Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Origins of structural and electronic transitions in disordered silicon

Nature volume 589, pages 59–64 (2021)Cite this article

Subjects

Abstract

Structurally disordered materials pose fundamental questions1,2,3,4, including how different disordered phases (‘polyamorphs’) can coexist and transform from one phase to another5,6,7,8,9. Amorphous silicon has been extensively studied; it forms a fourfold-coordinated, covalent network at ambient conditions and much-higher-coordinated, metallic phases under pressure10,11,12. However, a detailed mechanistic understanding of the structural transitions in disordered silicon has been lacking, owing to the intrinsic limitations of even the most advanced experimental and computational techniques, for example, in terms of the system sizes accessible via simulation. Here we show how atomistic machine learning models trained on accurate quantum mechanical computations can help to describe liquid–amorphous and amorphous–amorphous transitions for a system of 100,000 atoms (ten-nanometre length scale), predicting structure, stability and electronic properties. Our simulations reveal a three-step transformation sequence for amorphous silicon under increasing external pressure. First, polyamorphic low- and high-density amorphous 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 polycrystalline structure, consistent with experiments13,14,15 but not seen in earlier simulations11,16,17,18. A machine learning model for the electronic density of states confirms the onset of metallicity during VHDA formation and the subsequent crystallization. These results shed light on the liquid and amorphous states of silicon, and, in a wider context, they exemplify a machine learning-driven approach to predictive materials modelling.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on Springer Link
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Vitrification of supercooled liquid silicon.
Fig. 2: a-Si at high and very-high pressure.
Fig. 3: Electronic fingerprints of structural transitions.

Similar content being viewed by others

Data availability

Original data supporting this work, including coordinates for all reported structural models, are openly available in the Zenodo repository (https://doi.org/10.5281/zenodo.4174139).

Code availability

The QUIP code, which provides the interface for carrying out GAP-driven simulations with LAMMPS, is publicly available at https://github.com/libAtoms/QUIP; additional information may be found there. The GAP code is available freely for non-commercial research at http://www.libatoms.org/gap/gap_download.html.

References

  1. Elliott, S. R. Medium-range structural order in covalent amorphous solids. Nature 354, 445–452 (1991).

    ADS  CAS  Google Scholar 

  2. Sheng, H. W., Luo, W. K., Alamgir, F. M., Bai, J. M. & Ma, E. Atomic packing and short-to-medium-range order in metallic glasses. Nature 439, 419–425 (2006).

    ADS  CAS  PubMed  Google Scholar 

  3. Xie, R. et al. Hyperuniformity in amorphous silicon based on the measurement of the infinite-wavelength limit of the structure factor. Proc. Natl Acad. Sci. USA 110, 13250–13254 (2013).

    ADS  CAS  PubMed  PubMed Central  Google Scholar 

  4. Keen, D. A. & Goodwin, A. L. The crystallography of correlated disorder. Nature 521, 303–309 (2015).

    ADS  CAS  PubMed  Google Scholar 

  5. Hedler, A., Klaumünzer, S. L. & Wesch, W. Amorphous silicon exhibits a glass transition. Nat. Mater. 3, 804–809 (2004).

    ADS  CAS  PubMed  Google Scholar 

  6. Wilding, M. C., Wilson, M. & McMillan, P. F. Structural studies and polymorphism in amorphous solids and liquids at high pressure. Chem. Soc. Rev. 35, 964–986 (2006).

    CAS  PubMed  Google Scholar 

  7. Sheng, H. W. et al. Polyamorphism in a metallic glass. Nat. Mater. 6, 192–197 (2007).

    ADS  CAS  PubMed  Google Scholar 

  8. Debenedetti, P. G., Sciortino, F. & Zerze, G. H. Second critical point in two realistic models of water. Science 369, 289–292 (2020).

    ADS  CAS  PubMed  Google Scholar 

  9. Cheng, B., Mazzola, G., Pickard, C. J. & Ceriotti, M. Evidence for supercritical behaviour of high-pressure liquid hydrogen. Nature 585, 217–220 (2020).

    CAS  PubMed  Google Scholar 

  10. Deb, S. K., Wilding, M., Somayazulu, M. & McMillan, P. F. Pressure-induced amorphization and an amorphous–amorphous transition in densified porous silicon. Nature 414, 528–530 (2001).

