IBM, Mitsubishi Chemical, and Keio Train Generative Quantum Eigensolvers on 70-Qubit Hardware
quantumcomputingreport.com Nov 12, 2026

IBM, Mitsubishi Chemical, and Keio Train Generative Quantum Eigensolvers on 70-Qubit Hardware

AI-summarised brief · reviewed before publication

A consortium of IBM Quantum, Mitsubishi Chemical, Keio University, and NVIDIA trained autoregressive quantum circuit generators on IBM’s 156‑qubit Heron processor, creating 70‑qubit generative quantum eigensolver (GQE) policies. The policies, trained via a two‑stage reinforcement‑learning curriculum, were applied zero‑shot to simulate organotin photoresist molecules for extreme‑ultraviolet lithography. The approach uses Sample‑based Quantum Diagonalization to sample key bitstrings, after which classical processors reconstruct the active subspace and diagonalize the Hamiltonian. The work demonstrates scalable quantum‑classical hybrid training on large‑scale hardware.

💡 Why It Matters

  • · The study shows that deep quantum circuits can be trained directly on noisy, large‑scale hardware, bypassing the gate‑depth limits of traditional ansätze.
  • · This paves the way for practical quantum simulations of complex molecules that are intractable for classical methods.