Breaking the EUV Barrier

As companies around the world invest heavily in developing AI infrastructure, the demand for increasingly powerful semiconductors has intensified too. Behind those chips, sits a technology that has become one of the semiconductor industry’s most important strategic bottlenecks—and yet which remains virtually unheard of by the general public: extreme ultraviolet (EUV) lithography. 

Following his most recent research paper on the topic, we spoke to Eric Benoist, Tech & Data Research Specialist at Natixis CIB, to understand more about the race to compete—and how technological and geopolitical implications are influencing progress.

The most advanced AI processors depend on cutting-edge semiconductor manufacturing, where packing smaller, denser features onto silicon wafers is critical to boosting performance and efficiency. EUV lithography is absolutely central to that process and has become an important part of the global competition over AI hardware.

Today, Dutch giant ASML holds a de facto monopoly on this technology. It is the sole manufacturer capable of building and maintaining the incredibly complex machines that global foundries rely on to print nanometer-scale processors.

But because of strict US and Dutch export controls—implemented on national security grounds—China has never had access to the EUV market, which has had a tremendous impact on its ability to deploy viable AI infrastructure at home.

Can China make advanced chips without EUV? The answer is yes, up to a point.

To bypass the lack of EUV, Chinese manufacturers have relied on deep ultraviolet (DUV) lithography—the generation directly below EUV—paired with increasingly complex multi-patterning techniques.

The basic idea is relatively straightforward: instead of trying to print the smallest features in a single exposure, the pattern is built through several deposition and etching steps. This workaround has unlocked significant progress for China, particularly in smartphone processors.

But for chips with a much larger physical footprint—such as AI accelerators—each additional manufacturing step adds cost and complexity, creating more opportunities for fatal defects.

Ultimately, DUV-based manufacturing is a band-aid—and one that is becoming far too expensive and unreliable for high-end AI logic.

There have been unconfirmed reports of a prototype EUV machine being completed in a high-security facility in Shenzhen in early 2025—a system allegedly capable of generating EUV light, though whether it can actually produce working silicon remains entirely unproven.

While Chinese sources are targeting mass production by 2028, the lack of independently verifiable information leaves industry experts highly skeptical.

Sourcing the talent is not even the issue. China has assembled teams of brilliant scientists and former ASML engineers to work on various EUV-emitting laser technologies. But according to academic evidence, none of these experimental systems can yet match the sheer power output of ASML’s machines.

Despite what we’ve just discussed, no. EUV optics are extraordinarily demanding. The system relies on a complex network of about ten mirrors, each assembled with atomic precision just to make EUV light reflection possible. At this scale, the slightest imperfection ruins everything. To put this in perspective: if one of these mirrors were enlarged to the size of Germany, its greatest surface imperfection would stand a tenth of a millimeter high!

There is simply no evidence that China has yet replicated the optical column required for commercial production.

Photoresists—the light-sensitive chemical coatings used to stencil circuit patterns onto the silicon—are another major bottleneck. Japan dominates this high-end EUV market, where the materials require extreme purity and tightly controlled formulations.

The important point is that producing EUV light is only one piece in the puzzle. A commercially viable EUV system requires much more effort to operate reliably at the throughput required by a modern semiconductor fab.

This may ultimately be the more interesting question, as there are several alternative approaches being explored. Electron-beam lithography, for example, can achieve excellent resolution without the complex optical projection system used by EUV. The problem is throughput. Electron beams write patterns serially, making them far too slow for leading-edge, high-volume wafer production.

Nanoimprint lithography takes a completely different approach. Instead of projecting a pattern onto the wafer, a physical template is pressed into the resist.

Canon has already commercialised the technology, while Chinese company Prinano has reported sub-10-nanometre features across 300-millimetre wafers.

There are still major questions around defects, contamination, template wear and overlay accuracy. But nanoimprint could have useful applications in areas such as photonics and optical silicon, where the manufacturing requirements are different from those of leading-edge logic.

X-ray lithography remains highly speculative, but potentially the most disruptive option.

X-rays have shorter wavelengths than EUV, meaning they could theoretically achieve even finer resolution.

The idea itself is not new. It was extensively investigated by IBM in the 1980s and ’90s, but optical lithography improved so quickly that the industry ultimately chose to pursue EUV instead. Today, the X-ray approach is seeing a resurgence. In the US, a well-funded startup named Substrate is currently leading the charge—and while there is no public evidence of a Chinese equivalent, this silence should not be mistaken for inactivity.

The technical challenges remain formidable. Powerful and stable X-ray sources are difficult to produce and involve large, unpractical particle accelerators, while the required optics and mask technologies are also demanding.

But if those problems can be overcome, the implications are significant. A successful X-ray approach would change the economics of advanced chip manufacturing and weaken the importance of today’s EUV infrastructure.

In the immediate future, no, a shift is unlikely. China’s domestic EUV efforts remain at the prototype stage and the alternative technologies being explored each have substantial limitations. None currently offers a credible replacement for EUV in high-volume leading-edge logic production.

But that does not mean the current landscape can be taken for granted. China is pursuing several routes simultaneously and has a proven track record of rapidly closing the gap with Western competitors.
While many observers bet that China is at least 10 to 15 years away from a commercial EUV breakthrough, we wouldn’t be surprised to see it happen much sooner and we do not think the market is ready for it…