    ADS  CAS  PubMed  Google Scholar 

  11. McMillan, P. F., Wilson, M., Daisenberger, D. & Machon, D. A density-driven phase transition between semiconducting and metallic polyamorphs of silicon. Nat. Mater. 4, 680–684 (2005).

    ADS  CAS  PubMed  Google Scholar 

  12. Daisenberger, D. et al. Polyamorphic amorphous silicon at high pressure: Raman and spatially resolved X-ray scattering and molecular dynamics studies. J. Phys. Chem. B 115, 14246–14255 (2011).

    CAS  PubMed  Google Scholar 

  13. Pandey, K. K., Garg, N., Shanavas, K. V., Sharma, S. M. & Sikka, S. K. Pressure induced crystallization in amorphous silicon. J. Appl. Phys. 109, 113511 (2011).

    ADS  Google Scholar 

  14. Garg, N., Pandey, K. K., Shanavas, K. V., Betty, C. A. & Sharma, S. M. Memory effect in low-density amorphous silicon under pressure. Phys. Rev. B 83, 115202 (2011).

    ADS  Google Scholar 

  15. Haberl, B., Guthrie, M., Sprouster, D. J., Williams, J. S. & Bradby, J. E. New insight into pressure-induced phase transitions of amorphous silicon: the role of impurities. J. Appl. Cryst. 46, 758–768 (2013).

    CAS  Google Scholar 

  16. Durandurdu, M. & Drabold, D. A. Ab initio simulation of first-order amorphous-to-amorphous phase transition of silicon. Phys. Rev. B 64, 014101 (2001).

    ADS  Google Scholar 

  17. Morishita, T. High density amorphous form and polyamorphic transformations of silicon. Phys. Rev. Lett. 93, 055503 (2004).

    ADS  PubMed  Google Scholar 

  18. Daisenberger, D. et al. High-pressure X-ray scattering and computer simulation studies of density-induced polyamorphism in silicon. Phys. Rev. B 75, 224118 (2007).

    ADS  Google Scholar 

  19. Behler, J. First principles neural network potentials for reactive simulations of large molecular and condensed systems. Angew. Chem. Int. Ed. 56, 12828–12840 (2017).

    CAS  Google Scholar 

  20. Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O. & Walsh, A. Machine learning for molecular and materials science. Nature 559, 547–555 (2018).

    ADS  CAS  PubMed  Google Scholar 

  21. Deringer, V. L., Caro, M. A. & Csányi, G. Machine learning interatomic potentials as emerging tools for materials science. Adv. Mater. 31, 1902765 (2019).

    CAS  Google Scholar 

  22. Behler, J., Martoňák, R., Donadio, D. & Parrinello, M. Metadynamics simulations of the high-pressure phases of silicon employing a high-dimensional neural network potential. Phys. Rev. Lett. 100, 185501 (2008).

    ADS  PubMed  Google Scholar 

  23. Bonati, L. & Parrinello, M. Silicon liquid structure and crystal nucleation from ab initio deep metadynamics. Phys. Rev. Lett. 121, 265701 (2018).

    ADS  CAS  PubMed  Google Scholar 

  24. Bartók, A. P., Payne, M. C., Kondor, R. & Csányi, G. Gaussian approximation potentials: the accuracy of quantum mechanics, without the electrons. Phys. Rev. Lett. 104, 136403 (2010).

    ADS  PubMed  Google Scholar 

  25. Bartók, A. P., Kermode, J., Bernstein, N. & Csányi, G. Machine learning a general-purpose interatomic potential for silicon. Phys. Rev. X 8, 041048 (2018).

    Google Scholar 

  26. Deringer, V. L. et al. Realistic atomistic structure of amorphous silicon from machine-learning-driven molecular dynamics. J. Phys. Chem. Lett. 9, 2879–2885 (2018).

    CAS  PubMed  Google Scholar 

  27. Bernstein, N. et al. Quantifying chemical structure and machine-learned atomic energies in amorphous and liquid silicon. Angew. Chem. Int. Ed. 58, 7057–7061 (2019).

    CAS  Google Scholar 

  28. Hejna, M., Steinhardt, P. J. & Torquato, S. Nearly hyperuniform network models of amorphous silicon. Phys. Rev. B 87, 245204 (2013).

