Session N28: Computational Design, Understanding and Discovery of Novel Materials V

Machine Learning Based Electronic Structure Prediction: From Nanostructures to Complex Alloys

In this talk, I will describe our work on using specialized first principles calculations with machine learning tools to enable prediction of the electronic structure of various nanomaterials and bulk systems. I will focus on two related but independent directions. In the first, I will show how helical and cyclic symmetry adapted density functional theory calculations may be used to train interpretable machine learning models of the electronic fields of quasi-one-dimensional materials. The descriptors in this framework are global geometry and strain parameters. Through examples involving distorted carbon nanotubes, I will show how the framework can be particularly accurate in its prediction, even with limited training data. I will discuss the use of this framework for automated materials discovery, and in multiscale modeling. In the second, I will discuss the use of high-throughput first principles calculations to train machine learning models of bulk systems featuring some degree of disorder in atomic arrangements. The descriptors in this framework are local in nature and the prediction of electronic fields occurs in a pointwise manner spatially. I will discuss the use of this framework for prediction of the electronic structure of compositionally complex alloys, from which, various material properties of engineering interest may be inferred.

I will end with a discussion of ongoing and future research directions.   

Novel high-mobility candidates from the Materials Cloud 2D database

Over the past two decades, 2D materials have attracted tremendous interest thanks to their novel science and promise of possible nanotechnology applications.

Here, we focus on phonon-limited charge transport in gated 2D semiconductors to identify high mobility candidates. We improve and extend a previous study [1] exploring the expanded version of our curated portfolio of 2D structures (https://www.materialscloud.org/discover/2dstructures/). This gives us 1205 novel easily exfoliable candidates, from which we select 42 semiconductors. We combine density-functional perturbation theory - accounting for dimensionality (2D boundary conditions) and electrostatic doping - with Boltzmann transport beyond the relaxation-time approximation. Special attention is dedicated to the role of dimensionality and doping in driving electron-phonon interactions. We discuss the most interesting candidates, and the entire discovery workflow powered by the AiiDA materials informatics infrastructure.

[1] T. Sohier et al., 2D Mater. 8, 1015025 (2021).

*SNSF, NCCR MARVEL

    

A novel first principles framework for the study of chiral matter

Chiral nanomaterials, particularly, quasi-one-dimensional structures with helical symmetries, exhibit fascinating electric, magnetic, optical, and transport properties. If strategically engineered, these structures' unique properties could impact the design of novel quantum, electromagnetic and photonic devices. In this talk, I will describe a self-consistent first principles simulation framework for the discovery and characterization of such materials. Key ingredients of our technique include: 1) the use of helical Bloch waves to block-diagonalize the Kohn-Sham Density Functional Theory Hamiltonian, 2) discretization and solution of the governing equations of Kohn-Sham theory over a symmetry adapted unit cell, by means of specialized basis functions, and 3) solution of the electrostatics problem by means of a mixed spectral-finite difference formulation. I will also outline various applications of our simulations framework, including its use in the study of layered one-dimensional materials possibly featuring strongly correlated electronic states.   

Trends in MXene structure due to elemental composition: A DFT study

MXenes are a tunable class of materials (M_n+1C_nT_z or M_n+1N_nT_z), where the transition metal (M) and termination group (T) can be modulated to change the material’s properties. Density functional theory (DFT) predictions are fundamental in understanding MXene properties and selecting the optimal composition for a given application. For years, MXene synthesis required a fluorinated solution, which limited termination groups to F, O, and OH. Recent developments in fluorine-free synthesis, however, have led to the discovery of more than six new termination groups. We present a DFT analysis of 90 MXenes, including 9 transition metals and 9 termination groups, the majority of which have not been previously studied. It was presumed that MXenes retain the trigonal symmetry of their parent materials, but recent studies showed that some MXenes are more stable with hexagonal symmetry. We consider a total of 8 possible structures for each MXene- four terminal group locations for each of the two symmetry types. We are the first to report on both preferred symmetry and termination group location for this many MXenes. Our results show that structural preferences of MXenes are highly dependent on the elemental properties of each metal atom and termination group, allowing us to make generalizations and predictions of how a given element may impact the structure of any MXene.   

