The Insane Engineering of Terafab
Description
What would it take to build a chip factory with more than a hundred million square feet of floor space? We go inside: the air, the pipes under the floor, the robots overhead, the tin droplets and mirrors of an EUV machine, and the math of a thousand steps. Then we ask what abundant chips could make possible, from robots to orbit to the Moon.
Terafab is a proposed project and almost all of it has yet to be built. Figures come from public filings. The campus is our own rebuild of SpaceX's published concept, not footage. Factory interiors show a typical fab, not a published Terafab floor plan. Future scenes are possibilities, not project commitments, and are labeled on screen.
Chapters:
00:00 The factory of the future
01:22 Why the world needs more chips
02:53 Inside the world's most advanced factories
04:40 How we manufacture billions of transistors
07:40 Scaling the impossible
09:40 The world that could come next
12:03 The first step
The Illustrated System: how things actually work,
Transcript
Read auto-generated transcript (2746 words)
Kind: captions Language: en Imagine a factory so large that you could spend an entire [music] day walking through it and still not understand everything happening inside. A place [music] where some of the most sophisticated machines ever invented work together to manufacture something smaller than your fingernail. Computer chips, but these aren't just chips for phones or laptops. >> [music] >> They're the chips that could power millions of intelligent robots, cars that drive themselves, [music] entire computing systems operating above the earth, and perhaps one day the machines that help us build [music] cities on the moon and Mars. This is the idea behind Terafab, [music] a proposed semiconductor manufacturing complex in Texas with a vision that could eventually encompass more than a hundred million square feet of floor space. [music] If every proposed phase proceeds, the investment could reach a hundred and nineteen billion dollars. [music] But here's the extraordinary thing. Almost everything you just saw still needs to be built. And yet many of the technologies that could make it possible already exist. The lasers, the robots, [music] the machines capable of manufacturing billions of microscopic switches. So, what would it [music] actually take to bring this enormous factory to life? And more importantly, what could humanity build if it [music] succeeds? For decades, we've been making computers smaller, [music] faster, and more powerful. But something interesting is happening. We're starting to imagine computers doing things that used to require a human being. Take a humanoid robot. It needs to see the world through cameras, >> [music] >> understand objects, plan movements, maintain its balance, and react when something unexpected happens. [music] All of that requires computation. Now imagine one robot becomes a thousand, >> [music] >> then a million. The same applies to autonomous vehicles. Every car needs a computer that [music] can understand its surroundings and make decisions in real time. Then there are the giant data centers running artificial [music] intelligence. And SpaceX has proposed something even more ambitious, >> [music] >> computing infrastructure in orbit. All of these ideas depend on access to powerful, efficient chips. And Elon Musk believes the world's existing chip manufacturers won't be able to expand production quickly enough for the future he imagines. That's the motivation behind Terafab. [music] Its long-term ambition is to manufacture every year chips whose combined operating power could reach 1 terawatt. That's not how much electricity the factory itself would consume, it's the power those chips would use when they're operating. >> [music] >> An extraordinary amount of computing capacity. But before we can understand how a factory could [music] produce that much, we need to understand what actually happens inside one. From the outside, a semiconductor factory looks like a large industrial building. >> [music] >> But inside, it's more like one enormous machine. Every part of the building exists to support the manufacturing process. [music] Start at the top, air is pushed through extremely fine filters and into the clean room below. >> [music] >> It flows around the manufacturing equipment through openings in the floor and eventually circulates back through the system. Why? Because [music] when you're building structures only a few billionths of a meter across, an ordinary speck of dust is enormous. >> [music] >> And keeping the air clean is only the beginning. [music] Beneath the manufacturing floor is another world. Pumps, pipes, chemical delivery systems, >> [music] >> cooling equipment, exhaust treatment, electrical connections, thousands [music] upon thousands of components supporting the machines above them. >> [music] >> Now, zoom out. A fab needs [music] extraordinarily pure water, stable electricity, carefully controlled temperatures, specialized gases and chemicals, [music] and systems that can keep operating reliably day after day. At [music] Terafab's proposed scale, even the infrastructure surrounding the factory would become an enormous engineering project. >> [music] >> But there's something surprising about the production floor itself. There aren't crowds of people carrying components between machines. [music] Much of the movement is automated. Sealed containers travel [music] overhead. Robotic arms transfer materials between tools, and the entire process revolves around one surprisingly ordinary-looking object, a circular piece of silicon, about a foot wide. >> [music] >> This is a silicon wafer. >> [music] >> It's thinner than a dime, and its surface is polished until it's almost perfectly smooth. Eventually, this single disk can [music] contain many individual computer chips. But first, it has to travel through an astonishing manufacturing process, one involving well over a thousand carefully [music] controlled steps, depending on how they're counted. And one of the most remarkable happens inside [music] this machine, an extreme ultraviolet lithography machine, or EUV. [music] It's one of the most sophisticated pieces [music] of manufacturing equipment ever created, and its job is extraordinary, to help [music] print microscopic patterns onto silicon. Here's how it works. Inside the machine, tiny droplets of molten tin [music] travel through a chamber. A laser pulse hits a droplet, flattening it. Then, a second, [music] more powerful pulse strikes it. The tin becomes plasma, [music] releasing extreme ultraviolet light. And this happens tens of thousands of times every second. [music] The light is so difficult to handle that ordinary glass lenses won't work. >> [music] >> Instead, it must be reflected using extraordinarily precise mirrors. The resulting pattern is projected [music] onto a light-sensitive coating on the wafer, and after further processing, one microscopic layer begins to take shape. [music] One layer, but a modern computer chip contains many layers. >> [music] >> Think of it as constructing a city. Except, every building is microscopic, and instead of roads connecting