Abstract
Crystal cleavage provides a process to create atomically smooth materials surfaces. To identify possible cleavage planes in periodic materials, we generate bonding networks for crystals from MaterialsProject with five or fewer atoms in the primitive unit cell. For three-dimensional crystals, we systematically break up to three bonds in the bonding network. Successful cleavage reduces the bonding network to two periodic dimensions, creating a layer of the cleaved crystal. We identify 3,897 unique cleavage surfaces across 1,847 crystals and characterize the likelihood of cleavage by calculating the work of adhesion and the relaxed surface energy of these layers using density functional theory to discover 2,924 potential substrates. We discuss the distribution of lattice parameters across these substrates and epitaxially match them against BaSnO$_3$. Following, we identify distinct trends in the work of adhesion of these layers and relate them to metallic and covalent/ionic bonding. For the lowest energy surfaces, we fully optimize the structure, including dispersion interactions, and identify 17 2D materials with surface energies below that of freestanding SnSe.
Introduction
In the field of thin-film electronics, the substrate—the base material upon which a film is grown—is a critical component. High-quality crystalline film growth (epitaxy) requires the substrate's surface to match the film’s lattice structure and symmetry. Historically, substrate discovery has been a slow, trial-and-error process limited to a small library of commercially available crystals. Lattice mismatch often leads to strain and defects, degrading electronic performance. For instance, the transparent conductor $BaSnO_3$ currently lacks well-matched substrates, with standard options like $SrTiO_3$ imposing significant strain. This work presents a high-throughput computational pipeline to systematically identify potential new substrates by simulating crystal cleavage on thousands of known materials.
Methods: Simulated Crystal Cleavage
The researchers developed a multi-step workflow starting with bonding network analysis using the Topological Scaling Algorithm (TSA) to map atomic connectivity. To identify potential cleavage planes, two novel simulated cleavage approaches were implemented:
- Periodic Bonds: Breaking all instances of a specific bond type between two atoms within a unit cell.
- Periodic Atom: Breaking all bonds connected to specific atoms.
If breaking these bonds resulted in a 2D-layered motif, the system's stability was assessed using Density Functional Theory (DFT) to calculate the Work of Cleavage ($E_{cleave}$) and Surface Energy ($E_{surf}$). Finally, a geometric algorithm was used for lattice matching against target materials like $BaSnO_3$.
Key Insights and Results
The orientation-agnostic algorithm successfully expanded the library of potential substrates. From an initial set of approximately 9,000 crystals in the Materials Project database, the team identified 4,693 unique cleavage surfaces across 2,133 bulk materials. This resulted in a library of 3,991 potential substrates with surface energies comparable to or better than industry standards like AlN or ZnO.
Bonding Trends and Energetics
A significant finding was the energetic divide based on precursor bonding types. Metallic precursors generally exhibited lower energy per bond (~1.3 eV) compared to covalent or ionic precursors (~1.8 eV). This suggests that the less localized nature of metallic bonds reduces the energetic "cost" of cleavage.
Breakthroughs in Substrate Matching
The study identified 42 potential substrates for $BaSnO_3$. Notably, materials such as $Rb_2NiO_2$ and $NiO$ were predicted to reduce epitaxial strain to just +0.3%, a significant improvement over the -5.4% strain from $SrTiO_3$. Furthermore, the majority of identified stable surfaces (4,626) had low Miller indices ($≤ 1$), confirming that the most stable cleavage planes are often the simplest ones.
Limitations and Future Outlook
While this high-throughput approach significantly expands the search space for substrates, several limitations remain. The study was restricted to crystals with $≤ 5$ atoms in the primitive cell to manage computational costs, excluding more complex ternary compounds. Additionally, while partially-relaxed surface energies were calculated, exhaustive searches for complex surface reconstructions were not performed. Future work should incorporate phonon calculations to ensure vibrational stability and consider chemical reactivity between precursors and substrates during experimental growth processes like MBE or MOCVD. The complete dataset is available at MaterialsWeb.org to facilitate further experimental validation.
Summary updated