The World's Most Important Machine

Veritasium. The World's Most Important Machine. Hosted by Derek Muller and Casper. Duration: 55 minutes.

The most complicated commercial product humanity has ever built costs 400 million dollars, and there is only one company in the world that can make it. This is the story of ASML's extreme ultraviolet lithography machine, the device that saved Moore's Law and makes every advanced microchip on the planet possible.

The Problem That Nearly Killed Moore's Law

Every microchip starts as purified sand, melted into ultrapure silicon, pulled into a single crystal ingot, sliced into wafers, and then patterned layer by layer using light. The process is called photolithography. You coat a wafer with a light sensitive material called photoresist, shine light through a patterned mask, and the exposed areas weaken. Rinse them away, etch the silicon underneath, deposit metal, and you have one layer of a chip. Repeat this dozens of times, stacking transistors on the bottom and metal wires above, and you get a finished chip.

For decades, the industry kept shrinking transistors by using shorter wavelengths of light. Shorter wavelength means smaller features, as described by the Rayleigh Equation. By the late 1990s they had settled on 193 nanometer deep ultraviolet light, and that carried the industry all the way to around 2015. But then they hit a wall. The features they needed to print were smaller than what 193 nanometer light could resolve, even with tricks like immersion lithography and multiple patterning. Moore's Law, the observation that the number of transistors on a chip doubles every two years, was about to die.

A Crazy Idea From Japan

Back in the 1980s, a Japanese scientist named Hiroo Kinoshita had a radical idea. Why not use much shorter wavelengths, around 10 to 14 nanometers, essentially soft X-rays? In theory this would let you print vastly smaller features. But the problems were enormous. At these wavelengths, nearly everything absorbs the light. Air absorbs it. Glass lenses absorb it. You would need to work in a vacuum, and you could not use conventional optics at all.

Kinoshita found a paper by Jim Underwood and Troy Barbee about special multilayer mirrors that could reflect X-rays. The trick was stacking extremely thin alternating layers of two materials with different refractive indices. Each boundary reflects a tiny fraction of the light, and if the layers are tuned so the reflected waves line up constructively, you get meaningful reflection. Kinoshita spent two years building curved multilayer mirrors for 11 nanometer light and managed to print lines four microns thick. It was a proof of concept.

In 1986, he presented his results to the Japanese Society of Applied Physics. The audience refused to believe him. They thought bending X-rays to form images was a "big fish story." Kinoshita was devastated.

Lawrence Livermore and the American Believers

Across the Pacific at Lawrence Livermore National Lab, a nuclear weapons facility east of San Francisco, a scientist named Andrew Hawryluk was using similar multilayer mirrors to study X-ray light from fusion reactions. In 1987, a visiting professor from Cornell challenged him with the question, "Can you do anything useful with this stuff?" Hawryluk was so inflamed that over Christmas break he wrote a white paper applying the mirrors to chip lithography.

Five months later he presented at a conference. Like Kinoshita, he was laughed off the stage. Every expert he admired lined up at the microphone to explain why it would never work. He told his boss he would never speak of it again. But three days later, Bill Brinkman, the Executive Vice President of AT&T and Bell Labs, called and invited him to present. At Bell Labs, Hawryluk finally found fellow believers.

The US government was encouraging national labs to commercialize Cold War research, so Bell Labs partnered with Livermore and two other labs to keep developing what they now called extreme ultraviolet lithography, or EUV. By 1993, Kinoshita opened the first international EUV conference near Mount Fuji, declaring that technology would advance from the micro to the nano to the pico.

The Engineering Test Stand and Industry Panic

In 1996 the US government cut funding for EUV. Intel, Motorola, AMD, and other chipmakers panicked. They estimated 193 nanometer tools would fall behind Moore's Law by 2005. So they pooled 250 million dollars, the largest ever private investment in a Department of Energy research project, to keep EUV alive.

By 2000, the labs produced the Engineering Test Stand, the first fully functioning EUV prototype. It used a 1,700 watt laser firing into xenon gas to produce 9.8 watts of 13.4 nanometer EUV light, which bounced off eight mirrors to print 70 nanometer features. It proved EUV could work, but it could only process about 10 wafers per hour. Commercial viability required hundreds of wafers per hour, running nonstop, 365 days a year.

The fundamental problem was light loss. Each mirror reflected about 70 percent of the light. After nine bounces including the reticle, only 4 percent of the original light reached the wafer. You could reduce the number of mirrors, but you needed at least six to control optical aberrations. Even with six mirrors and a reticle, only 8 percent of the light survived. The source needed to be dramatically more powerful.

One by one, American companies walked away. That left a single company to commercialize EUV: ASML.

ASML Takes the Bet

ASML, originally a Philips spinoff operating out of a shed in a small Dutch town, had joined the US EUV consortium early. Now the entire project landed on their shoulders. They partnered with Zeiss, the legendary German optics company. Zeiss would handle the mirrors. ASML would build the light source.

The mirrors needed to be silicon and molybdenum multilayers tuned for 13.5 nanometers. An alternative using beryllium offered higher reflectivity at 11 nanometers, but beryllium is extremely toxic. Zeiss perfected the mirrors using sputtering, bombarding target material with ions to deposit atoms on the mirror surface, then smoothing imperfections with an ion beam technique they described as "shaking atoms until they fall into the right hole." The resulting mirrors are the smoothest objects ever made. If you scaled one up to the size of Germany, the tallest bump would be about a millimeter. For the newer high NA machines, if the mirror were the size of the Earth, the tallest bump would be no thicker than a playing card.

