Researchers from IonQ, Oak Ridge National Laboratory, NVIDIA, and the University of Tennessee, Knoxville have demonstrated an artificial intelligence method that dramatically reduces the cost of tuning quantum optimization circuits, the company says. The team trained a generative model to directly write these circuits, bypassing a traditionally iterative process that often required effort for each adjustment.
“Better answers in hybrid quantum optimization have traditionally come with a steep tuning tax,” said Dr. Martin Roetteler of IonQ, co-author of the study, “In this benchmark, generative AI replaced the iterative tuning loop, and as the quantum subproblems grew the solution quality improved.” This approach allows solution quality to improve as problem size increases, a counterintuitive result that could unlock new scales for hybrid quantum optimization.
AI-Driven Circuit Synthesis Reduces Quantum Optimization Costs
The conventional process of tailoring quantum circuits for hybrid optimization demanded hundreds of iterations of trial-and-error parameter tuning, a significant bottleneck researchers have now bypassed with an artificially intelligent approach. The team demonstrated that solution quality improved as the size of quantum subproblems increased, a counterintuitive outcome given that larger problems typically amplify tuning demands, according to NVIDIA. This advancement hinges on a trained generative model, initially exposed to high-performing circuits generated through the traditional trial-and-error method.
Researchers ran the conventional process across numerous sampled problems, retaining only near-optimal circuits to train a transformer model, similar to those powering large language models, but adapted for circuit design. The resulting model then produces candidate quantum circuits without repetitive parameter adjustments; in experiments, it sampled ten circuits per subproblem, selecting the best-scoring candidate to refine the overall solution.
This contrasts sharply with prior methods, where circuit-finding time on a 100-decision variable problem rose from approximately 34 seconds on 4 qubits to over 11 minutes on 12 qubits. The generative approach maintained a consistent runtime of nearly 28 seconds regardless of problem size, a substantial improvement achieved by running both approaches on identical infrastructure.
Every circuit was simulated using the NVIDIA cuQuantum SDK through the NVIDIA CUDA-Q platform on a single NVIDIA H200 GPU within the Oak Ridge Leadership Computing Facility’s Defiant2 system, ensuring a controlled comparison focused on the circuit-generation methods themselves, not quantum versus classical solvers.
NVIDIA’s involvement extends beyond providing the computational infrastructure; the company views AI-driven algorithm development as important for accelerating the path to practical quantum applications, the firm reports. “Drawing on accelerated computing and AI to make breakthroughs in quantum algorithms is one of the most promising ways to reach useful quantum applications as quickly as possible,” said Sam Stanwyck, Director, Quantum Product at NVIDIA.
The CUDA-Q platform, and tools like the recently launched cuda-quantum and cudaq-algorithms, are designed to facilitate this integration, enabling developers to build quantum algorithms architected around AI from the outset. This aligns with NVIDIA’s broader strategy of creating a robust quantum computing ecosystem, supported by partnerships with companies like Quantinuum, IQM, and IonQ, and bolstered by a portfolio of over 10 quantum-related patent families.
The research team’s success in scaling hybrid quantum optimization has significant implications for tackling previously intractable problems, the company states. “Take that cost away and you can work at the size where the answer is meaningful,” Roetteler explained. 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.
The study, validated at a benchmark scale, demonstrates the potential for AI to become a new computational layer for quantum circuit synthesis, automating the design and optimization of circuits for increasingly complex challenges. Researchers are now focused on extending this framework to real-world scientific and engineering applications, and scaling it across larger high-performance computing systems to address problems of even greater magnitude.
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.
Dr. Martin Roetteler, IonQ’s Vice President of Quantum Applications R&D and a co-author of the paper
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
