IBM, Mitsubishi Chemical and Keio Train AI to Design Quantum Circuits on 70-Qubit Hardware

At a glance

A research team from IBM Quantum, Mitsubishi Chemical, Keio University and NVIDIA trained an AI that builds quantum circuits on a real quantum computer. The trained AI then calculated molecules used in tin-based photoresists for extreme ultraviolet (EUV) chipmaking without further training.

Key points

1. 1

   The AI was trained directly on IBM's 156-qubit ibm_kobe processor.

2. 2

   Training covered active spaces of 46 and 70 qubits.

3. 3

   Training used 120 million hardware shots across 600 calibration jobs.

4. 4

   The trained AI was applied without retraining to 11 active-space tiers of methyltin trihydroxide molecules.

Why it matters

Calculating candidate chip materials by computer in advance could cut the cost and time of lab experiments. It is also a real example of using quantum computers for practical work.