The world’s most complex machine - Works in Progress Magazine

The phones we carry around in our pockets have two million times more memory and are thousands of times faster than the room-sized computers that guided the Apollo mission to the Moon. This incredible shrinking act has been driven by our ability to make transistors smaller and smaller.Each transistor is a microscopic switch that can alternate between a one and a zero, the basic language of all computing. Billions are packed onto tiny silicon chips called semiconductors. The more transistors that fit onto a chip, the more logic and memory circuits it holds, and the more it can do. Get the print magazineSubscribe for $100 to receive six beautiful issues per year.SubscribeAdvanced semiconductors are, arguably, the most important technology in the world. Over the last five years, they have even emerged as a geopolitical flashpoint between the US and China. But for all this rivalry, any country or company that hopes to manufacture semiconductors is dependent on a single firm: ASML. Dubbed ‘a relatively obscure Dutch company’ by the BBC in 2020, ASML makes the only machines in the world capable of stenciling the transistors onto chips with the precision necessary to fit billions on a 30-centimeter wafer.These machines are roughly the size of double-decker buses. To ship one requires 40 freight containers, three cargo planes, and 20 trucks. They are the world’s most complex objects. Each contains over one hundred thousand components, all of which have to be perfectly calibrated for the machine to produce light consistently at the right wavelength.While ASML is now the sole supplier of these machines, and will be for some time to come, it started out as a laggard in the chipmaking industry. Overtaking its competition required many things rarely associated with European companies: close collaboration with the American government, selling large stakes to foreign competitors, and a huge gamble on an unproven technology. Let there be lightThe key to ASML’s success is a technology called photolithography (sometimes just called lithography). The technique involves transferring a pattern onto a semiconductor wafer by exposing it to light. In the 1950s, the first chipmakers had tried to draw these patterns by hand, but anything that physically touches the wafer scratches it, dirties it, or warps the pattern.

Scientists working independently for Bell Labs and the US military realized that they could use light to print identical patterns without making physical contact with the wafer.To make chips, engineers start with a thin wafer of semiconductor material, usually silicon. This wafer is coated with a chemical called photoresist, which reacts when exposed to light. In photolithography, light is projected through a detailed pattern onto the photoresist-coated wafer, softening the exposed areas. The wafer is washed to remove any softened areas, revealing the silicon underneath. It is then moved to an etching machine that blasts it with charged chlorine or bromine gas, carving the desired pattern into the exposed silicon. These features are later filled with metal, such as tungsten and copper, to connect the transistor to power. These etched layers then combine into an intricate network of transistors. Over time, the semiconductor manufacturing ecosystem has developed increasingly sophisticated etching using ever smaller wavelengths of light. Smaller wavelengths diffract less, allowing the light to travel in straighter lines and print sharper, tinier details without blurring. These allow for more precise pattern projections that, in turn, allow smaller and more densely packed transistors. Early lithography relied on mercury vapor lamps that were similar to streetlights, while more modern machines rely on lasers created using argon and fluorine gases. By 2010, such lasers made it possible to create a 22-nanometer feature through multiple exposures using a 193-nanometer wavelength.The most advanced version of this technology, extreme ultraviolet lithography, is used to make the very smallest chips. The smallest in 2025 were marketed as three nanometers, roughly 25,000 times thinner than a human hair. To make them, a droplet of liquid tin is released into a chamber and hit with a single pulse of light, which melts and flattens it. As the droplet continues to fall, a second, more powerful pulse vaporizes the tin, creating an extremely hot plasma that emits light at the narrow wavelengths needed for extreme ultraviolet lithography. The light beam is then concentrated by reflecting it across a series of slightly concave mirrors so flawless that, if scaled to the size of Germany, their imperfections would be measured in millimeters. Engineers need to use mirrors, rather than the glass lenses used in standard lithography, as almost all solid materials absorb light at such short wavelengths.