neighborhoods, there are [music] tiny electrical pathways connecting billions of switches. These switches are called transistors. [music] They control the flow of electrical current, and modern AI processors can contain more than 200 billion of them across their assembled silicon dies. >> [music] >> Now, imagine manufacturing all of those structures with extraordinary precision again and [music] again. New material is deposited. Patterns are defined. Selected regions are etched away. [music] Surfaces are cleaned and polished. And the process repeats. Sometimes for months. >> [music] >> But there's one more challenge, accuracy. Imagine a manufacturing process with a thousand independent steps. And suppose each step succeeds 99.9% of the time. [music] That sounds almost perfect. But multiply those probabilities together, and in this simplified example, only about 37% would complete [music] every step without failure. Real semiconductor yields are far more complicated than that. But the lesson is remarkable. To manufacture modern chips successfully, [music] thousands of processes have to work together with extraordinary consistency. And the fact that humanity can already do this is one of the great achievements of modern engineering. Now, imagine doing it on a much larger scale. >> [music] >> Once you can manufacture an advanced chip, [music] the next question becomes, how do you manufacture an extraordinary number of them? Because you can't [music] simply build 10 identical factories and expect 10 times the output overnight. >> [music] >> Every machine needs to be installed, calibrated, connected to the [music] other systems, and carefully tuned to produce consistent results. Imagine thousands of wafers moving through the factory [music] at different stages of production. Some are being coated. Some are being exposed to ultraviolet light. >> [music] >> Others are being etched, cleaned, polished, or inspected. If one critical machine slows down, the effects [music] can spread through the production line. If one process produces too many defects, weeks of manufacturing work could [music] be lost. The factory has to operate as a coordinated system. And Terafab proposes [music] bringing several major semiconductor disciplines into one enormous manufacturing vision. [music] Logic chips perform calculations. Memory stores the information those calculations need. [music] And advanced packaging brings multiple pieces of silicon together into powerful computing systems. These are different manufacturing challenges requiring different tools and expertise. [music] Bringing them together could create new opportunities for integration. But it also makes the [music] engineering far more ambitious. Now, add everything outside the clean room. The power infrastructure, >> [music] >> the water purification plants, the cooling systems, the specialized suppliers, [music] and the thousands of engineers and technicians needed to make it all work. [music] At this scale, you're not just constructing a factory, you're building [music] an entire industrial ecosystem. One designed to turn raw materials into some of the most sophisticated objects humanity knows how to manufacture. [music] And if a system like this can dramatically expand the supply of computing power, the consequences could reach far beyond [music] the factory gates. Imagine what happens if powerful computing becomes easier to manufacture, [music] more efficient, and more widely available. The first changes might be surprisingly practical. Robots moving heavy materials through factories, [music] machines inspecting dangerous industrial equipment, autonomous systems helping build infrastructure [music] in places that are difficult or unsafe for people. A robot doesn't need to look like a person to [music] be useful, but it does need to understand its environment. And that requires computing. [music] The same goes for autonomous transportation. Imagine fleets of vehicles coordinating deliveries, machines handling repetitive transport [music] tasks, and industrial equipment that can operate with increasing independence. >> [music] >> Of course, more chips alone won't make these systems possible. They also need better software, reliable sensors, [music] efficient batteries, and rigorous safety engineering. But abundant computing could help make them more practical. Now, take that idea beyond Earth. [music] SpaceX has proposed the possibility of operating computing infrastructure [music] in space. It's an ambitious idea. Launching hardware is expensive. Getting rid of heat [music] in space is difficult. And maintaining complex equipment in orbit creates entirely new problems. But consider the engineering possibilities [music] if those challenges can be addressed. Computing systems powered by large solar arrays, >> [music] >> specialized machines operating in orbit, and autonomous spacecraft capable of performing increasingly complex tasks. Now, imagine going farther. Before a large settlement on the moon could support [music] many people, enormous amounts of work would have to be done. Landing areas prepared, power systems [music] installed, habitats assembled, equipment inspected and maintained. Intelligent machines could help with that work. And [music] the same idea could eventually extend to Mars. >> [music] >> None of this is guaranteed by building a semiconductor factory. But these are examples of the futures that more capable, more [music] accessible computing could help enable. And that's what makes the ambition behind Terafab interesting. It's not just about manufacturing more computer [music] chips. It's about expanding the foundation on which future technologies could be built. [music] Of course, none of this happens all at once. The journey starts much smaller. Tesla [music] has described work on a research fab in Austin, where the goal is to learn how to manufacture [music] chips before attempting production at a much larger scale. And the first milestones are surprisingly simple. >> [music] >> Can the equipment be installed? Can the machines work together? Can they produce a functioning chip? And can the process be repeated, [music] improved, and eventually scaled? These questions sound ordinary, but they're the same kinds of problems semiconductor engineers have spent generations learning to solve. [music] And that's perhaps the most hopeful part of this story. There was a time when putting billions of microscopic switches onto [music] a piece of silicon would have sounded impossible. Today we manufacture them routinely. The lasers are real, the mirrors [music] are real, the robotic manufacturing systems are real. >> [music] >> Terafab's full vision remains an ambition, but the technologies that could form its foundation already demonstrate what human engineering is capable [music] of. Maybe the next great leap in computing won't come from one miraculous invention. Maybe it will come from learning to manufacture extraordinary technology on a scale we've never reached before. >> [music] >> And if we can do that, the most exciting question isn't how big the factory will be. >> [music] >> It's what we'll build with everything that comes out of it. >> [music]