    ADS  Google Scholar 

  29. Dahal, D., Atta-Fynn, R., Elliott, S. R. & Biswas, P. Hyperuniformity and static structure factor of amorphous silicon in the infinite-wavelength limit. J. Phys. Conf. Ser. 1252, 012003 (2019).

    CAS  Google Scholar 

  30. Jia, W. et al. Pushing the limit of molecular dynamics with ab initio accuracy to 100 million atoms with machine learning. In SC’20: Proc. Int. Conf. High Performance Computing, Networking, Storage and Analysis (ed. Cuicchi, C.) 5 (2020).

  31. Khaliullin, R. Z., Eshet, H., Kühne, T. D., Behler, J. & Parrinello, M. Nucleation mechanism for the direct graphite-to-diamond phase transition. Nat. Mater. 10, 693–697 (2011).

    ADS  CAS  PubMed  Google Scholar 

  32. Bartók, A. P., Kondor, R. & Csányi, G. On representing chemical environments. Phys. Rev. B 87, 184115 (2013).

    ADS  Google Scholar 

  33. De, S., Bartók, A. P., Csányi, G. & Ceriotti, M. Comparing molecules and solids across structural and alchemical space. Phys. Chem. Chem. Phys. 18, 13754–13769 (2016).

    CAS  PubMed  Google Scholar 

  34. Imai, M., Mitamura, T., Yaoita, K. & Tsuji, K. Pressure-induced phase transition of crystalline and amorphous silicon and germanium at low temperatures. High Press. Res. 15, 167–189 (1996).

    ADS  Google Scholar 

  35. Moras, G. et al. Shear melting of silicon and diamond and the disappearance of the polyamorphic transition under shear. Phys. Rev. Mater. 2, 083601 (2018).

    CAS  Google Scholar 

  36. Hu, J. Z. & Spain, I. L. Phases of silicon at high pressure. Solid State Commun. 51, 263–266 (1984).

    ADS  CAS  Google Scholar 

  37. Shanavas, K. V., Pandey, K. K., Garg, N. & Sharma, S. M. Computer simulations of crystallization kinetics in amorphous silicon under pressure. J. Appl. Phys. 111, 063509 (2012).

    ADS  Google Scholar 

  38. Xu, M. et al. Pressure-induced crystallization of amorphous Ge2Sb2Te5. J. Appl. Phys. 108, 083519 (2010).

    ADS  Google Scholar 

  39. Wu, M., Tse, J. S., Wang, S. Y., Wang, C. Z. & Jiang, J. Z. Origin of pressure-induced crystallization of Ce75Al25 metallic glass. Nat. Commun. 6, 6493 (2015).

    ADS  CAS  PubMed  Google Scholar 

  40. Sun, J., Ruzsinszky, A. & Perdew, J. P. Strongly constrained and appropriately normed semilocal density functional. Phys. Rev. Lett. 115, 036402 (2015).

    ADS  PubMed  Google Scholar 

  41. Stillinger, F. H. & Weber, T. A. Computer simulation of local order in condensed phases of silicon. Phys. Rev. B 31, 5262–5271 (1985).

    ADS  CAS  Google Scholar 

  42. Ben Mahmoud, C., Anelli, A., Csányi, G. & Ceriotti, M. Learning the electronic density of states in condensed matter. Phys. Rev. B (in the press).

  43. Mott, N. F. & Davis, E. A. Electronic Processes in Non-crystalline Materials (Oxford Univ, Press, 2012).

  44. Beye, M., Sorgenfrei, F., Schlotter, W. F., Wurth, W. & Fohlisch, A. The liquid–liquid phase transition in silicon revealed by snapshots of valence electrons. Proc. Natl Acad. Sci. USA 107, 16772–16776 (2010).

    ADS  CAS  PubMed  PubMed Central  Google Scholar 

  45. Barkalov, O. I. et al. Pressure-induced transformations and superconductivity of amorphous germanium. Phys. Rev. B 82, 020507 (2010).

    ADS  Google Scholar 

  46. Mignot, J. M., Chouteau, G. & Martinez, G. High pressure superconductivity of silicon. Physica B+C 135, 235–238 (1985).

