KAIST nanophotonics
이세욱 (Seuk Lee), 박상효 (Sanghyo Park)
물리학과 (Physics)
Deep Neural Network Trained Composite Pulse Sequences for Cold Atom in Optical Tweezer Array
To realize scalable quantum computing, precise quantum state control over each qubit is essential. Cold-atom quantum systems based on tweezer arrays have attracted due to their scalability. However, in these systems, the local addressing of the gate beam is limited by its spatially varying amplitude, which affects its performance. When implementing local addressing, smaller beam sizes result in larger amplitude errors, while larger beams reduce the error but increase crosstalk with neighboring atoms. This trade-off relationship is a challenge. To overcome this trade-off, we introduce a new degree of freedom by utilizing composite pulse sequences. However, the conventional composite pulses designed to correct static errors are not optimized for the dynamic errors present in the tweezer system caused by atomic motion. In order to devise composite pulses that are optimized for the dynamic errors of the system, we employed deep neural networks to find new composite pulse sequences that outperform the conventional one in terms of system fidelity. We determine the optimal beam size and corresponding fidelity for three different pulses and compare their performance. Our results show that we can implement composite gates that are robust to spatially varying amplitudes, overcoming optimization tradeoffs and increasing system fidelity.