How microchips are becoming more and more powerful – and making artificial intelligence even better

Optical lithography has been the key technology for the production of microchips for more than 40 years. When using deep ultraviolet (DUV) technology, this process reached its technical and economic limits. “In order to achieve the processing power required by digitalization or artificial intelligence, while also reducing energy consumption and manufacturing costs per chip during chip production, we had to think in a completely new way,” reports Peter Kürz.

The researchers found the solution in the wavelength of light. ZEISS pioneer Ernst Abbe had already observed in the 19th century that the shorter the wavelength of light in an image, the better the resolution. For microchip manufacturing, this means the shorter the wavelength, the finer the structures on the wafer. But the wavelengths could not be shortened much further with DUV technology. So a new solution had to be found. With the help of EUV technology, it was finally possible to reduce the wavelength by a factor of 14 – from 193 to 13.5 nanometers. Structures 5,000 times thinner than a human hair can be imaged on a wafer using this technology.

What now sounds so simple in theory is high technology in practice. “We’re talking about the world’s most precise mirrors and mechatronic systems that keep these mirrors extremely accurate. Scientifically, but also technically, there are enormous challenges,” Kürz explains. The EUV light creates a plasma that is 40 times hotter than the surface of the sun.

All these technologies are brought together in a complex machine – weighing a good 180 tonnes, the size of a school bus, and consisting of more than 100,000 individual parts. “This is among the most sophisticated machines ever built,” Kürz says. An innovation of this magnitude therefore requires a good team – and strong partners. The laser in the EUV source, for example, comes from the company Trumpf. The machine and EUV light source were developed by ASML – the world’s only manufacturer of EUV lithography machines. Together with ZEISS, it is responsible for the production of 80 percent of all microchips worldwide.

Artificial intelligence played a dual role here. On the one hand, AI is used, and on the other, AI is enabled. “Artificial intelligence helps us to recognize when we need to replace components in our production machines,” Kürz explains. But he sees the bigger connection in the role of ASML and ZEISS as enablers of digitalization, and in particular of this new technology, artificial intelligence. “AI requires vast amounts of data and enormous processing power. So this means more and more powerful microchips are needed. And we make this possible,” he says. He sees digitalization as the basis of technological progress. “This development makes changes possible that were thought to be impossible,” says the expert. He is driven by the need to push the boundaries of what is possible. AI enables this.

Instead of being satisfied with the results, the 57-year-old is already working with his team on the next stage of chip production development, i.e. a new generation of chips. The resolution of the optics should be improved even further with the help of high-NA-EUV optics, which offer a larger aperture angle. “If we increase the aperture angle of the optics – the numerical aperture (NA) – we can further increase the transistor density on the microchips, making the chip even more powerful,” Kürz explains. So things are looking good for the future of artificial intelligence and semiconductors. And Gordon Moore can also be pleased because it appears his law will be relevant for at least the next decade.