The Pulse
IonQ’s AI-Generated Circuits Hold Circuit-Finding Time Near 28 Seconds
IonQ said researchers from Oak Ridge National Laboratory, NVIDIA, and the University of Tennessee, Knoxville trained a generative model to write quantum optimization circuits directly, keeping circuit-finding time nearly constant as benchma
AI.info Team ·
IonQ’s DQAOA-GPT Replaces the Tuning Loop
IonQ said on September 16 that researchers from Oak Ridge National Laboratory, NVIDIA, and the University of Tennessee, Knoxville have trained a generative model to write quantum optimization circuits directly. The work, presented at IEEE Quantum Week in Toronto, targets one of the main costs in hybrid quantum optimization: repeatedly running a circuit, measuring its output, changing parameters, and running it again.
The research introduces DQAOA-GPT, a hybrid system that combines distributed quantum approximate optimization with GPT-based circuit generation. Rather than searching for circuit parameters through an iterative variational process, the model produces candidate circuits for smaller pieces of a larger optimization problem.
“Better answers in hybrid quantum optimization have traditionally come with a steep tuning tax. In this benchmark, generative AI replaced the iterative tuning loop, and as the quantum subproblems grew the solution quality improved. Take that cost away and you can work at the size where the answer is meaningful. The result provides a potential path toward scaling hybrid quantum optimization, unlocking completely new capabilities and scales that align with IonQ’s existing and future quantum computing hardware generations,” said Dr. Martin Roetteler, IonQ’s Vice President of Quantum Applications R&D and a co-author of the paper.
100 Variables, Ten Candidates Per Subproblem
The team tested the approach on dense higher-order unconstrained binary optimization problems with as many as 100 decision variables. Distributed quantum optimization breaks those larger problems into smaller subproblems, solves them separately, and combines the results into a global solution.
Training began with conventional optimization runs across many sampled problems. Researchers retained near-optimal circuits and used those examples to train a transformer model on circuit instructions rather than text. During inference, DQAOA-GPT generated ten candidate circuits for each subproblem; researchers simulated and scored all ten, then used the best candidate to update the overall solution.
IonQ said model-generated answer quality roughly doubled as subproblems grew on the 100-variable benchmark. The announcement does not provide the specific accuracy figures cited in some accounts.
Circuit-Finding Time Stayed Near 28 Seconds in Simulation
The largest difference appeared in circuit-finding time. With the prior state-of-the-art method, the time required to find circuits increased from about 34 seconds for four qubits to more than 11 minutes for 12 qubits. DQAOA-GPT stayed near 28 seconds across the tested sizes.
Both methods ran on the same infrastructure, allowing the researchers to compare circuit-generation workflows rather than quantum computing with classical optimization. The experiments used NVIDIA’s cuQuantum software through CUDA-Q on a single NVIDIA H200 GPU in the Oak Ridge Leadership Computing Facility’s Defiant2 system.
The reported result is a comparison of two quantum optimization workflows. It does not establish that the generative method beats classical optimization, and the circuits were simulated rather than executed on quantum hardware.
Benchmark Results, Not Hardware Demonstration
That distinction limits what the September 16 announcement establishes. The study validates the circuit-generation method at benchmark scale, but it does not show the circuits running on an IonQ processor or demonstrate an advantage over established classical solvers.
The preprint, submitted to arXiv on July 22, describes DQAOA-GPT as a framework for hybrid high-performance-computing and quantum-computing environments. Its authors say larger GPU resources and parallel execution could support bigger combinatorial optimization problems, but those applications have not yet been demonstrated in the paper.
ORNL Plans Real-World Tests
ORNL led the study, with authors from its National Center for Computational Sciences and Materials Science and Technology Division, IonQ, NVIDIA, and the University of Tennessee. The research paper lists Seongmin Kim, Abhinav Rijal, Yuri Alexeev, Nora Bauer, Martin Roetteler, Mina Yoon, George Siopsis, and In-Saeng Suh as authors.
“AI can become a new computational layer for quantum circuit synthesis, enabling the automatic design and optimization of quantum circuits for increasingly complex problems. We are now extending the framework to real-world scientific and engineering applications and scaling it across larger HPC systems to address problems of even greater scale and complexity,” said Dr. In-Saeng Suh and Dr. Seongmin Kim, National Center for Computational Sciences, ORNL.
For now, the concrete result is narrower: a transformer trained on strong example circuits reduced the growth of circuit-generation time in a simulated 100-variable optimization benchmark. The paper and its code-facing framework are available in the DQAOA-GPT preprint, while IonQ’s announcement gives the collaboration’s measured runtime comparison.