Mapping 2D Sliding Mechanisms of Bilayer and Bulk Ni-doped MoS2 from First Principles

The van der Waals forces between layers of 2D materials allow for easy sliding, which is used in solid lubrication and also relevant to sliding ferroelectricity, strain engineering, stacking of 2D materials, and nanomechanical devices. Motivated by our previous study [arXiv:2209.15629] in 1D, we explore the 2D sliding potential energy surface (PES) landscape of Ni-doped MoS₂ which is found to have favorable properties for space lubrication. We use previously identified stable and metastable doped structures: Mo/S substitutions and octahedral/tetrahedral intercalations. We determine favored sliding pathways in different directions from the PES and nudged elastic band calculations. The PES have extrema generally related to high-symmetry stackings, but the octahedral intercalated structure also has deep minima away from symmetric points where the structure changes significantly. We analyze the sliding PES in terms of bonding changes, symmetry breaking, relation between bilayer and bulk interactions, and the effects of load. We also compare our results to the registry index model, as generalized to the case of doped materials. These findings augment our understanding of sliding in doped 2D materials and how it could be tuned.   

Computational dissection of 2-dimensional tetrahexagonal InN alloys: Anisotropic mechanical, electronic, and charge carrier transport properties for photocatalytic water splitting

2D materials with unique physical properties lead to new possibilities in future nanomaterial-based applications. Among them, 2D structures which are suitable to be the solar-driven catalyst for water splitting reactions have become excessively important since the demand for clean energy sources increases. Apart from the conventional crystals with well-known symmetries, recent studies showed that materials that have exotic decorations could possess superior features. In this respect, we report novel 2D tetrahexagonal (th-) InN crystal and its ordered alloys that can be utilized as effective catalysts for water splitting reactions. Proposed structures possess robust stability with a versatile mechanical response. After a critical tensile strain value, all monolayers exhibit strain-induced Negative Poisson's ratio in a particular crystal direction, making them half-auxetic materials. The examined materials are indirect semiconductors with desired bandgaps and band edge positions for water splitting applications. Due to their structural anisotropy, they have direction-dependent mobility that can keep the photogenerated charge carriers separated by reducing their recombination probability, which boosts the photocatalytic process. Relatively high absorption capacity in the wide spectral range underlines their potential performance. The novel properties of 2D th-InN and its alloys, indicate that they can be used for water splitting applications in near future.    

Machine learning of borophene-boride hetero-structures for borophene synthesis

Borophene, two-dimensional (2D) boron, exhibits versatile properties which may lead to a variety of applications, such as CO₂ reduction, hydrogen and oxygen evolution reactions, and superconductivity. However, a big challenge in borophene field is synthesis. Though borophene has been synthesized on several metal substrates, its strong interaction with substrates limit the achievement of free-standing borophene. Recently, it was found that boride is formed between borophene and substrate during borophene synthesis on Al (111). Metal borides have the potential to be superior substrates, compared to metals, for borophene synthesis and separation. Here to search for good substrates for borophene synthesis, we firstly generate a dataset of approximately 100 borophene-boride hetero-structures by density functional theory (DFT), then perform machine learning (ML) to analyze the interaction in the hetero-structures. For ML, random forest algorithm is used. Furthermore, we use the ML model in a larger dataset of hypothetical borides to predict good substrates. This work allows us to explore alternative routes of borophene synthesis.

    

Expanding two-dimensional databases by combining exhaustive search and machine learning

We report an exhaustive search of 2d structures for binary and ternary compositions. The methodology consists of a combinatorial exploration of all possible Wyckoff positions of

layered groups while restricting our search to neutral compounds which follow the electronegativity Pauling test. Most elements of the periodic table were considered except

radioactive elements and rare gases. A selected set of oxidation states were used based on experimental feedback, allowing our materials to be proner to synthesis. A screening procedure is developed where pre-optimized structures are obtained by using a crafted universal potential, which is trained by using structures, energies, and forces fully optimized from DFT. We report an unprecedented amount of unreported two-dimensional structures containing the vast majority of elements in the periodic table. Using the obtained structures, we are to decorate nearly all Platonic and Archimedean tesselations and their dual Laves or Catalan tilings. We also report a vibrant space of polymorphism over materials with similar chemical compositions. We further accelerate the exploration of the available chemical space by training a machine learning prototype search by using the structures and energies found in the exhaustive search. We can then report more than 6500 compounds with energies below 250 meV/atom of the convex hull, only over binaries and ternaries.   

Leveraging Deep Neural Networks and Density Functional Theory to guide two-dimensional material synthesis using Chemical Vapor Deposition.

The computational Condensed Matter community has made many strides in recent years towards the discovery of new two-dimensional (2D) materials, utilizing Density Functional Theory (DFT). With the theoretical discovery of these materials, experimentalists must then devise new techniques for efficiently synthesizing said materials. In recent years, modern synthesis approaches have begun to utilize Bayesian Optimization (BO) to guide experimental growth parameter convergence. Unfortunately, given the computational costs, this approach limits the amount of data than can be utilized for model convergence and it requires unguided sample collection to initially train the model. Given the large parameter space for experimental synthesis of 2D materials, we instead propose employing a Deep Neural Network (DNN) to reduce computational costs, where the initial sample collection for training the model is derived using DFT, and the model is guided by Time-Dependent DFT (TD-DFT) computationally derived Raman spectra through, Placzek approximations, of the desired 2D material. In this presentation, we will discuss the TD-DFT and Neural Network computational techniques used, the preliminary results of Chemical Vapor Deposition (CVD) synthesis of 2D materials, and the plans for future work--to include source sharing and further optimizations to the technique.   

