# Chip Fabs in Space: Technically Possible, Completely Impractical

## Метаданные

- **Канал:** Asianometry
- **YouTube:** https://www.youtube.com/watch?v=_2rkRDg0d60
- **Источник:** https://ekstraktznaniy.ru/video/26155

## Транскрипт

### Segment 1 (00:00 - 05:00) []

In July 2025, a startup called Varda Space Industries raised $187 million to make drugs in orbit. They are not the only startup trying to make stuff in space. Starcloud - formerly called Lumen Orbit - raised $21 million to try and assemble a 5-gigawatt data center in space. And tech giants Elon and Bezos have been talking a great deal about putting AI data centers in space. A few people on the internet have riffed off those trends and ideas to also ask, “Why not do semiconductors in space too? ” Bezos himself said all the way back in 2016: "We can build gigantic chip factories in space". Sure we can. But it's not going to be practical. In this video, Jon spends way too much time on space fabs. ## AI Data Centers in Space Let's start with the reasons why people think we should be putting AI data centers in space. Part of this recent push seems to be frustration with how long the data center buildouts have been taking here on Earth due to energy shortages or local opposition. With regards to the latter blocker, there are no local residents in space that I personally know of so presumably we bypass that. With regards to the former - energy - there is something kind of compelling. Up in space, a data center taking a specific earth orbit can face the Sun for extended periods of time. And without clouds or atmosphere, solar panels generate anywhere from 3-10 times more power. There are other cited reasons. Elon has said that space is cold, implying that we can vent the waste heat into it. Though many have noted that in a vacuum, heat dissipation can only happen via radiation. More on this later. And data connections with space-based fabs are assumed to work faster because signals can travel faster through the vacuum of space than within optical fibers. I guess he might be right. Or not. I don't know. This video isn't about data centers. It's about space fabs. ## The Vacuum Before we begin, I want to point you to Nick Pfeiffer and Professor Glenn Chapman at Simon Frasier University. Their work in the late 1990s and early 2000s really set the tone for the existing space-based semiconductor research field. It was not the only source I used for this video but my favorite. Go check them out. Perhaps the single most significant benefits of space-based semiconductor manufacturing are the cleanliness and vacuum. Semiconductor manufacturing fundamentally demands cleanliness. Stray particles like pollutants from the outside environment or previous steps falling onto the wafer or interfering with equipment cause defects that hurt yield and profits. Well, if you want nothing, then you will find plenty of that in space. The vacuum of space in Low Earth Orbit is measured at about ten to the minus five to the minus eight pascals. The most abundant thing up there is atomic oxygen produced by the sun's UV rays hitting the oxygen in the atmosphere. More on this later, it is not a trivial thing. In semiconductor manufacturing, a "high vacuum" environment is classified as being at ten to the negative one to negative five pascals. And an "Ultra-high vacuum" environment is classified as being between ten to the negative seven to negative nine pascals. These are non-standard measurements so there is some variation but the point is that Low Earth Orbit meets the particle density requirements for an advanced fab. On Earth, fabs maintain big HVAC systems to change out the air, filter the air of particles, and maintain temperature. About half of a cleanroom's operating costs are from HVAC systems - though I want to note this is a very rough number. And certain tools require a vacuum. I remember visiting the massive High-NA EUV machine at ASML. The second thing you notice about it - after how big it is - is how loud it gets. Those are the pumps maintaining a vacuum environment for the EUV light. Other tools that require some form of vacuum include those for chemical vapor deposition, dry etch, ion implantation, and sputter physical vapor deposition. Up in orbit, we can get the strong vacuum for basically free. Just open a side-vent to the outside. It is simple, saves energy and avoids

