[Paper Review] Miniaturizing neural networks for charge state autotuning in quantum dots

Stefanie Czischek, Victor Yon+9 more|arXiv (Cornell University)|Jan 8, 2021

Advanced Memory and Neural Computing46 references21 citations

TL;DR

This paper proposes ultra-compact feed-forward neural networks trained on synthetic stability diagrams to detect charge-state transition lines in quantum dot devices. The networks, with as few as 25 input neurons, achieve high accuracy in classifying transition lines and are small enough for direct implementation on existing memristor crossbar arrays, enabling low-power, on-chip autotuning for scalable quantum dot quantum computers.

ABSTRACT

A key challenge in scaling quantum computers is the calibration and control of multiple qubits. In solid-state quantum dots, the gate voltages required to stabilize quantized charges are unique for each individual qubit, resulting in a high-dimensional control parameter space that must be tuned automatically. Machine learning techniques are capable of processing high-dimensional data - provided that an appropriate training set is available - and have been successfully used for autotuning in the past. In this paper, we develop extremely small feed-forward neural networks that can be used to detect charge-state transitions in quantum dot stability diagrams. We demonstrate that these neural networks can be trained on synthetic data produced by computer simulations, and robustly transferred to the task of tuning an experimental device into a desired charge state. The neural networks required for this task are sufficiently small as to enable an implementation in existing memristor crossbar arrays in the near future. This opens up the possibility of miniaturizing powerful control elements on low-power hardware, a significant step towards on-chip autotuning in future quantum dot computers.

Motivation & Objective

• Address the challenge of high-dimensional gate voltage tuning in multi-qubit quantum dot systems.
• Develop minimal neural networks capable of detecting charge-state transition lines in stability diagrams.
• Enable on-chip implementation of autotuning by designing networks compatible with emerging memristor hardware.
• Demonstrate transferability of models trained on synthetic data to real experimental quantum dot devices.
• Minimize experimental measurement costs by covering only small regions of the stability diagram through patch-based scanning.

Proposed method

• Train small feed-forward neural networks (FFNNs) on synthetic stability diagrams generated from a simulation package.
• Use 5×5 pixel patches of pre-processed stability diagrams as input to the FFNNs for transition line classification.
• Design networks with very few trainable parameters—on the order of tens to hundreds—suitable for memristor crossbar arrays.
• Implement a patch-shifting strategy to scan across the stability diagram, identifying transition lines incrementally.
• Use a single output neuron to classify whether a transition line is present in a given patch.
• Train the networks using supervised learning with ground-truth labels from simulated data, then evaluate on real experimental data.

Experimental results

Research questions

• RQ1 Can extremely small neural networks detect charge-state transition lines in quantum dot stability diagrams with high accuracy?
• RQ2 Can models trained on synthetic data generalize effectively to real experimental quantum dot devices?
• RQ3 What is the minimal network size required to achieve robust performance in transition line detection?
• RQ4 Can the patch-based scanning strategy reduce experimental measurement costs by focusing on small regions of the stability diagram?
• RQ5 Are these compact networks compatible with existing memristor crossbar hardware for on-chip deployment?

Key findings

• Neural networks with as few as 25 input neurons (5×5 patches) achieve high accuracy in detecting charge-state transition lines in stability diagrams.
• The models trained on synthetic data generalize effectively to real experimental silicon metal-oxide-semiconductor quantum dot devices.
• The minimal network size required for high success rates is small enough to be implemented on existing memristor crossbar arrays.
• Using arrays of connected patches increases the success rate of autotuning while minimizing the number of required measurements.
• The approach significantly reduces experimental measurement costs by covering only small, targeted regions of the stability diagram.
• The entire autotuning pipeline, including transition detection, is compatible with low-power, on-chip deployment using emerging memristor hardware.