    ADS  CAS  Google Scholar 

  47. Helfrecht, B., Cersonsky, R. K., Fraux, G. & Ceriotti, M. Structure-property maps with Kernel principal covariates regression. Mach. Learn. Sci. Technol. 1, 045021 (2020).

    Google Scholar 

  48. Stukowski, A. Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool. Model. Simul. Mater. Sci. Eng. 18, 015012 (2010).

    ADS  Google Scholar 

  49. Deringer, V. L. & Csányi, G. Machine learning based interatomic potential for amorphous carbon. Phys. Rev. B 95, 094203 (2017).

    ADS  Google Scholar 

  50. Caro, M. A., Csányi, G., Laurila, T. & Deringer, V. L. Machine learning driven simulated deposition of carbon films: from low-density to diamondlike amorphous carbon. Phys. Rev. B 102, 174201 (2020).

    ADS  CAS  Google Scholar 

  51. Blöchl, P. E. Projector augmented-wave method. Phys. Rev. B 50, 17953–17979 (1994).

    ADS  Google Scholar 

  52. Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169–11186 (1996).

    ADS  CAS  Google Scholar 

  53. Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 59, 1758–1775 (1999).

    ADS  CAS  Google Scholar 

  54. Deringer, V. L., Pickard, C. J. & Csányi, G. Data-driven learning of total and local energies in elemental boron. Phys. Rev. Lett. 120, 156001 (2018).

    ADS  CAS  PubMed  Google Scholar 

  55. Harl, J. & Kresse, G. Accurate bulk properties from approximate many-body techniques. Phys. Rev. Lett. 103, 056401 (2009).

    ADS  PubMed  Google Scholar 

  56. Harl, J., Schimka, L. & Kresse, G. Assessing the quality of the random phase approximation for lattice constants and atomization energies of solids. Phys. Rev. B 81, 115126 (2010).

    ADS  Google Scholar 

  57. Schimka, L. et al. Accurate surface and adsorption energies from many-body perturbation theory. Nat. Mater. 9, 741–744 (2010).

    ADS  CAS  PubMed  Google Scholar 

  58. Bartók, A. P. et al. Machine-learning approach for one- and two-body corrections to density functional theory: applications to molecular and condensed water. Phys. Rev. B 88, 054104 (2020).

    ADS  Google Scholar 

  59. Ramakrishnan, R. et al. Big data meets quantum chemistry approximations: the Δ-machine learning approach. J. Chem. Theory Comput. 11, 2087–2096 (2015).

    CAS  PubMed  Google Scholar 

  60. Bernstein, N., Csányi, G. & Deringer, V. L. De novo exploration and self-guided learning of potential-energy surfaces. npj Comput. Mater. 5, 99 (2019).

    ADS  Google Scholar 

  61. Pickard, C. J. & Needs, R. J. High-pressure phases of silane. Phys. Rev. Lett. 97, 045504 (2006).

    ADS  PubMed  Google Scholar 

  62. Pickard, C. J. & Needs, R. J. Ab initio random structure searching. J. Phys. Condens. Matter 23, 053201 (2011).

    ADS  PubMed  Google Scholar 

  63. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865–3868 (1996).

    ADS  CAS  PubMed  Google Scholar 

  64. Plimpton, S. Fast parallel algorithms for short-range molecular dynamics. J. Comput. Phys. 117, 1–19 (1995).

    ADS  CAS  MATH  Google Scholar 

  65. Parrinello, M. & Rahman, A. Polymorphic transitions in single crystals: a new molecular dynamics method. J. Appl. Phys. 52, 7182–7190 (1981).

    ADS  CAS  Google Scholar 

  66. Martyna, G. J., Tobias, D. J. & Klein, M. L. Constant pressure molecular dynamics algorithms. J. Chem. Phys. 101, 4177–4189 (1994).

    ADS  CAS  Google Scholar 

  67. Shinoda, W., Shiga, M. & Mikami, M. Rapid estimation of elastic constants by molecular dynamics simulation under constant stress. Phys. Rev. B 69, 134103 (2004).

    ADS  Google Scholar 

  68. Bringuier, S. dump2VDOS.py. Python code http://www.u.arizona.edu/~stefanb/Codes/dump2VDOS.py (2014).

  69. Drabold, D. A. & Sankey, O. F. Maximum entropy approach for linear scaling in the electronic structure problem. Phys. Rev. Lett. 70, 3631–3634 (1993).