### Segment 2 (05:00 - 10:00) [5:00]

potentially damaging pump vibrations. The resulting environment is thus very clean. To keep any aforementioned atomic oxygen from wafting in and interacting with wafers, Pfeiffer recommends placing those side-vents in the space fab’s wake. So away from incoming oxygen atoms. ## Space Radiation Something that came to mind when I heard about this was the space radiation. Space radiation describes protons and atomic nuclei traveling through the universe. Much of it is solar wind from the Sun but there are other particles from nearby stars and galaxies. The earth's atmosphere and magnetic field deflects many of these particles. And most simply pass through things untouched, but a percentage of these energized particles do leave an impact. In doing so, they knock around atoms and potentially shift the chips' crystal lattice. This leaves behind vacancy defects - a missing atom in the lattice - or interstitial defects - an additional atom in the lattice. This can cause the chip to degrade in performance and eventually fail. It is why electronic products sent into space have to be hardened and why solar cells in space slowly degrade over time. Pfeiffer addresses this, saying that a 200-millimeter wafer receives less than 10 rads of effective radiation damage if it does not have an electric field applied to it. Such damage, he says, can be easily repaired with a simple annealing. This seems reasonable. But I note that this assertion was made 25 years ago. As Moore's Law progresses and transistors get smaller, even a small amount of lattice damage can have significant consequences. ## Temperature Control Another thing that concerns me are the wild temperature swings in space. As I hinted at earlier, fab HVAC systems not only include filters and pumps for cleaning the air or producing a vacuum but also chillers regulating the temperature. More energy is spent on the latter. If 50% of a cleanroom's operating cost is HVAC, then 30 of that 50% are chillers. Heat can come from the outside, like the sun's rays or via dissipation from the inside, like heat-producing equipment. Without an atmosphere, temperatures can swing from negative 220 to positive 220 degrees Celsius. These tools release a lot of heat - some are literal ovens - while being super temperature sensitive. The wafers themselves are just as sensitive too. Variations on their surfaces create physical distortions and can mess with the delicate process chemistry. As mentioned earlier, the primary method to remove heat in a vacuum is to radiate it away. The International Space Station uses liquid ammonia to actively circulate heat to eight big radiators - each about as wide and long as a city bus. Engineers have been dealing with thermal challenges in space for as long as we have been sending stuff to space. And maybe the orbit’s stability helps too. I am no satellite designer. But the sheer amount of heat to dissipate away looks like a significant obstacle. ## In Microgravity The second defining thing about space is the microgravity. Low earth orbit is not zero-gravity. A satellite orbiting 400 kilometers above the earth's surface is just six percent further away than a guy on the ground. So gravity's pull on the satellite is 88% that of the guy. So the weightlessness comes from the fact that the satellite is constantly falling towards the earth. It creates something equivalent to a zero-gravity environment - Einstein's principle of equivalence at work here - but not the same. Ergo people like to say microgravity rather than zero-gravity. Gases and liquids act differently in microgravity. Like, the relative density of substances does not matter when they mix. Which means that we do not have any sedimentation or buoyancy. Particles don’t fall. Bubbles don’t rise. Without gravity to pull them down, liquids are most affected by their surface tension. When by themselves they take a spherical shape. When in a container, they no longer fit the shape of those containers. And on surfaces they spread out into thin films and tenaciously cling on. ## Liquids and Microgravity Don't Mix Some process steps are gravity neutral or even do better in microgravity. The various chemical and physical vapor deposition methods fall into this category. They deposit thin layers of material onto surfaces. CVD via gaseous chemical precursors. PVD via smashing particles off a source so they drift to the target.

