China’s chipmakers have shown that EUV lithography isn’t the only route to advanced processors. Their progress with older DUV equipment is real; so is the price of extending it. Multiple patterning can bring a chip’s features closer together, but each added step increases manufacturing time and the risk of defects. The central question is no longer simply whether DUV can make advanced chips. It’s how much inefficiency manufacturers can absorb before the economics stop working.
ASML’s extreme ultraviolet systems use light with a 13.5-nanometer wavelength to print fine features on silicon. Sales of those machines are restricted for China, leaving its chip industry to rely on deep ultraviolet tools. Some analysts estimate China trails Western rivals in EUV technology by at least five years. Yet the equipment gap hasn’t prevented Chinese manufacturers from making chips at increasingly advanced densities.
Teardowns of Huawei’s Mate 60 Pro in 2023 identified its Kirin 9000S as a product of SMIC’s second-generation 7nm-class process, made without EUV. More recent analysis of Huawei’s Kirin 9030 points to further progress: SMIC’s DUV-based process approaches the density associated with 5nm manufacturing. It still falls short of leading processes from TSMC and Samsung, but the direction is clear.
More exposures, more trade-offs
The method behind that progress is multiple patterning. Rather than print a complex layer in one pass, manufacturers split it into simpler patterns and expose the wafer several times. Combined, those patterns can approximate a layer that a single EUV exposure might produce.
Chris Mack, co-founder and chief technology officer of semiconductor firm Fractilia, says the economics change as the number of exposures rises. Early industry estimates put three exposures using 193-nanometer immersion lithography at roughly the same cost as one EUV exposure. “At three, it was about even,” Mack told Tom’s Hardware Premium. “But as soon as you hit four masks – 193 masks being replaced by one EUV mask – then EUV became more cost-effective.”
Every exposure also has to align accurately with patterns already on the wafer. More exposures mean more opportunities for misalignment and process variation to accumulate. “If you have, say, four 193 masks, you've got all those different overlays, and you've got all these different process variations that start to accumulate,” Mack said.
Manufacturers can reduce some of that risk by leaving more space between features. That makes production more forgiving, but it also makes each chip die larger. A wafer holds fewer large dies, and wafers can cost tens of thousands of dollars. The business depends on how many saleable chips come off each one. “You don't sell wafers, you sell die,” Mack said. “The question is: how many die can I fit onto that wafer, and how many of those die yield, and therefore how many chips can I sell per wafer?”
That distinction complicates claims that a processor has reached a particular process “node.” Node labels don’t offer a consistent measure of chip density across manufacturers. “If somebody says it's X node – the seven nanometer node, the five nanometer node, the three nanometer node – these words almost have no meaning,” Mack said. SMIC’s N+3 process, for example, is reported to represent meaningful progress toward 5nm-class manufacturing, while measurements indicate it remains less scaled than established 5nm processes from Samsung and TSMC. SMIC’s latest chips show how technical progress and manufacturing parity can remain separate questions.
Memory offers a different test
DRAM provides evidence that DUV can remain useful at advanced process generations. Micron produced competitive memory with DUV, multiple patterning and tight process control. Its 1-alpha DRAM generation was developed without EUV. Mack said Micron pushed 193-nanometer immersion lithography further than other memory makers, producing competitive chips while Samsung and SK Hynix used EUV on multiple layers.
That experience informs expectations for Chinese memory maker CXMT. Morgan Stanley has estimated that CXMT’s planned move to 1-alpha DRAM, intended for HBM3e, could be achievable without EUV. Micron’s results support the possibility, though memory’s regular, repeating structures make it easier to coordinate design and manufacturing than the intricate logic inside phone and computer processors. Mack put it plainly: “Memory is a lot easier than logic.”
There’s also a limit to what Micron’s example proves. The company has since adopted EUV for its newer 1-gamma DRAM process. Micron says that process delivers more than 30% greater bit density per wafer than its preceding 1-beta generation. The shift suggests DUV can carry a manufacturer far, while EUV becomes attractive as the cost of extending older tools grows.
EUV itself brings process challenges. Mack points to stochastic effects: small, random variations tied to the limited number of photons in an exposure. Increasing the EUV dose can improve pattern control, but that raises costs. No lithography approach removes every compromise; the practical choice depends on the density, yield and expense a manufacturer can tolerate.
Capacity can narrow the gap, not erase it
China’s manufacturers can continue extending DUV through more patterning, design choices and investment in production capacity. Those measures can increase chip supply even when each wafer yields fewer usable dies or takes longer to process. Morgan Stanley estimates Chinese DRAM maker CXMT may produce about one-third as many bits per wafer as Western competitors. Its analysts attributed only around a third of that difference to the older process node, while describing lithography access as a “hard ceiling” for Chinese memory development.
Douglas Fuller, a researcher at Copenhagen Business School, told Tom’s Hardware Premium that advanced logic might be engineered around EUV in theory, but the alternatives are unlikely to be efficient. He also sees room for DUV in memory, arguing that advanced packaging could let manufacturers use it for later DRAM generations by the end of the decade.
More capacity can offset some production limits. Morgan Stanley’s estimates assume CXMT remains on 1-alpha without EUV, yet grows its share of global DRAM bit supply from 14% in 2026 to 16% in 2028 by adding new factories. The bank also projects China’s AI chip self-sufficiency rising from 42% in 2025 to 70% by 2030, supported by new advanced-node capacity and improved yields. It expects gaps in EUV, high-bandwidth memory and advanced software to persist.
That’s the distinction to watch: China can make meaningful gains in chip capability and output without matching the economics of leading-edge manufacturing. DUV remains a tool for progress, especially when designs suit its limits and production scales up. But Mack doubts that multiple patterning can make the most advanced nodes commercially viable on its own. “I would be hard pressed to really believe that a three nanometer or a two nanometer node is going to be economically viable without EUV,” he said. The next measure of China’s progress will be not only which chips it can produce, but how efficiently it can turn them into saleable volume.