Transcription
Until now, advanced chipmaking has been defined by two companies. TSMC manufactures roughly 90% of the world's most advanced chips. ASML is the only company that can build the lithography machines capable of printing them. That balance held for a while.
Now, a new startup claims it can break both. Their new tool can print chips at the sub-nanometer scale in a single exposure at roughly half the cost. And here is the part the most people miss. They don't want to sell this machine. They want to build entirely new chip factories around it.
I've spent more than 10 years designing microchips. And typically, this industry is quite conservative. It moves in tiny incremental steps. But this one is not incremental because if this works, it doesn't just threaten ASML and TSMC. It puts the entire advanced chip manufacturing model at risk. Subscribe to the channel and let me explain.
Every breakthrough in microchips eventually runs into the same wall. How do you print it? Chip manufacturing uses hundreds of tools, thousands of steps, but one step dominates everything: lithography.
This is an EUV lithography machine. It is the most complex and expensive manufacturing tool humanity has ever built. Its job is simple to describe. It prints transistor features onto silicon wafers. It actually defines how small, how dense, and how powerful chips can become. At today's leading edge, those are just a few nanometers wide. See it for yourself. Here we are zooming into a 0.2 nanometer chip by firing electrons. What you are seeing are features just a few nanometers wide. This particular transistor is more than 10,000 times smaller than a human hair.
So how do you manufacture something so small with a machine so colossal, of the size of a bus? And the answer is light. You shine light through a mask onto a silicon wafer coated with light-sensitive chemistry. Where light hits, chemistry changes. Where it doesn't, material is removed. Layer by layer, a chip appears. That's how we turn sand into thinking machines. The idea is pretty simple. Making it work at nanometer scale is not.
As transistors shrunk, light became the limiting factor. Early lithography used deep ultraviolet light at 193 nanometers wavelength. It worked until physics stopped it. It turned out you can't reliably print features smaller than the wavelength you are using. So the industry made a radical shift to extreme ultraviolet lithography. EUV uses light with a wavelength of 13.5 nm, more than 10 times shorter than before. That single change, combined with a cascade of hard-won innovations, is what unlocked the most advanced chips on earth.
But EUV comes at a price. EUV light gets absorbed by almost everything: air, glass, lenses. That's why the entire system has to run in a near vacuum. To generate EUV, molten tin droplets are blasted with lasers inside a vacuum chamber. The light is then reflected using mirrors polished at near atomic precision. This took decades of research, much of it pioneered in the United States. Today, only one company can actually build these machines: ASML in the Netherlands. And each tool costs roughly $400 million. And despite that price, the economics still work. A single EUV machine can generate over $600 million worth of wafers per year. The real challenge isn't buying the machine, it's making it work. And at the leading edge, only a handful of companies can do that reliably. The list is really short: it's TSMC, Samsung, and Intel.
Which brings us to the real problem. As transistor nodes shrink, printing gets harder. Eventually, you can't print the smallest features in a single exposure. So, the industry is forced to rely on a very complicated workaround. It splits patterns into multiple passes. This is so-called multi-patterning. Imagine drawing fine lines with a marker that is too thick. In this case, you draw every other line first, then you shift the grid and repeat. This lets us produce patterns at least four times denser than the original. It works, but every extra pass adds masks, costs, and defect risks. Costs rise fast, and this is a real killer today.
To keep scaling alive, ASML pushed EUV even harder. The new version, High-NA EUV, still uses the same light but with far more aggressive optics, and that enabled scaling beyond 2 nanometers towards the Angstrom era. These machines are already running at TSMC, Samsung, and Intel, but the cost is extreme. Each tool is approaching half a billion dollars. The next step, Hyper-NA EUV, pushing EUV even harder. So instead of changing the source of light, they're extracting even more resolution from the same light by pushing a numerical aperture that makes the machines even larger, more complex, and of course, more expensive.
At some point, the machine becomes so expensive that the chips it enables stop making economic sense. Just think about it. By the end of this decade, leading-edge fabs are expected to cost $50 billion each. This will further concentrate advanced chipmaking in the hands of just a few companies with massive capital. Wafer costs are projected to climb to $100,000 per wafer. And at that point, advanced chips will become inaccessible to smaller companies and new entrants. You see what's happening? The industry keeps pushing the same technology even if the costs rise faster than the benefits.