Building a Sun on Earth

The source was the hardest part. ASML first tried discharge produced plasma, essentially ionizing metal vapor with an electric field. They got it to a few watts but could never scale it. So they switched to laser produced plasma. A powerful laser hits a target material, creating a plasma hotter than 220,000 degrees Celsius. Electrons get ripped from atoms, and when the laser shuts off, they recombine and emit EUV photons.

The Engineering Test Stand had used xenon gas, but xenon's conversion efficiency was terrible, only 0.5 percent. Most energy went into producing 11 nanometer light that the mirrors could not use. ASML switched to tin, which has a much higher emission peak right at 13.5 nanometers, giving 5 to 10 times better conversion.

But tin brought its own nightmare. Each explosion of a tin droplet sprays debris everywhere, and just 30 centimeters away sits a Zeiss mirror so smooth it took years to make. A single nanometer of tin contamination would ruin it. ASML fills the chamber with hydrogen gas, which slows tin particles and chemically reacts with them to form stannane, a gas that can be flushed away. But the hydrogen itself gets heated by 50,000 tiny explosions per second, essentially miniature supernovas. Engineers discovered the shockwaves follow the same Taylor von Neumann Sedov formula that describes nuclear blasts and actual supernovas. They needed to flush hydrogen at 360 kilometers per hour, faster than a Category 5 hurricane.

The Breakthrough of the Double Pulse

By 2013, ASML reached 50 watts but needed 200. The tin droplets were too dense. Most EUV light was getting reabsorbed by neutral atoms before reaching the collector mirror. The breakthrough was hitting each droplet twice. The first pulse flattens the spherical droplet into a pancake shape. The second, more powerful pulse vaporizes the pancake into plasma. The larger surface area and lower density meant dramatically more usable EUV light escaped.

By 2014 they hit 100 watts. But the industry had moved the goalposts. Improvements in multi-patterning with older technology meant EUV would only be commercially justified at 200 watts and 125 wafers per hour. ASML was being "crucified at every conference" as Jos Benschop put it. Customers were running thin on patience.

In a moment of desperation, when Kinoshita visited ASML, Jos took him to dinner. Across from the restaurant was a chapel. They went in and Kinoshita lit three candles for the three EUV suppliers. Jos swears there is a strong correlation between the candle lighting and the power going up, though he admits it is not causal.

The Race to 200 Watts

The precision required is staggering. Each tin droplet is the size of a white blood cell, moving at 250 kilometers per hour. The laser must hit it three times in 20 microseconds, and it does this 50,000 times per second. As one engineer put it, it is like landing a golf ball in a hole 200 meters away, every single time, while shooting through a tornado of hydrogen.

By 2015, ASML board members were summoned to Korea by customers who were "really fed up." Either demonstrate 200 watts or go away. When they boarded the plane, the experiment was still running. When they landed, the first 200 watt result had come in. As Jos said, "This is how close we came."

The Final Fix and Commercial Triumph

One last problem remained. High energy photons and hydrogen ions were degrading a special coating on the collector mirror, requiring cleaning every 10 hours. An engineer noticed that every time they opened the machine, the mirrors seemed cleaner. Oxygen from the air was doing the cleaning. They started adding tiny amounts of oxygen to the vacuum system, and the collector lifetime extended dramatically.

By 2016, orders poured in. ASML's first commercial machines could print 13 nanometer features. Then they doubled down on the next generation, the high NA machine with a numerical aperture of 0.55, investing in it before the first generation was even commercially proven. Intel put in 4.1 billion dollars and Samsung and TSMC added 1.3 billion combined.

Inside the Machine

The video crew got rare access to ASML's clean room, one of the cleanest environments on Earth. In any cubic meter, there can be no more than 10 particles of 0.1 microns. For context, hospital operating rooms allow 10,000 such particles. The machine itself ships in 250 containers across 25 trucks and seven Boeing 747s. It contains 100,000 parts from 5,000 suppliers, 3,000 cables, 40,000 bolts, and two kilometers of hosing.

The carbon dioxide laser starts at a few watts and passes through four amplifiers to reach 20,000 watts, four times stronger than lasers that cut steel. The reticle moves back and forth at accelerations over 20 Gs, five times a Formula 1 car, to print 185 wafers per hour. The overlay accuracy, meaning how precisely one chip layer aligns to another, must be within one nanometer. That is five silicon atoms. The mirror pointing accuracy is measured in picoradians. As one engineer demonstrated, if you put a laser on the mirror and a dime on the Moon, you could choose which side of the dime to point at.

Key Takeaways

ASML's EUV lithography machine is arguably the most consequential piece of technology on Earth. Every advanced smartphone, data center chip, and AI processor depends on it. It took over 30 years to develop, from Kinoshita's first dismissed presentation in 1986 to commercial production in 2016. Multiple teams across Japan, America, the Netherlands, and Germany contributed breakthroughs in multilayer mirrors, tin droplet plasma sources, hydrogen cleaning systems, and mirror precision.

The story is ultimately about unreasonable people. As the video concludes with a quote from George Bernard Shaw, "The reasonable man adapts himself to the world. The unreasonable one persists in trying to adapt the world to himself. Therefore, all progress depends on the unreasonable man." Kinoshita, Hawryluk, van den Brink, and the thousands of engineers at ASML refused to accept that EUV was impossible, even when the entire industry laughed at them. And now their machine makes the modern world possible.