    ADS  CAS  PubMed  Google Scholar 

  70. Kwon, I., Biswas, R., Wang, C. Z., Ho, K. M. & Soukoulis, C. M. Transferable tight-binding models for silicon. Phys. Rev. B 49, 7242–7250 (1994).

    ADS  CAS  Google Scholar 

  71. Drabold, D. A., Li, Y., Cai, B. & Zhang, M. Urbach tails of amorphous silicon. Phys. Rev. B 83, 045201 (2011).

    ADS  Google Scholar 

  72. Skilling, J. The eigenvalues of mega-dimensional matrices. In Maximum Entropy and Bayesian Methods (ed. Skilling, J.) 455–466 (Kluwer, 1989).

  73. Jaynes, E. T. Probability Theory: The Logic of Science (Cambridge Univ. Press, 2003).

  74. Mead, L. R. & Papanicolaou, N. Maximum entropy in the problem of moments. J. Math. Phys. 25, 2404–2417 (1984).

    ADS  MathSciNet  Google Scholar 

  75. Bandyopadhyay, K., Bhattacharya, A. K., Biswas, P. & Drabold, D. A. Maximum entropy and the problem of moments: a stable algorithm. Phys. Rev. E 71, 057701 (2005).

    ADS  CAS  Google Scholar 

  76. Weiße, A., Wellein, G., Alvermann, A. & Fehske, H. The kernel polynomial method. Rev. Mod. Phys. 78, 275–306 (2006).

    ADS  MathSciNet  MATH  Google Scholar 

  77. Drabold, D. A., Ordejón, P., Dong, J. & Martin, R. M. Spectral properties of large fullerenes: from cluster to crystal. Solid State Commun. 96, 833–838 (1995).

    ADS  CAS  Google Scholar 

  78. Willatt, M. J., Musil, F. & Ceriotti, M. Feature optimization for atomistic machine learning yields a data-driven construction of the periodic table of the elements. Phys. Chem. Chem. Phys. 20, 29661–29668 (2018).

    CAS  PubMed  Google Scholar 

  79. Imbalzano, G. et al. Automatic selection of atomic fingerprints and reference configurations for machine-learning potentials. J. Chem. Phys. 148, 241730 (2018).

    ADS  PubMed  Google Scholar 

  80. Blum, V. et al. Ab initio molecular simulations with numeric atom-centered orbitals. Comput. Phys. Commun. 180, 2175–2196 (2009).

    ADS  CAS  MATH  Google Scholar 

  81. Heyd, J., Scuseria, G. E. & Ernzerhof, M. Hybrid functionals based on a screened Coulomb potential. J. Chem. Phys. 118, 8207–8215 (2003).

    ADS  CAS  Google Scholar 

  82. Krukau, A. V., Vydrov, O. A., Izmaylov, A. F. & Scuseria, G. E. Influence of the exchange screening parameter on the performance of screened hybrid functionals. J. Chem. Phys. 125, 224106 (2006).

    ADS  PubMed  Google Scholar 

  83. Borlido, P. et al. Large-scale benchmark of exchange–correlation functionals for the determination of electronic band gaps of solids. J. Chem. Theory Comput. 15, 5069–5079 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

  84. Musil, F., Willatt, M. J., Langovoy, M. A. & Ceriotti, M. Fast and accurate uncertainty estimation in chemical machine learning. J. Chem. Theory Comput. 15, 906–915 (2019).

    PubMed  Google Scholar 

  85. Bundy, F. P. Phase diagrams of silicon and germanium to 200 kbar, 1000 °C. J. Chem. Phys. 41, 3809–3814 (1964).

    ADS  CAS  Google Scholar 

  86. Funamori, N. & Tsuji, K. Pressure-induced structural change of liquid silicon. Phys. Rev. Lett. 88, 255508 (2002).

    ADS  PubMed  Google Scholar 

  87. Dharma-wardana, M. W. C., Klug, D. D. & Remsing, R. C. Liquid-liquid phase transitions in silicon. Phys. Rev. Lett. 125, 075702 (2020).

