Zeiss lenses are doing the printing inside China's flagship domestic DUV machine

An SMIC capacity planner deciding whether next quarter's wafers go through a tool nobody has run at volume faces a choice shaped less by the absence of a Chinese prototype than by the absence of anyone willing to eat the cost when a first-of-kind tool takes a line down. Lin Qingyuan, a semiconductor analyst at Bernstein, told the Financial Times that Huawei is playing a project management role in the DUV push, and that a prototype on its own achieves nothing without thousands of vendors and customers moving together under a central orchestrating power ^[2]. Huawei is investing across the supplier base and helping those suppliers get machines into Chinese fabs, SMIC among them ^[3].

What is being pitched is a domestic DUV machine, but what is being funded is the underwriting of the adoption risk that kept fabs on imported tools.

The optics replacement is funded at a scale worth reading carefully. Huawei's venture arm Habo and a second Huawei-backed fund have both put money into Fujian Zhiqi Photonics, an ultra-precision optics maker spun out of the Shenzhen-listed group Castech in 2022, with the fund taking a 37% stake for about $15m in 2023 ^[8]. Divide $15m by 0.37 and the implied value of the whole company is roughly $41m ^[1]. Castech, which traces its origins to a Chinese Academy of Sciences institute and makes optical crystals for laser systems, has disclosed that both ASML and Huawei are customers ^[9].

Optics is one of two gaps. A DUV machine carries tens of thousands of parts across roughly a dozen major component groups, and China lags in two of them ^[10]. The other is the light source, a market dominated by Japan's Gigaphoton and by Cymer, which ASML owns ^[11]. Lin's description of the difficulty is the one an operator recognises: building a light source is easy, but having one that works very stably at such high power is difficult ^[11]. Beijing RSLaser Opto-Electronics, another Habo investment, has supplied sources for older Shanghai Micro Electronics Equipment machines, and Chinese media report it is now working with SMEE on a source for a 28nm tool ^[12].

Time-to-value is where the arithmetic stops flattering the program. SMIC has been testing a 28nm Yuliangsheng tool and pairs it with multi-patterning, running each wafer through the machine repeatedly, to reach 7nm chips ^[14]. A fab buying throughput therefore gets less throughput per tool than the specification suggests, on machines Lin says will take several years to approach ASML productivity ^[16].

Any substitution plan an operator is handed comes down to which line on the bill of materials still comes from the party being substituted away from, and what the funded replacement for that line is actually worth. Here the answers are the projection lenses and about $41m of implied value, set against a dozen machines shipping this year. The program has a project manager and a schedule, but the lens grinder still owns the critical path.