### Segment 3 (10:00 - 15:00) [10:00]

These work. Some might even work better in microgravity. An issue with CVD for instance is understanding the complex mixing of gases of varying densities at different temperatures: Convection. That is no longer an issue. Crystal-growing is another thing. This has been studied for some sixty years. The general theoretical expectation is that we can grow larger, purer, and better crystals than we can on earth. Single-crystal silicon ingots can be grown using a method called the floating-zone method. We melt a portion of polysilicon feedstock and let it gently pour down to merge with a silicon seed crystal. It is very pure because the melt never touches a container's sidewalls. But on earth, we are limited to wafers of about 200 millimeters cuz gravity. Microgravity changes this. The big problem are when things involve liquids. In microgravity, liquids lose their predicted behavior - causing breaking changes to the tool. We must either re-engineer the tool or take on a "dry" alternative. Also. In a vacuum, exposed liquids start boiling off as their molecules start pulling apart - releasing potentially contaminating vapors. An unfortunate side effect of having a vacuum everywhere. ## No Wet Cleans Unfortunately, we use a lot of liquids in the semiconductor space. Wet cleans, to start. A large portion of our existing process node involves just cleaning contaminants like particles, metallic and ionic residues, and organic residues off the wafer. Wet cleans use ultrapure water - or in the case of a style of cleaning called RCA clean, liquid acids and alkalines - for this cleaning. They are performed in baths or with sprays. The best dry alternative to RCA clean is probably plasma etch. The fab can produce a weakly ionized plasma - perhaps using atomic oxygen collected from space - and expose that plasma to the wafer. This is already done to strip photoresist layers off the wafer after lithography. The major challenges are that (1) It just takes longer; And (2) It is an ETCH process. We can try to control it but there will be risk of damaging the chip itself. ## No Liquid Photoresist Another wet-based process that will fail is spin-coating. Lithography requires a photoresist. When exposed to light, the resist hardens and retains the chip's design information during the subsequent etch step. Today, the photoresist is a liquid, applied onto the wafer using a spin-coat machine. The machine spins the wafer at a very high rate. We then slowly pour the resist onto the surface, causing it to spread out to a nice, even layer. In microgravity, the liquid photoresist will not stay on the wafer. The machines also use vacuum suction cups to hold the wafer, which too will fail. So we must turn to non-liquid "dry" photoresists. In 2000, Chapman and Pfeiffer proposed an all-dry resist system that uses a bi-layer of inorganic carbon and aluminum and can be etched using reactive ion etching. Overall it works, except it is not very sensitive to UV light, which is not good for throughput. I have not heard anything about it since the mid-1990s. Fortunately, the industry is already moving towards other dry resist solutions to adapt to the stricter conditions of EUV lithography. Metal oxide resists from JSR and Lam Research are applied using vapor deposition. They are also slower, but there is no choice. ## No Immersion and EUV Today's leading edge fabs largely use EUV and immersion 193-nanometer lithography machines. The majority of layers are done using the latter - which is a problem because immersion depends on pumping ultrapure water between the lens and wafer to raise the tool's Numerical Aperture. The use of liquid water is untenable in vacuum, it boils off immediately so we need a pressurized mini-chamber. Plus in microgravity, the water acts in such unexpected ways to necessitate an extensive redesign. Like to prevent defect-causing bubbles, for instance. To me this is a no-go from the very start. How about an EUV machine? Maybe this is better. EUV light already needs a vacuum - though again I worry that oxygen atoms in LEO space can enter and degrade the mirror surfaces. And in microgravity, the big EUV mirrors won't weigh anything

### Segment 4 (15:00 - 20:00) [15:00]

which in theory can reduce the tool's overall size and complexity. Just for fun, can we even send an EUV machine into space? Per ASML, an EUV NXE:3300 machine (so not High-NA EUV) is 150,000 kilograms and has a volume of 765 cubic meters. This may or may not include the drive laser, cooling plant, hydrogen gas support system and other subfloor equipment. Falcon Heavy cannot send that up, so you would have to wait for Starship. The problem however is that the EUV machine famously uses an immense amount of energy - one machine eats about 1-2 megawatts. Producing a megawatt in space wouldn't be impossible, though it would certainly need a massive solar panel. The real question to ask is how are you going to dissipate 1 megawatt of heat generated by the machine? That is 14 times more than what the International Space Station’s radiators reject. Back of the envelope, such a radiator array might get as big as a football field. There are other things that I am thinking about. How do you catch and retain the tin droplets in space without gravity? How do you prevent all the space fab’s jitters and random movements and counter-reactions from ruining the precise lithography overlay? I can do this all day but let's keep going. Let us just assume that any space fab maker goes and rebuilds ASML's 193i or EUV machines... but smaller, more energy efficient, more physically stable, and needing less maintenance. Easy. ## No Wet Etches After patterning the wafer, we need to do the etch step. There are two types of etch: Wet and Dry. The wet etch dunks and shakes the patterned wafer inside a tub of acid etchant. Since it involves a tub of liquid, we cannot use that. Bubbles formed by the chemical reactions also can’t be easily shaken off due to the lack of buoyancy - preventing fresh etchant from reaching the wafer surface and creating defects. Fortunately the industry has largely adopted dry etch tools - Reactive Ion Etch, plasma ashing, etc - to take advantage of certain directional etch benefits compared to their wet alternatives. But wet etch is still used for a few process steps where those benefits are not necessary. These would need to be replaced with dry etch. And at this point, I am starting to wonder - between the cleaning and etching - how many plasma etch tools we are going to need on this space fab. ## No CMP Another significant missing process step is Chemical-Mechanical Polishing or CMP. We use CMP for "planarization", i. e. making things flat. ICs have multiple layers stacked on top of each other, so a non-flat surface can lead to serious defects. CMP combines the mechanical action of a grinding pad with a reactive chemical slurry to grind down layers of material to produce a very flat and level surface. Since that slurry is liquid, it cannot be used in microgravity. The thing is that CMP works so well that "dry" alternatives have not been explored. There are variations - adding ultrasonics or changing pads - but the slurry has always been there. So we need something entirely new. What that is, I have no suggestions. ## New Wafer Handling Systems Another issue to consider is wafer handling: How do we transport wafers between tools and hold them inside the tool? And not damage the wafers too? I mentioned one method earlier in passing: Vacuum suction. This one is untenable in a vacuum environment because we need atmospheric pressure outside of the suction. The brain-dead simplest way to do it on earth is with a spatula-like effector that scoops up the wafer like a pancake. This obviously does not work in microgravity. The robot arm can pinch the wafer at the edges, but that risks damage. So probably your best bet is something along the lines of an electrostatic chuck. This creates an electric field to securely attach the wafer to the clamp. They are already used in vacuum environments like the EUV machine. We cannot use this all the time however, because the electric field's presence can attract space radiation - possibly leading to more damage. In any case, every tool in our space fab has to be re-engineered for this new wafer handling system. ## Logistics and Lift Costs Unlike data centers, fabs need to be constantly supplied with a variety of raw ingredients like chemicals, silicon wafers, metals, and precursor gases. They also need spare parts and access to maintenance. When a tool like a lithography machine breaks down - and it often does - we need to get it back online as fast