    ADS  CAS  PubMed  Google Scholar 

  88. Desgranges, C. & Delhommelle, J. Unraveling liquid polymorphism in silicon driven out-of-equilibrium. J. Chem. Phys. 153, 054502 (2020).

    CAS  PubMed  Google Scholar 

  89. Needs, R. J. & Martin, R. M. Transition from β-tin to simple hexagonal silicon under pressure. Phys. Rev. B 30, 5390–5392 (1984).

    ADS  CAS  Google Scholar 

  90. Laaziri, K. et al. High-energy X-ray diffraction study of pure amorphous silicon. Phys. Rev. B 60, 13520–13533 (1999).

    ADS  CAS  Google Scholar 

Download references

Acknowledgements

V.L.D. acknowledges a Leverhulme Early Career Fellowship and support from the Isaac Newton Trust. Parts of the simulations reported here were carried out during his previous affiliation with the University of Cambridge (until August 2019). N.B. acknowledges support from the Office of Naval Research through the US Naval Research Laboratory’s core basic research programme, and computer time through the US DOD HPCMPO at the AFRL DSRC. D.A.D. acknowledges support from the US NSF under award DMR 1506836. M.C. and C.B.M. acknowledge support by the Swiss National Science Foundation (project no. 200021-182057), and by the NCCR MARVEL, funded by the Swiss National Science Foundation. This work used the ARCHER UK National Supercomputing Service via a Resource Allocation Panel award (project e599) and the UKCP consortium (EPSRC grant EP/P022596/1). All structural drawings were created using OVITO48. We thank A. P. Bartók for technical help.

Author information

Author notes

  1. Stephen R. Elliott

    Present address: Trinity College, Cambridge, UK

Authors and Affiliations

  1. Department of Chemistry, Inorganic Chemistry Laboratory, University of Oxford, Oxford, UK

    Volker L. Deringer

  2. Center for Materials Physics and Technology, US Naval Research Laboratory, Washington, DC, USA

    Noam Bernstein

  3. Engineering Laboratory, University of Cambridge, Cambridge, UK

    Gábor Csányi

  4. Laboratory of Computational Science and Modeling, IMX, École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

    Chiheb Ben Mahmoud & Michele Ceriotti

  5. National Centre for Computational Design and Discovery of Novel Materials (MARVEL), École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

    Chiheb Ben Mahmoud & Michele Ceriotti

  6. Department of Chemistry, Physical and Theoretical Chemistry Laboratory, University of Oxford, Oxford, UK

    Mark Wilson

  7. Department of Physics and Astronomy, Ohio University, Athens, OH, USA

    David A. Drabold

  8. Department of Chemistry, University of Cambridge, Cambridge, UK

    Stephen R. Elliott

Authors
  1. Volker L. Deringer
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Noam Bernstein
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Gábor Csányi
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Chiheb Ben Mahmoud
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Michele Ceriotti
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Mark Wilson
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. David A. Drabold
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Stephen R. Elliott
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

V.L.D., G.C. and S.R.E. initiated the project. V.L.D. and N.B. performed the ambient-pressure simulations; V.L.D. performed the high-pressure simulations; V.L.D., N.B. and G.C. carried out further validation with other machine learning potentials. D.A.D. performed the tight-binding electronic-structure computations. C.B.M. and M.C. performed the electronic DOS machine learning predictions and the KPCovR analysis. V.L.D., M.W., D.A.D. and S.R.E. analysed the data and developed the main conclusions regarding high-pressure phases. All authors contributed to discussions. V.L.D. drafted the paper, and all authors contributed to its final version.

Corresponding author

Correspondence to Volker L. Deringer.

Ethics declarations

Competing interests

G.C. is listed as an inventor on a patent filed by Cambridge Enterprise Ltd related to SOAP and GAP (US patent 8843509, filed on 5 June 2009 and published on 23 September 2014). The other authors declare no competing interests.

Additional information

Peer review information Nature thanks Davide Donadio, Paul McMillan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Machine learning-driven modelling beyond the nanometric length scale.

The fully relaxed a-Si structure with 100,000 atoms is shown. The smaller boxes on the left show the size of a 512-atom system from a recent study26, marking the limit of current DFT methods for simulations over several nanoseconds, and that of a 4,096-atom system in our pilot GAP-MD studies26. All boxes are drawn to scale.

