Title: Error-robust quantum logic optimization using a cloud quantum computer interface

Abstract: We describe an experimental effort designing and deploying error-robust single-qubit operations using a cloud-based quantum computer and analog-layer programming access. We design numerically-optimized pulses that implement target operations and exhibit robustness to various error processes including dephasing noise, instabilities in control amplitudes, and crosstalk. Pulse optimization is performed using a flexible optimization package incorporating a device model and physically-relevant constraints (e.g. bandwidth limits on the transmission lines of the dilution refrigerator housing IBM Quantum hardware). We present techniques for conversion and calibration of physical Hamiltonian definitions to pulse waveforms programmed via Qiskit Pulse and compare performance against hardware default DRAG pulses on a five-qubit device. Experimental measurements reveal default DRAG pulses exhibit coherent errors an order of magnitude larger than tabulated randomized-benchmarking measurements; solutions designed to be robust against these errors outperform hardware-default pulses for all qubits across multiple metrics. Experimental measurements demonstrate performance enhancements up to: $\sim10\times$ single-qubit gate coherent-error reduction; $\sim5\times$ average coherent-error reduction across a five qubit system; $\sim10\times$ increase in calibration window to one week of valid pulse calibration; $\sim12\times$ reduction gate-error variability across qubits and over time; and up to $\sim9\times$ reduction in single-qubit gate error (including crosstalk) in the presence of fully parallelized operations. Randomized benchmarking reveals error rates for Clifford gates constructed from optimized pulses consistent with tabulated $T_{1}$ limits, and demonstrates a narrowing of the distribution of outcomes over randomizations associated with suppression of coherent-errors.