The light eventually hits the mask, which contains the pattern to be printed on the chip. As the pattern on the mask is usually several times larger than what is wanted on the chip, the light is then reflected by a second system of mirrors.  Path of light through an extreme ultraviolet lithography scanner. Image ASML. After the light reflects from the mask, it carries the pattern as a bundle of rays spreading out from each point. The next mirrors tip these rays inward so that, instead of spreading widely, they reunite over a shorter distance. When the rays from each point come together sooner, the picture they form is physically smaller. By repeating this with several carefully shaped mirrors, engineers shrink the pattern by a fixed amount while keeping it in focus. After being shrunk four times, it hits the wafer.The great shrinking actLonger wavelengths act like a blunt chisel, suitable for rough shaping, but they struggle to capture finer details. The longer light waves are larger relative to the tiny features on the reticle that they must reflect from. When a wave meets something smaller than itself, it naturally spreads and bends around its edges instead of casting a sharp shadow. To create the same details, the blunt chisel needs to go over the same spot a number of times (creating blurrier edges). Lithography had to take wavelengths all the way to the extreme ultraviolet range to achieve the high resolution patterning needed for cutting-edge process nodes.Wavelengths as low as 13.5 nanometers can achieve more precise patterns in a single exposure. In fact, extreme ultraviolet lithography can combine three or four photolithography patterning cycles into a single one on a seven-nanometer node. Without EUV, producing five-nanometer nodes might require as many as one hundred different steps. Extreme ultraviolet lithography was able to produce more accurate patterns on wafers than older techniques even if they were used multiple times.Today, ASML dominates the overall market for lithography and has an effective monopoly in extreme ultraviolet lithography. Its EUV machines sell for more than $120 million. With a market capitalization of over $400 billion, ASML is one of Europe’s most valuable companies. But it wasn’t always like this.OriginsASML started off life within Philips, the Dutch consumer electronics giant.

During the 1970s, Philips had roughly 20 percent of the global electronics market and was a major chipmaker. In this era, lithography machines used wavelengths of over 400 nanometers to pattern 1,000-nanometer features. The industry struggled to shrink features without losing accuracy or letting dust and flaws creep in. Philips began to work on its own prototype, drawing on its expertise in optics and precision mechanics. By the early 1980s, the project was running into trouble. The company was looking to cut costs and engineers estimated that they would need over $280 million in today’s money to finish the machine’s development and production.In 1984, Philips spun out Advanced Semiconductor Materials Lithography (which later dropped the full name in favor of its acronym) as a joint venture with ASM International, a Dutch conglomerate that sold equipment to the semiconductor industry. The business originally struggled. It had no market share and no brand recognition. Its first product, the PAS 2000, was a commercial failure. The machine used oil pressure, like that in power steering, to move the table that held the wafer during exposure, rather than electric motors. This made it smooth and precise, but it was prone to leaking. At the first conference ASML attended, one industry executive told them: ‘The race has already been run. There’s no room for you here.’ ASML switched back to electric motors.The company took an unusual approach from the outset. While Japanese giants Nikon and Canon were vertically integrated, ASML outsourced key components like optics and motors so that it could focus on assembling and optimizing the final machine. Given this outsourcing, it made sense for ASML to embrace a modular design with clearly defined subsystems. This approach was mocked in European manufacturing circles. German engineers warned ASML’s leadership that they were ‘asking for trouble’ and would ‘lose all control’ if they didn’t make critical components themselves. But ASML had no choice: it lacked the capital, expertise, and time to build these subsystems from scratch.By 1988, ASML was on the verge of collapse. ASM International had already pulled out, and Philips considered shutting it down. It was saved by a single Philips board member, Gerd Lorenz, who was particularly worried about Europe’s growing dependence on Asia for strategic technology.

Lorenz argued that Europe needed a stake in chip manufacturing. This was enough to convince Philips to give ASML more time, but didn’t fix its fundamental problem: it was still an inferior supplier with no competitive edge.ASML used the time it was given to develop the PAS 5500, released in 1991 and the company’s first commercial breakout. While Nikon’s contemporary photolithography system was more precise, ASML’s modular design meant that machines could be fixed quickly on site. This reduced downtime and, by making it easy to replace parts when they broke, it was possible to extend the machine’s life. This was a key factor that led John Kelly, IBM’s director of semiconductor R&D, to push IBM to order the PAS 5500 over the Japanese machines. ASML had gone global. The first breakthroughsASML’s success depended on two projects in the late 1990s and 2000s that gave it a huge advantage in research and development. The first was a public-private partnership, started in 1997, called the Extreme Ultraviolet Limited Liability Company. The Extreme Ultraviolet Limited Liability Company began life as a rescue mission. Before 1997, basic semiconductor research was carried out in a small handful of research labs, all dependent on government grants.The original program for EUV research was a ‘virtual national lab’ that combined Lawrence Livermore National Laboratory, Sandia National Laboratories, and the Lawrence Berkeley National Laboratory. Each covered a different component: Livermore focused on mirrors and optics, Sandia on the light source and systems engineering, and Berkeley on advanced equipment for testing. But in 1996, Department of Energy budget cuts had placed the virtual national lab program on the chopping block. Intel, then the undisputed world leader in microprocessors, was keen to preserve the work and spearheaded the creation of the Extreme Ultraviolet Limited Liability Company, the largest public-private partnership of its kind in the history of the US Department of Energy. During its six-year life, the company invested over $270 million into extreme ultraviolet lithography development, funded by the sale of shares to member companies, giving them a right of first refusal to purchase the photolithography tools being produced.The company initially restricted membership to American firms.