustained colonies in space or on other planets - at least yet. Making electronic components "in the field" is nigh impossible.

[VirtualCLD]

intel had a really hard time rolling out its 14nm process. hence the haswell rehash (and why im still running a 22nm machine). so the traditional encarnation of moors law might be coming to an end.

As few points I would like add/comment on:

It is very true that Intel had a difficult time getting down to 14nm production chips, i.e. Chips available as a product and not just a demo. As you probably know, Intel uses a tick-tock development cycle where tick is when they shrink an existing architecture design down to a smaller process node and tock is where they design a new architecture in the existing process node. It was speculated that Intel may switch to a tick - tock - tock development cycle due to the time it takes to shrink process nodes. You'll notice that Intel quickly jumped from Broadwell (tick) to Skyline (tock) because designing in the existing 14nm process node was much easier than jumping from 22nm down to 14nm.

Finally,my random-guy-from-the-internet prediction is that while "Moore's law" is coming to an end in Si CMOS technology, I trust ITRS is right and that we will get down to 5nm process nodes. Beyond that, I just don't know.

[PB666]

^ what he said. Hydrofluoric acid and other toxic or volatile stuff in a pressurized zero-g environment - what could go wrong, right? That's also why we can't have a self-sustained colonies in space or on other planets - at least yet. Making electronic components "in the field" is nigh impossible.

Centrifugal force, take a donut and spin it on its axis and you have gravity.

the reason i place the link here is that this technique offers a way to reduce the weight and heat of electronics, not neccesarily build them in space, but on other planets the available semimetals may also differ and this allows for options of metals that might be used to make electronics in space.

One example is that we don't send people to build our colonies, but robots, which then have to build the infrastrucutre for people that arrive later, they would need to build based on the resources they could gather, thus they need options.

[Nibb31]

Centrifugal force, take a donut and spin it on its axis and you have gravity.

And then figure out a way to reliably inject and extract fluid out of your spinning donut tank. Add the complexity of having a motor, joint, seal, cooling system, lubricant, etc... And a counter-rotating mechanism to cancel the torque. And don't forget that the whole contraption has to run continuously.

Nothing is simple when it comes to space hardware. The equipment that VirtualCLD describes is already heavy and complex and expensive. Making it work in space would make it orders of magnitude more complex and expensive.

[PB666]

And then figure out a way to reliably inject and extract fluid out of your spinning donut tank. Add the complexity of having a motor, joint, seal, cooling system, lubricant, etc... And a counter-rotating mechanism to cancel the torque. And don't forget that the whole contraption has to run continuously.

Nothing is simple when it comes to space hardware. The equipment that VirtualCLD describes is already heavy and complex and expensive. Making it work in space would make it orders of magnitude more complex and expensive.

Not the Manf unit, the entire station segment.