MAViS: Modular Automated gate-virtualization of two-dimensional semiconductor quantum dot arrays.

Arrays of gate-defined semiconductor quantum dots are promising candidates for scalable quantum processors, where qubit operations—initialization, manipulation, and readout—are controlled via metallic gates. The capacitive crosstalk between these gates hinders independent tuning of every quantum dot array parameter, affecting high-fidelity qubit control. Although virtual gates alleviate control crosstalk, determining them efficiently and accurately as the devices grow in size and complexity becomes challenging. To overcome these, we propose MAViS — a Modular Automated Virtualization System, for autonomously constructing a complete stack of virtual gates in real-time. Our method exploits machine-learning techniques to rapidly extract from two-dimensional charge stability diagrams features necessary for virtualizing plunger and barrier gates in the low- and high-tunnel coupled regimes. Subsequently, computer vision and regression models help determine all capacitive couplings self-consistently. Through MAViS, we successfully virtualize a dense ten-dot 2D quantum-dot array in germanium and discover that in the strong-tunnel coupling regime, linear virtual barrier gates provide insufficient compensation, necessitating quadratic corrections. Furthermore, our method also allows autonomous estimation of charge noise, measures of disorder, and calibration of quantum-dot arrays in real-time.