[Paper Review] Machine learning-enabled tomographic imaging of chemical short-range atomic ordering

Yue Li, Timoteo Colnaghi+14 more|arXiv (Cornell University)|Mar 23, 2023

Advanced Materials Characterization Techniques4 citations

TL;DR

This study introduces a machine learning-enhanced atom probe tomography (APT) framework to quantitatively reconstruct 3D chemical short-range order (CSRO) in CoCrNi medium-entropy alloys. By combining high-resolution APT data with convolutional neural networks and Monte Carlo simulations, the method identifies multiple CSRO configurations and establishes quantitative links between processing parameters and material properties, enabling atomic-scale design of advanced materials.

ABSTRACT

In solids, chemical short-range order (CSRO) refers to the self-organisation of atoms of certain species occupying specific crystal sites. CSRO is increasingly being envisaged as a lever to tailor the mechanical and functional properties of materials. Yet quantitative relationships between properties and the morphology, number density, and atomic configurations of CSRO domains remain elusive. Herein, we showcase how machine learning-enhanced atom probe tomography (APT) can mine the near-atomically resolved APT data and jointly exploit the technique's high elemental sensitivity to provide a 3D quantitative analysis of CSRO in a CoCrNi medium-entropy alloy. We reveal multiple CSRO configurations, with their formation supported by state-of-the-art Monte-Carlo simulations. Quantitative analysis of these CSROs allows us to establish relationships between processing parameters and physical properties. The unambiguous characterization of CSRO will help refine strategies for designing advanced materials by manipulating atomic-scale architectures.

Motivation & Objective

• To overcome the challenge of quantitatively characterizing chemical short-range order (CSRO) in complex solid-state materials.
• To develop a data-driven approach that extracts 3D CSRO morphology, density, and configurations from near-atomically resolved APT data.
• To establish quantitative relationships between processing parameters and physical properties via CSRO microstructure.
• To validate observed CSRO configurations using state-of-the-art Monte Carlo simulations.
• To enable atomic-scale design of materials by manipulating CSRO architectures.

Proposed method

• Employing machine learning, specifically convolutional neural networks, to analyze and reconstruct 3D chemical short-range order from high-sensitivity APT data.
• Using a deep learning model trained on simulated APT data to identify and classify distinct CSRO configurations in experimental datasets.
• Integrating experimental APT data with Monte Carlo simulations to validate the formation energetics and stability of observed CSRO structures.
• Applying tomographic reconstruction techniques to achieve 3D quantitative analysis of elemental distributions and local ordering.
• Using statistical analysis to correlate CSRO domain density, morphology, and atomic configurations with processing history.
• Validating model predictions through comparison with simulated APT data and known thermodynamic behavior.

Experimental results

Research questions

• RQ1 What are the dominant chemical short-range order (CSRO) configurations in CoCrNi medium-entropy alloys?
• RQ2 How do processing parameters influence the formation and distribution of CSRO domains?
• RQ3 Can machine learning accurately reconstruct 3D CSRO from near-atomically resolved APT data?
• RQ4 What is the quantitative relationship between CSRO microstructure and mechanical or functional properties?
• RQ5 How do simulated CSRO configurations compare with experimentally observed ones in terms of stability and formation energy?

Key findings

• Multiple distinct chemical short-range order (CSRO) configurations were identified in the CoCrNi medium-entropy alloy using machine learning-enhanced APT.
• The observed CSRO configurations were validated through agreement with state-of-the-art Monte Carlo simulations, confirming their thermodynamic stability.
• A quantitative link was established between processing parameters and the density and morphology of CSRO domains.
• The machine learning model successfully reconstructed 3D CSRO with high fidelity, enabling detailed statistical analysis of atomic-scale ordering.
• The method revealed that CSRO is not uniformly distributed but forms specific, energetically favorable configurations influenced by processing history.
• The unambiguous characterization of CSRO provides a pathway for designing materials by engineering atomic-scale architectures.