Xanadu and Mitsubishi Chemical take their quantum-EUV chipmaking bet to Phase 2

The process used to etch sub-nanometer features onto the silicon behind AI training chips, flagship phones, and high-performance computing now runs into a 92-electron-volt problem. In extreme ultraviolet (EUV) lithography, the photons that are supposed to draw those lines on a photoresist also knock loose a shower of secondary electrons that scatters the line they were meant to etch. Xanadu and Mitsubishi Chemical are betting a fault-tolerant quantum computer (FTQC) can simulate that blur well enough to design the next photoresist — and Canada's NRC IRAP and Japan's Strategic Innovation Promotion Program (SIP) are putting public money behind the bet in Phase 2 of their joint partnership.

EUV lithography fires 92 eV photons at a light-sensitive photoresist. As circuit widths shrink, the wavelength of the light has to shrink with them, and the current generation of steppers sits in the deep EUV band. When those photons hit the resist, they don't expose it cleanly. They trigger photoabsorption, secondary electron cascades, and Auger decay — three quantum-mechanical phenomena that blur the line and cap how small and dense features can get. Classical computers can model some of that chemistry, but not accurately enough to predict which photoresist molecule will resist the blur best. The partners are targeting a specific, well-bounded sub-problem: the 92 eV photoabsorption cross-section of candidate photoresist molecules.

The plan is to hand that sub-problem to a quantum computer. The partners are sizing the algorithm for early fault-tolerant quantum hardware — machines that can correct their own errors well enough to run long quantum simulations — and aiming for fewer than 500 physical and logical qubit allocations. Output from those FTQC algorithms feeds directly into Mitsubishi Chemical's classical multi-scale macroscopic models to predict electron blur and screen for new, blur-resistant chemistries. In practice, that would let a photoresist candidate be tested in simulation rather than spun onto a wafer in a cleanroom, a process that is slow and expensive.

That pipeline is now Phase 2, announced 2026-08-25 and co-funded by Canada's National Research Council's Industrial Research Assistance Program (NRC IRAP) and Japan's SIP. The Japanese side runs through AIST (the National Institute of Advanced Industrial Science and Technology) and G-QuAT, the Global Research and Development Center for Business by Quantum-AI Technology. Phase 1, completed earlier, validated that quantum algorithms could model the optical properties and light-matter interactions of photoresists at all. Phase 2 turns that into a production-ready computational pipeline with public money on the line.

Two things have to hold for the bet to pay off. First, the <500-qubit target is an allocation for early FTQC hardware that does not yet exist at production scale; if that hardware cannot reach 92 eV cross-section accuracy on the relevant timeline, the pipeline fails. Second, classical methods — time-dependent density functional theory, multireference perturbation theory — keep improving, and if any of them reach EUV-photoresist accuracy first, the quantum bet becomes moot. Neither has happened yet, but the bar is real.

The test is now on a calendar. HPCwire's coverage of the Phase 2 announcement notes that the pipeline is being scoped to specific photoresist chemistries rather than the open-ended materials-discovery problem. The partners are running a defined bet, with defined funding, on a defined physical problem — and a clear falsifier if the quantum side slips behind classical.