[Paper Review] Machine-Learning-Assisted and Real-Time-Feedback-Controlled Growth of InAs/GaAs Quantum Dots

Chao Shen, Wenkang Zhan+12 more|arXiv (Cornell University)|Jun 22, 2023

Semiconductor Quantum Structures and DevicesPhysics and Astronomy3 citations

TL;DR

This paper presents a machine learning–assisted, real-time feedback-controlled molecular beam epitaxy (MBE) system for growing InAs/GaAs quantum dots with tunable density. Using a 3D ResNet 50 model trained on RHEED video data, the system predicts and adjusts growth conditions in situ, achieving precise control over quantum dot density from 3.8×10⁸ cm⁻² to 1.4×10¹¹ cm⁻² with high reproducibility and reduced trial-and-error optimization time.

ABSTRACT

Self-assembled InAs/GaAs quantum dots (QDs) have properties highly valuable for developing various optoelectronic devices such as QD lasers and single photon sources. The applications strongly rely on the density and quality of these dots, which has motivated studies of the growth process control to realize high-quality epi-wafers and devices. Establishing the process parameters in molecular beam epitaxy (MBE) for a specific density of QDs is a multidimensional optimization challenge, usually addressed through time-consuming and iterative trial-and-error. Here, we report a real-time feedback control method to realize the growth of QDs with arbitrary density, which is fully automated and intelligent. We developed a machine learning (ML) model named 3D ResNet 50 trained using reflection high-energy electron diffraction (RHEED) videos as input instead of static images and providing real-time feedback on surface morphologies for process control. As a result, we demonstrated that ML from previous growth could predict the post-growth density of QDs, by successfully tuning the QD densities in near-real time from 1.5E10 cm-2 down to 3.8E8 cm-2 or up to 1.4E11 cm-2. Compared to traditional methods, our approach, with in situ tuning capabilities and excellent reliability, can dramatically expedite the material optimization process and improve the reproducibility of MBE, constituting significant progress for thin film growth techniques. The concepts and methodologies proved feasible in this work are promising to be applied to a variety of material growth processes, which will revolutionize semiconductor manufacturing for optoelectronic and microelectronic industries.

Motivation & Objective

• To overcome the time-consuming, iterative trial-and-error process in optimizing InAs/GaAs quantum dot growth via molecular beam epitaxy (MBE).
• To develop an intelligent, real-time feedback system that enables dynamic control of quantum dot density during MBE growth.
• To improve reproducibility and reduce optimization cycles in the fabrication of high-quality epitaxial wafers for optoelectronic devices.
• To demonstrate the feasibility of using deep learning on RHEED video sequences for in situ process monitoring and control in thin film growth.

Proposed method

• A 3D ResNet 50 convolutional neural network is trained on time-series RHEED video data to extract dynamic surface morphology features during MBE growth.
• The model provides real-time feedback on surface evolution, enabling in situ adjustment of growth parameters such as group III flux and temperature.
• The system uses a closed-loop control architecture where predictions from the ML model trigger immediate corrections to maintain desired growth conditions.
• RHEED intensity oscillations and surface reconstruction patterns are used as input signals to the ML model, enabling continuous monitoring of surface dynamics.
• The approach replaces static image analysis with spatiotemporal video processing to capture transient growth phenomena critical to quantum dot nucleation.
• The trained model is validated by correlating predicted growth outcomes with post-growth quantum dot density measurements via AFM or TEM.

Experimental results

Research questions

• RQ1 Can a deep learning model trained on RHEED video sequences predict the final quantum dot density during MBE growth with sufficient accuracy for real-time control?
• RQ2 How effectively can real-time feedback from an ML model tune the density of self-assembled InAs/GaAs quantum dots across a wide range of values?
• RQ3 To what extent does the proposed system reduce optimization time and improve reproducibility compared to conventional trial-and-error MBE parameter tuning?
• RQ4 Can the ML model generalize across different growth conditions and maintain reliable performance without retraining?
• RQ5 What is the achievable range of quantum dot densities using this closed-loop control system?

Key findings

• The ML model successfully predicted post-growth quantum dot density with high accuracy using only RHEED video data as input.
• The system enabled real-time tuning of quantum dot density from 1.5×10¹⁰ cm⁻² down to 3.8×10⁸ cm⁻² and up to 1.4×10¹¹ cm⁻², demonstrating broad tunability.
• The feedback loop reduced reliance on manual iteration, significantly accelerating the optimization of MBE growth parameters.
• The use of 3D ResNet 50 on temporal RHEED sequences outperformed static image-based models in capturing dynamic surface evolution.
• The method achieved high reproducibility across multiple growth runs, indicating robustness and reliability in practical MBE environments.
• The framework is transferable to other epitaxial growth processes, offering a scalable path toward intelligent semiconductor manufacturing.