A picture worth a thousand atoms: 2D image to 3D nanostructure mapping by merging simulated and measured data
Figure: Scanning transmission electron microscopy is an integrative technique which leads to loss of structural information of surface nanostructures. This information can be reconstructed with the help of electronic structure simulations and machine learning methods.
A picture worth a thousand atoms: 2D image to 3D nanostructure mapping by merging simulated and measured data
Understanding how local atomic structure and long-range emergent magnetic and electronic properties in defective 2D materials are connected is critical for the development of next generation functional materials. However, modern atomically-resolved imaging techniques only give an integrated snapshot of the structure, without revealing the details of the three-dimensional morphology or the stability: There are many ways to arrange atoms that give essentially the same 2D image.
The project will employ electronic structure calculations of well-characterized 2D materials, simulations of electron microscopy images, and machine learning methods to reconstruct the 3D atomic positions of materials from a 2D microscopy image. The student will work closely with experts at national spectroscopy and imaging facilities to deliver scientific software applicable to experimental imaging data.
Supervisors
Primary: Prof. Reinhard Maurer, Chemistry
Prof. Richard Beanland, Physics
A transcript of the video is available by clicking this link - transcript opens in another windowLink opens in a new window
Background
The controlled engineering of defects and dopants in 2D materials and nanostructures is fundamental to their application in technologies ranging from quantum electronics [1] to energy storage and catalysis [2]; for example, defects in MoS2 produce quantum light in optical emission experiments [3]. No single analytical technique can measure both local structure and electronic properties over orders of magnitude in length and complementary methods are needed. Scanning transmission electron microscopy (STEM) measurements provide atomic-scale topographs of a material. However, since this intensity of the measured signal is integrated along the incident electron beam and most information with respect to depth is lost, the three-dimensional information of the nanostructure is not directly accessible.
The project will employ first principles electronic structure calculations of well-characterized functionalized graphene [4] and supported metal nanostructures [5], as well as multiple scattering simulations of TEM images to generate a labelled database of 3D structures and images. This data establishes the mapping of structure to image. The inverse problem, namely the reconstruction of three-dimensional structures will be achieved with the combined use of machine learning interatomic potentials [5] and generative machine learning models [6], such that the most likely structures that give rise to an image can be determined. The student will work closely with experts at national spectroscopy facilities (Diamond Light Source, STFC) to transfer the model from simulated to realistic experimental data.
Project Aims
The aim of this project is to create a computational tool based on experimental input, simulated data, and machine learning methodology to extract 3D atomic structure information from 2D identical location STEM images. STEM image data will be augmented with atomistic structure data from electronic structure theory and STEM image simulations. All data will be combined into an automated workflow that predicts thermodynamically stable nanoparticle structures that are most likely to give rise to the observed STEM images.
Project Outcomes
- Generation of a database of nanostructures from electronic structure theory calculations and transmission electron microscopy image simulations
- Development of a machine learning model capable of inferring 3D atomic structure from two-dimensional TEM projection images
- Application of the new approach to the characterization of metal nanostructures stabilized on defective graphene films
- Development of an automated 3D structure analysis software applicable for a broad range of scientific end users
Skills that the student will acquire
- Electronic structure theory calculations of materials, atomistic molecular simulation methods
- Experience with machine learning methods
- Expertise in surface science characterization techniques and multiple scattering simulations of transmission electron microscopy images.
- Software development in Python
The project is 50% funded by the STFC Ada Lovelace PhD program and will be in close collaboration with STFC Scientific Computing. The student will benefit from extended secondments at STFC.
References:
[1] Nano Res. 16, 3104–3124 (2023)
[2] Chem. Soc. Rev. 52, 1723-1772 (2023)
[3] Nature Communications 12, 3522 (2021)
[4] Phys. Rev. Lett. 132, 196201 (2024)
[5] Digital Discovery, 2022, 1, 463-475
[6] Nature Computational Science 3, 139–148 (2023)
How to apply
This is a fully-funded 4-year PhD position based in the HetSys Centre for Doctoral Training at the University of Warwick. All applications must be made through the University's postgraduate application form with a deadline of 20 January 2025. Please see our How to ApplyLink opens in a new window page for further details on the application process. For further information about student funding, the integrated HetSys training programme and what life is like in the HetSys CDT, please visit the Study with Us page.