So here is the real question. What if the problem isn't how far we can push EUV, but that we are keep pushing EUV at all? What if there is a better alternative? What if smaller features could be printed in a single shot at 1/10th of the price?
This is where Substrate, a US-based startup, proposes a different path. Instead of pushing EUV optics even harder, they're abandoning EUV entirely and betting on X-ray lithography. The idea itself isn't new. Researchers have explored it for decades. The execution is: X-rays have much shorter wavelengths than anything used in today's fabs, but they're extremely hard to control. For a long time, the optics simply didn't exist, and generating stable X-rays meant using synchrotrons. These were machines hundreds of meters long, often occupying entire buildings, not something you could ever put inside a factory.
Inside the synchrotron, electrons are accelerated to almost the speed of light. Powerful magnets then bend their paths, and this motion creates extremely bright X-ray light. What changed is not the concept, but the supporting technology. Over the years, the optics improved. X-ray sources became more compact and better controllable. And all this progress accumulated, and then Substrate pulled all these pieces together.
Instead of adding more masks and more steps through multi-patterning, they asked a simpler question: What if we stop increasing complexity and just print it all at once? You can think about it this way: EUV is like sketching a drawing line by line. X-ray lithography aims to stamp the entire drawing in a single exposure. Here, we are dealing with electromagnetic radiation in the range of a few angstroms, which is 0.01 to 10 nm range, which makes it up to 1,000 times shorter than the wavelength of EUV. And this is ideal for scaling because this practically means you can draw smaller features. In practice, it's extremely difficult. X-rays pass through most materials instead of bending where you want them to, making them notoriously hard to control. That's why X-ray lithography lived in research papers for decades. The physics is elegant. Manufacturing has been hard. Substrate claims that they've crossed now enough of these barriers to turn X-ray lithography into a real manufacturing tool.
If this works, it doesn't just change one tool. It reshapes who can afford to build factories at all. Next, we will break down how do they generate this light and the first Substrate results and what it will take them to build a new semiconductor factory around this tool.
While fabs are hitting physical limits, AI is disappearing into everyday life. At CES 2026, the world's largest tech show, everything evolved around AI. DREO showed what AI looks like when it moves directly into everyday products. At the center is DREO's new AIoT ecosystem where intelligence lives directly inside the device. They launched a new generation of AI-powered home and kitchen appliances, including the AI Sensory Lab. Inside, lighting, airflow, humidity, and temperature shift automatically, recreating the atmosphere of a place you remember. This is AI that understands context. Then they brought it into the kitchen. DREO unveiled an AI-powered cooking experience that lowers the barrier to cooking to the point where it becomes effortless, even for me. You type what you want to cook or speak naturally, and the system turns this text and voice into instant, context-aware cooking guidance. And then you taste it: food made with Chef Maker. That's what DREO showed at CES. AI that turns climate, comfort, and cooking into something you don't have to think about.
Now, we are back to the chip factories, to the core problem. How do you produce X-ray light inside a fab without building a particle accelerator the size of a city block? This is where things get really interesting. They're not sharing much details, but it seems they're using a compact particle accelerator built directly into the lithography system. Not a kilometer-scale synchrotron, but something that actually fits inside a factory.
Inside the system, electrons are accelerated to near the speed of light using radio frequency cavities. Those electrons are then sent through precisely arranged magnetic structures. As they pass through these magnetic fields, they are forced to wiggle. And when electrons wiggle at these energies, they emit intense bursts of X-ray light. It's the same idea used in a synchrotron, just compressed down by orders of magnitude to fit inside a factory tool.
Substrate haven't shared much technical details on the tool itself, likely for competitive reasons, but they have shared the results. And this is where a conversation shifts from theory to evidence. What do they actually have to show? So far, Substrate has demonstrated printing of 12 nanometer features. Those are directly relevant for building sub-2 nanometer transistors. They also claim they can use single patterning for all the layers. In other words, printing in one shot what today requires multiple complex passes.
According to the Substrate data, they already achieved the resolution comparable to ASML's most advanced High-NA EUV system. And some of the numbers here are what really caught my attention. When we talk about lithography quality, consistency matters a lot. They report consistent feature sizes across the entire wafer with accuracy down to about 0.25 nanometers. That consistency measured in a fraction of an atom.