### Segment 5 (20:00 - 24:00) [20:00]

as possible. Every minute of downtime costs a commercial fab tens of thousands of dollars. An engineer has to put on a clean suit and go into the cleanroom to see what's going on. And if something needs replacing, a new part needs to be available within hours. They cannot wait to take the next SpaceX rocket up. And as for lift costs, Pfeiffer's 2000 thesis did an economic analysis for the operating costs of an all-dry space fab producing 5,000 wafers a month. It assumes that we redesign all the semiconductor tools to be "vacuum-native", removing pumps, structural supports, and subsystems - which was estimated to reduce the machines' mass and volume by anywhere between 40-60%. The financial analysis does not take into account R&D costs. The base case assumed lift costs of about $5,000 per kilogram and concluded that it would cost twice as much to make a wafer in a space fab as compared to a ground-based fab running wet processes. Again, only operating costs. In the years since, the launch of reusable spacecraft like Falcon, Starship and Electron has lowered lift costs from $10,000 per kilogram to potentially as low as $1,000 to $2,000 per kilogram. Maybe even $500 in the not-so-distant future. But Pfeiffer also ran a scenario that reduced launch costs to about $1,000 per kilogram. He also slimmed down the thickness of the wafers to make them cheaper to transport to the fab. The model still said that space-based operating costs would be 12% higher than that on Earth. That was 25 years ago. Back then, the leading edge was 130 nanometers. 193-nanometer DUV machines were just entering the cleanroom. And fabs were still largely fabricating 200 millimeter wafers. A leading edge fab cost only a billion dollars. Things have changed. Tolerances are stricter. The lithography issues alone are untenable. And today, the most compelling economic benefit of going to space - the free vacuum - is largely a solved problem in the fab. I asked people at several tool vendors about how much engineering mindshare and mass/volume they spend on vacuum conditions like UHV. The answer was very little. A space fab would be unfathomably cool, but this juice is not worth the squeeze. ## Conclusion Okay so where do I fall on the Space Fab? I think it is technically possible. But so is me learning how to dunk. I don't think it's anywhere economically practical. A space node cannot use existing process nodes, nor can it use existing semiconductor manufacturing equipment. So we have to develop both from scratch at the same time. A complete custom job from the ground floor. Which might excite the Silicon Valley "disruption" boys, but feels like a nation-state type project. More Huawei than two dudes in a garage. Moreover, we did not even mention test and packaging - much to the consternation of my friends in the Test industry. Neither are anywhere near as automated as front-end. So we need to send the chips back to Earth and pray the wafers don't break on the way down. At the end of all that, you get a silicon chip that is worse and more expensive than what Intel, Samsung and TSMC have been making at massive scale for over half a century. Why not just send the chips up into space? What makes more sense than space fabs are special material-making facilities that leverage microgravity and relatively free energy to produce stuff like silicon wafers. More well-studied, smaller in scope, and the conditions cannot be as easily achieved on the ground. Let's not rush to put the most sophisticated and demanding mass manufacturing ever attempted into a literally alien world.