Automated in situ optimization and disorder mitigation in a quantum device

We investigate automated in situ optimization of the potential landscape in a quantum point contact device, using a $3 \times 3$ gate array patterned atop the constriction. Optimization is performed using the covariance matrix adaptation evolutionary strategy, for which we introduce a metric for how “step-like” the conductance is as the channel becomes constricted. We first perform the optimization of the gate voltages in a tight-binding simulation and show how such in situ tuning can be used to mitigate a random disorder potential. The optimization is then performed in a physical device in experiment, where we also observe a marked improvement in the quantization of the conductance resulting from the optimization procedure.

💡 Research Summary

This paper presents a machine learning-based methodology for the automated in situ optimization of quantum devices, specifically targeting the enhancement of conductance quantization and the mitigation of disorder effects. The study focuses on a Quantum Point Contact (QPC) device defined in a GaAs-based two-dimensional electron gas (2DEG). The unique aspect of the device is a 3x3 array of square top-gates (“pixel gates”) patterned over the constriction, flanked by two outer split-gates. This configuration provides fine-grained control over the electrostatic potential landscape within the QPC channel.

The core optimization is performed using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), a derivative-free, stochastic optimization algorithm well-suited for noisy experimental landscapes. A key innovation is the design of a custom loss function that quantifies the “step-like” quality of the conductance trace. This function rewards large jumps between plateaus (using a cube root of conductance differences) and penalizes a conductance trace that is either too smooth or where the channel is completely pinched off. Crucially, the loss function contains no prior information about the expected quantized conductance values (e.g., integer multiples of 2e²/h), allowing the physics of the QPC to dictate the solution rather than enforcing a pre-defined pattern.

The methodology is first validated through tight-binding simulations using the KWANT software. The potential from the gate voltages is calculated electrostatically. The optimization parameters are not the individual pixel gate voltages but their Fourier modes, which correspond to broader shapes of the potential landscape, making the optimization more robust. The simulations demonstrate two key results: 1) In a disorder-free system, the optimized gate configuration produces significantly sharper conductance steps compared to a simple sweep of the average gate voltage. 2) When a spatially correlated random disorder potential is introduced—strong enough to smear out the conductance staircase—the CMA-ES algorithm can find a new gate configuration that effectively mitigates the disorder and restores clear conductance plateaus. The optimized solutions consistently yield conductance plateaus at integer values without this being explicitly encoded, confirming the algorithm discovers physically valid configurations.

The study then successfully transitions from simulation to experiment. The same optimization algorithm is deployed on a physical device with an identical gate design fabricated on a GaAs heterostructure. The algorithm operates in a fully automated closed loop: it suggests gate voltage configurations, measures the resulting conductance, evaluates the loss, and updates its search strategy—all without human intervention. The experimental results show a marked improvement in the quantization of the conductance after optimization, directly demonstrating in situ mitigation of device-specific, unknown disorder inherent to the material and fabrication process.

In conclusion, this work establishes a general and practical framework for using evolutionary algorithms like CMA-ES to autonomously tune complex quantum devices. It proves that such an approach can not only improve device performance beyond manual tuning but also actively compensate for unpredictable disorder, a major hurdle in semiconductor quantum technologies. This paves the way for automated calibration and optimization of larger-scale quantum systems, such as quantum dot arrays and spin qubit processors, enhancing reproducibility and performance.