Semiconductors · materials

Pick the winning EUV photoresist before the fab

Resist candidates ranked before any were made, so only the lead goes to synthesis and wafer testing.

Headline result

4 of 4 ranked correctly

Four formulations ranked in the right order on quantum hardware, every energy difference within 0.3%

Measured on quantum hardwareClassical computation

Still from the Merlin Quantum film: a processor package under test

01 · The decision

The decision on the table

Choosing which photoresist formulation to take to the fab means synthesising and testing each candidate: weeks of lab and wafer time, and cost, per option. The published chemistry method for this class of material also needed 173 GPU-hours of training before it could run.

02 · What we did

What we did, with proprietary methods

We modelled an industrial tin-oxo resist cluster and its counter-anion variants with proprietary methods, trained the chemistry model on one GPU, and ranked the candidates on IBM quantum hardware with the protocol frozen before the run and a null control in the same run.

Four resist formulations ranked before synthesis by predicted electronic gap (lower activates more easily under EUV light). The ranking held on quantum hardware; every difference within 0.3%; protocol frozen before the run.

Accuracy that improves with scale: 74.0% of the target at 12 qubits, 90.4% at 40 qubits, with circuits kept to three entangling layers. GPU computation plus hardware sampling.

03 · What it was worth

What it was worth

Four formulations ranked in the right order on quantum hardware, every energy difference within 0.3%
Key result · Semiconductors · materials

Only the lead candidate goes to the lab

Baseline: Synthesise and test every candidate
Condition: Four formulations ranked in the right order on quantum hardware, every energy difference within 0.3%

Compute cut from days to minutes

Baseline: A published 173 GPU-hours of training
Condition: Minutes on one GPU, same challenge method

Twice the chemistry in one model

Baseline: About 20 qubits in the published work
Condition: 40 qubits, every step within chemical accuracy of its exact reference

Outcome

What happened next

The ranking was entered in the GIC 2026 Advanced Materials track: Award winner · GIC 2026 · 2nd worldwide. For a client, the next step is the one the ranking exists for: only the lead formulation goes to synthesis and wafer testing.

Decision it enabled

Only the lead candidate goes to the lab

Instead of: Synthesise and test every candidate

Verified by · Judged by an independent panel at GIC 2026; the ranking is a device result on IBM Heron.

05 · How we know

How we know

01Formulation ranking: 4 of 4 in exact order, rank correlation 1.0, every energy difference within 0.3% (IBM Heron, protocol frozen before the run).
02Three counter-anion variants ranked on 120 qubits in a single run; measured order equal to exact order.
03AI-driven quantum chemistry on the tin-oxo cluster at 40 qubits, every step within chemical accuracy of its exact reference (GPU computation plus hardware sampling).
04Award winner, 2nd place worldwide, GIC 2026 Advanced Materials track (Mitsubishi Chemical Group × AIST). A judged result; it does not imply endorsement of any product.
05Not claimed: that the material will perform in the fab, or that the material has been solved completely.

StatusMeasured on quantum hardwareClassical computationRanking measured on quantum hardware; training cost and the 40-qubit chemistry are classical GPU computation with hardware sampling.

Book a 30-minute call See this case in the app

Next story · Pharma · life sciences

Find where a protein can be drugged before the assay

Within 1.4% of exact

Pharma · life sciences

Find where a protein can be drugged before the assay

Candidate allosteric sites read from a structure file alone, with the site model validated against exact references. A method result: no drug-ranking gain is claimed.

Measured on quantum hardwareClassical computationRead the story