AI-Driven Pulse Compilation for Robust, Site-Selective Control of Atomic Qubits

In this presentation, we report a deep learning framework designed to compile composite pulses robust to amplitude fluctuations due to atomic motion in optical tweezers <span style="font-size:10.8333px">[1]. A deep neural network, trained on the interaction dynamics between a trapped atom and a focused control beam, autonomously generates robust pulses yielding an order-of-magnitude improvement in control fidelity across the arbitrary qubit rotations . We further validate the robustness of these pulses against experimental imperfections, including optical aberrations and beam misalignment. This approach establishes a generalized strategy for AI-driven pulse compilation, readily extendable to other quantum platforms limited by coherent errors, such as trapped ions and color centers.