Janus Ge2PX (X: N, As, Sb, Bi) Monolayers: First-principles Investigation of Electronic and Thermomelectric Properties

2D materials with unique features lead to new possibilities in future nanomaterial-based applications. One of their intriguing derivatives is Janus monolayers, constructed by substituting all atoms at one side of their binary counterpart with a different atom. As a result of broken out-of-plane mirror symmetry created by the addition of a third element, 2D Janus crystals can exhibit fascinating physical properties. In this respect, we systematically investigate the structural, electronic, and thermoelectric (TE) properties of Ge2PX (X = N, As, Sb, Bi) monolayers using first-principles methods. Our results show that the considered systems are dynamically stable. The electronic structure calculations at the level of HSE+SOC indicate that Ge2PAs and Ge2PSb are indirect semiconductors with a bandgap of 1.59 and 0.37 eV, respectively. Using Landauer formalism, the ballistic transport and TE properties of ternary Ge2PX structures are studied in a wide range of temperatures. Ge2PN possesses the highest phonon thermal conductance, while Ge2PBi has the lowest phonon thermal conductance owing to their phonon transmission spectra. Not only low phonon thermal conductance values but also high n-type power factors maximize n-type ZT values. Also, strain engineering can substantially improve the p-type ZT values at low temperatures. Our study suggests that ternary Ge2PAs and Ge2PSb can be promising n-type thermoelectric candidates at high temperatures.   

Exploring MAX and MAB phases using first-principles methods

Ternary transition metal carbides and nitrides with the so-called MAX structures have received much attention due to their various electronic properties, and continue to be intensely studied. Many members of this family were shown to be functionalized to form 2D materials called MXenes. These 2D materials are promising for a variety of applications such as spintronics and energy storage. Similar to the MAX phases, ternary borides --the MAB phases-- are also found to exist with B substituting the X element. While the crystal and electronic structures of the MAB materials are similar to those of MAX phases, there have been relatively fewer theoretical studies of their electronic structures. In this talk, we present a comparative study of the paramagnetic phases of Cr₂AlC and Cr₂AlB₂ with their derived MXene and MBene structures using, Density functional theory, Dynamical Mean Field Theory, and Wannier function analysis.   

Elastic Properties in Monolayer Covalent-Organic Frameworks

Covalent-Organic Frameworks (COFs) are crystalline porous materials that are based on organic monomeric units, so called building blocks. As a multitude of different building blocks can be combined in reticular chemistry, manifold different porous structures with tailored properties have been synthesized in recent years. Through current experimental progress, monolayer COF materials have been synthesized, providing a new class of 2D materials. However, these materials have defects and grain boundaries which make it challenging to describe properties of realistic materials computationally. To approach this issue, we show in this work how to use a surrogate model to calculate elastic properties of 2D COFs based on density functional based tight binding (DFTB) calculations. This allows us to model defective systems at low computational cost and paves the way for multiscale modeling. Furthermore, this approach enables us to predict elastic properties from the properties of the monomeric building blocks.   

Elemental Amorphous Carbon versus Binary Amorphous Boron Nitride Monolayers

The structure of amorphous materials has been debated since the 1930s as a binary question: Zachariasen continuous random networks (Z-CRNs) vs. Z-CRNs containing crystallites. It was recently demonstrated, however, that amorphous diamond can be synthesized in either form. Here we address the question of the structure of single-atom-thick amorphous monolayers by kinetic Monte Carlo simulations that emulate chemical-vapor deposition (CVD) growth on a substrate. We find that crystallite-containing Z-CRN is by far the energetically preferred structure of elemental monolayer amorphous carbon (MAC), as recently fabricated, whereas the most likely structure of binary monolayer amorphous BN (ma-BN) is altogether different than either of the two long-debated options: it is a compositionally disordered “pseudo-CRN” comprising a mix of B−N and noncanonical B−B and N−N bonds and containing “pseudocrystallites”, namely, honeycomb regions made of noncanonical hexagons [1]. The ma-BN is thermally stable and insulating, and the thermal conductivity is two orders of magnitudes smaller than h-BN due to vibrational-mode localization [2].

[1] Y.-T. Zhang et al., https://doi.org/10.1021/acs.nanolett.2c02542 (2022).

[2] Y.-T. Zhang et al., Appl. Phys. Lett. 120, 222201 (2022).