Origins of structural and electronic transitions in disordered silicon

Abstract

Structurally disordered materials continue to pose fundamental questions [1][2][3][4] , including that of how different disordered phases ("polyamorphs") can coexist and transform from one to another 5-9 . As a widely studied case, amorphous silicon (a-Si) forms a fourfold-coordinated, covalent network at ambient conditions and much higher-coordinated, metalliclike phases under pressure 10-12 . However, a detailed mechanistic understanding of the structural transitions in disordered silicon has been lacking, due to intrinsic limitations of even the most advanced experimental and computational techniques. Here, we show how atomistic machine-learning (ML) models can break through this long-standing barrier, describing liquid-amorphous and amorphous-amorphous transitions with quantum-mechanical accuracy for a system of 100,000 atoms (ten-nanometre length scale). Our simulations reveal a three-step transformation sequence for a-Si under increasing external pressure. First, polyamorphic low-and high-density amorphous (LDA and HDA) regions are found to coexist, rather than appearing sequentially. Then, we observe a structural collapse into a distinct, very-high-density amorphous (VHDA) phase. Finally, our simulations indicate the transient nature of this VHDA phase: it rapidly nucleates crystallites, ultimately leading to the formation of a poly-crystalline structure, consistent with experiments [13][14][15] but not seen in earlier simulations 11,[16][17][18] . An ML model for electronic densities of states (DOS) confirms the onset of metallicity during VHDA formation and subsequent crystallisation. These results shed new light on liquid and amorphous states of silicon, and, in a wider context, they exemplify a holistic, ML-driven approach to predictive materials mod-

Machine learning potentials for extended systems: a perspective

“…The larger the cutoff, the smaller the contribution of the long-range interactions that are ignored, but the space in which the atomic energy function has to be fitted is larger, necessitating more training data. There are plenty of examples, and not coincidentally these constitute the earliest successes of MLPs, where long-range terms beyond 5-6Å or so can be essentially ignored, and the resulting potentials are still extremely useful [8,25,26]. For example in elemental systems, and many bulk materials, the degree of charge transfer is either minimal or screening lengths are very short.…”

“…The success of MLPs, built using compact low body order representations and nonlinear regressors (such as HDNNPs [23] or GAPs [31]), in correctly fitting finely nuanced ab initio potential energy surfaces [25,26] has moved the field forward significantly, and brought with it renewed interest in designing symmetric representations for materials. In particular, Moment Tensor Potentials [32] and later the Atomic Cluster Expansion [34] (see third panel on the right of Fig.…”

Toward Better and Smarter Batteries by Combining AI with Multisensory and Self‐Healing Approaches

“…Recent work by Deringer et al, using GAP molecular-dynamics simulations on 100 000 silicon atoms were capable of accessing the time scales needed for prediction of distinct electronic features that can be compared directly to ultrafast spectroscopic techniques, and the experimentally relevant length scales for description of (poly-)crystallization in amorphous silicon. [46] Applying such approaches to describe battery materials and interfaces, [47] will enable a direct comparison with experimental operando techniques revealing interfacial phenomena and establish structureproperty relations that are out of reach to existing methods.…”

A transferable active-learning strategy for reactive molecular force fields

“…3 Machine learning (ML) approaches have the potential to revolutionise force-field based simulations, aiming to provide the best of both worlds, [4][5][6] and have indeed begun to provide new insights into a range of challenging research problems. [7][8][9][10][11][12][13][14][15][16] The development of an ML potential applicable to the whole periodic table mapping nuclear coordinates to total energies and forces is, however, precluded by the curse of dimensionality. Within small chemical subspaces, models can be achieved using neural networks (NNs), 6,[17][18][19][20][21] kernel-based methods such as the Gaussian Approximation Potential (GAP) framework 22,23 or gradient-domain machine learning (GDML), 24 and linear fitting with properly chosen basis functions, 25,26 each with different data requirements and transferability.…”