A light source larger than a London double-decker bus is breaking records and redefining what is possible in computer chip manufacturing. Using extreme ultraviolet (EUV) light, the system can etch structures on silicon wafers as narrow as eight nanometres—believed to be the smallest ever achieved in a single step by a commercial chip-patterning machine. According to its manufacturer, ASML, the new technology can produce chips with nearly three times as many transistors as previous generations, a leap with profound implications for AI-powered data centers and next-generation computing.
The process works by projecting EUV light through a patterned mask onto a wafer coated with light-sensitive chemicals. The light hardens the chemicals in precise shapes, which are then etched to form transistors and ultra-fine wiring. This cycle is repeated to create all the components of a modern chip. While EUV lithography is not new, ASML’s record-setting device features enhanced optics capable of unprecedented precision, allowing engineers to pack smaller transistors into the same chip area without increasing power consumption.
The breakthrough, reported in Nature, was presented at the SPIE Advanced Lithography + Patterning conference in San Jose, California. The first machines, costing roughly $400 million each, have been delivered to major chipmakers including Intel and SK hynix. Maarten Voncken, ASML’s head of research metrology, emphasized the urgent need driving the innovation: “The demands we see are monumental in the number of chips that are needed and the scaling that is needed,” he said, citing the explosive growth of AI workloads.
For decades, the semiconductor industry has relied on Moore’s Law—the principle that transistor counts double roughly every two years—to guide development. But as AI and data-intensive applications expand, keeping up with Moore’s Law has become increasingly challenging. Smaller features on a chip mean more transistors in the same area, boosting computing power without dramatically increasing energy consumption. Achieving this requires both shorter wavelengths of light and sophisticated optical systems capable of directing it precisely.
The EUV light generated by ASML’s machines has a wavelength of just 13.5 nanometres, far below the 193 nm deep-ultraviolet light used in earlier decades. Such short wavelengths are difficult to manage, as they are absorbed by almost everything, including air and conventional lenses. Only mirrors made from nanometre-thin layers of silicon and molybdenum, produced by Zeiss in Germany, can reflect EUV light with sufficient accuracy. Even atomic-scale imperfections can degrade the image, highlighting the precision required in the system.
The latest EUV machine features larger and slightly re-shaped mirrors, along with a higher numerical aperture—a measure of light-handling capability—of 0.55 compared to 0.33 in older models. These enhancements allowed the creation of the smallest structures ever tested by Imec, an independent semiconductor research institute in Belgium. Voncken says ASML is already developing machines with a numerical aperture of 0.75, a stage branded as hyper-NA, to further shrink features. Beyond that, engineers would need to shift to X-ray wavelengths, which would necessitate a complete redesign of the lithography system.
Yet miniaturization has limits. Transistors smaller than today’s nodes risk leaking electrical charge, reducing performance. To continue increasing computing power, the industry is turning to three-dimensional designs. Researchers at Imec are exploring stacking transistors, placing one atop another, to multiply functionality without expanding chip footprint. However, densely packed 3D transistors generate heat that is difficult to dissipate, creating a thermal engineering challenge. Advanced cooling techniques will be essential to make these designs practical.
The stakes are enormous. AI models powering language understanding, image generation, and predictive analytics rely on enormous computational throughput. Data centers could run far more calculations without additional energy use if chips incorporate denser, smaller transistors. For technology companies racing to stay ahead in AI, the ability to mass-produce such chips represents a strategic advantage akin to building faster supercomputers overnight.
Jeffrey Bokor, an early pioneer of EUV lithography at Bell Labs, recalled that decades ago the technology “was regarded as lunatic fringe” due to the extreme difficulty of controlling EUV light. Today, the combination of powerful lasers, molten-tin plasma sources, and nanometre-precision mirrors has turned fringe science into commercial reality. The innovation has now moved from experimental laboratories to the heart of global semiconductor supply chains, bridging scientific ambition and industrial-scale production.
The breakthroughs described in Nature highlight how precision engineering and optical science are pushing the boundaries of what chips can do. As the AI revolution continues to accelerate, advances like ASML’s EUV system could redefine not just the speed and efficiency of computing, but the very trajectory of technological development in the coming decades. The industry is entering a new era where transistor density, three-dimensional stacking, and hyper-precise manufacturing will determine the winners in the global AI race.