If these numbers hold, the implications are significant. This means we will be able to pack more logic into a smaller area and print it all in one exposure using a tool which costs just $50 million instead of $500 million. In practical terms, that translates to better chips for AI and mobile applications at dramatically lower cost.
But that only works if you can control more than just this machine. And that's why Substrate doesn't want to sell these machines, but they want to build the entire manufacturing process around it. And there is a reason for that. At these nodes, lithography alone is not enough. X-rays don't behave like the light today's fabs are built around. They carry far more energy, which means the materials that work for EUV simply stop working here. Here, Substrate has to reinvent the photoresist itself. And the same applies to masks and optics. And then there is the risk of damage and noise. X-rays can push straight through materials. If they are not controlled perfectly, they can damage transistors or introduce subtle defects that destroy the yield.
And finally, there is the topic of throughput because currently Substrate has a demo, but full-scale semiconductor manufacturing is an entirely different universe. Making something work once in a lab is one thing. Making it work reliably across hundreds of millions of wafers at the most advanced nodes is something else entirely. That's why most chipmakers have dropped out of leading-edge manufacturing over the years. At mass production scale, a tool has to run fast all day, every day, around the year. And for EUV, it took more than a decade to make this jump. X-ray lithography would have to go through the same process.
So, Substrate doesn't plan to sell tools. Instead, they want to build their own factory in America, install their own machines, figure out the recipe, and then offer foundry services directly. And this will put them in direct competition with players like TSMC and Samsung. So, this is not just about inventing a new tool. It's about inventing an entirely new factory and the foundry model around it. And this will take at least another five years.
And this is where reality hits. There is a reason NVIDIA stays fabless and lets TSMC handle manufacturing. TSMC didn't just buy machines. They built recipes over decades. They invested hundreds of billions into yield learning and manufacturing discipline. And just imagine, all of that only pays off for them because of scale. TSMC runs around 30 factories and produces roughly 1.6 million wafers every month across many customers and products. And that combination of process mastery, volume, and packaging is why matching or especially beating TSMC is so hard.
So, we will see because this could be the start of a new era or just a lesson on how unforgiving chip manufacturing actually is. If Substrate's plan works, the implications go far beyond technology. Advanced chipmaking is now tightly linked to economic power and national security. And the most important promise here is simple: costs drop dramatically.
To see why that matters, just look at space. For many, many years, it was a government-only domain. Launches were rare and incredibly expensive: tens of thousands of dollars per kilogram to orbit. Then SpaceX changed just one assumption: rockets didn't have to be disposable. Reusability cut launch costs by roughly an order of magnitude, and that single shift changed everything. Lower prices unlocked new markets, faster iterations, entire industries that didn't exist before.
So the same logic applies here. If this new tool can cut the cost of advanced chip manufacturing by half, the consequences are enormous. For innovations, for startups like mine, this would make a huge difference because currently, tape-out in advanced nodes costs millions of dollars, which basically means you get just one shot. If you can drop costs, it means you can do more attempts, faster iterations. That will accelerate innovation and ultimately expand how much compute is available for AI and everything built on top of it.
It's also important to be clear that Substrate is not the only company who is working on particle accelerators as light sources. In the United States alone, there is also xLight and Inversion, and they are building particle accelerators. And there's also research ongoing in Europe, Japan, and China. But unlike Substrate, they are solving a different problem. Most of these efforts focus on the light source itself: how to generate brighter EUV or soft X-ray light to push existing lithography further. For example, xLight is building a free-electron light source designed to extend EUV, and this fits into the existing roadmap and eventually it will help ASML tools to go further, but it doesn't replace them.
Substrate is aiming much higher. They're trying to replace the entire lithography step with a completely different tool and then rebuild the whole chip manufacturing process around it. If they succeed, the effects compound quickly and accelerate the progress in computing. And this one will drive progress everywhere else.
Throughout history, advances in civilization have closely tracked advances in computing, which is why we care about technology and chips a little obsessively. And if you do too, remember to subscribe to the channel. And now watch this episode where I take you inside a secret chip factory to see how future transistors are being developed. Or watch this episode where I break down step-by-step what it takes to build a microchip factory from scratch. And I will see you there. Ciao.