IonQ, ORNL, NVIDIA, and UT Knoxville Advance AI-Driven Generative Quantum Circuit Synthesis

A research collaboration led by Oak Ridge National Laboratory (ORNL) alongside IonQ (NYSE: IONQ), NVIDIA, and the University of Tennessee, Knoxville (UT) has introduced DQAOA-GPT, a generative AI framework that synthesizes quantum optimization circuits directly to eliminate iterative parameter-tuning loops in distributed quantum algorithms. Presented at IEEE Quantum Week 2026 in Toronto, the paper received a Best Paper Award for demonstrating constant-time circuit synthesis for subproblem evaluations across scaling quantum domain widths.

The framework replaces the traditional trial-and-error variational loop of the Distributed Quantum Approximate Optimization Algorithm (DQAOA) with a transformer model trained on high-performing circuit profiles. For each subproblem, the generative transformer outputs candidate quantum circuits directly, which are evaluated in a fixed 10-candidate sampling step before updating global solution parameters. Benchmarked on a 100-variable higher-order unconstrained binary optimization (HUBO) problem using single NVIDIA H200 GPUs via NVIDIA CUDA-Q and the cuQuantum SDK, conventional variational circuit optimization times escalated from 34 seconds (4 qubits) to over 11 minutes (12 qubits), whereas the generative DQAOA-GPT approach maintained a constant synthesis runtime of approximately 28 seconds regardless of subproblem qubit count while doubling overall solution quality.