Extended Data Fig. 2 Compression of liquid silicon.

a, Overview of the transition pathways investigated in the present work. The quench (vitrification) and compression runs are discussed in the text. We focus here on additional data that we have collected for validation: namely, the description of the high-temperature liquid. We melted a structure at 1,800 K, above the melting point of diamond-type silicon, and then compressed it by simultaneously adapting the thermostat and barostat settings so as to trace the estimated phase coexistence lines given by Bundy85, in analogy to ref. 86. The temperature was reduced by 41 K GPa−1 to follow the diamond melting line, up to the estimated triple point at 15 GPa, after which the slope was inverted and followed the metallic silicon melting line (+14 K GPa−1)85. The compression rate was 0.5 GPa ps−1. b, Structure factors of liquid silicon during this compression run. Computed values from our simulations (red) are overlaid on experimental reference data by Funamori and Tsuji86 (black) for which the estimated temperatures are at about 50 K above the melting line86, closely mirrored by our computations. In the original experimental work86, arrows indicate the location of the maxima (labelled there as Q1 and Q2) and a shoulder in the first peak (Qh). The height of Qh gradually diminishes at higher pressure, and all these features are correctly described by our simulations. In the context of liquid–liquid transitions, we mention in passing very recent density-functional87 and empirical force-field based studies88; such research questions may become worthwhile targets for future GAP-driven studies as well. Reprinted figure with permission from ref. 86, copyright 2002 by the American Physical Society. Expt., experimental.

Extended Data Fig. 3 Vibrational densities of states (VDOS).

We obtained these by Fourier transformation of the velocity–velocity autocorrelation function, as described in Methods. Two characteristic features associated with the amorphous–amorphous transition under high pressure, observed in previous Raman spectroscopy experiments10,11,12, are reproduced by these simulations. First, the peak at large wavenumbers persists throughout the LDA/HDA coexistence but then disappears entirely. Second, the VHDA formation is associated with the formation of another peak at intermediate wavenumbers. It is noted that this feature appears in both the simulated VHDA and the polycrystalline sh system. The predictions here might be tested, in the future, by combined in situ X-ray diffraction and Raman spectroscopy; the former technique will easily be able to distinguish VHDA silicon from crystalline phases.

Extended Data Fig. 4 Reproducibility of VHDA formation with a separate machine learning potential.

a, Snapshots from a compression simulation using the same starting structure and protocol as for the main result (Fig. 2a–e), but now using a newly fitted GAP machine learning model based on SCAN meta-GGA input data (Methods). This simulation confirms the structural collapse at high pressure, seen in the third panel, and the subsequent crystallization. The SCAN-level machine learning potential initially nucleated β-Sn-like crystallites (N = 6; red colour on the atoms), which is explained in the following. b, Energy–volume curves for relevant crystalline allotropes of silicon, computed using the GAP-18 model (based on PW91 data; top) and the new SCAN-based model (bottom). In both cases, the sequence of dia → β-Sn-type → sh with increasing pressure (decreasing cell volume) is correctly reproduced, consistent with early DFT studies89. With SCAN, the β-Sn-like phase is favoured over a wider range of pressures; the crossover between the two E(V) curves is indicated by arrows in both panels. Note that the absolute energies for both allotropes are very similar, leading to a delicate balance between both. c, Oblique view of the simulation cell from the SCAN simulation after reaching 20 GPa. Initially, β-tin-like crystallites had formed (N = 6; red); then, an sh grain emerged (N = 8; orange). Note that the absolute pressure values at which the subsequent transitions occur are slightly different between the GAP-18 (Fig. 2a–e) and SCAN (Extended Data Fig. 4a) simulations, but the VHDA phases and subsequent formation of polycrystalline phases are clearly observed in both. The same is not the case with an established, empirically fitted interatomic potential, as shown in Extended Data Fig. 6b.

Extended Data Fig. 5 Beyond-DFT modelling with a Δ-GAP machine learning fit.

a, Schematic illustration of the approach, as discussed in Methods. The key ideas are that: (i) the RPA potential-energy surface (PES) can only be sampled at selected points, because of the computational cost, and that: (ii) the difference Δ(RPA–DFT), indicated by red shading, varies more smoothly than the full PES and is therefore more easily amenable to a machine learning fit. b, Example structural snapshots from a GAP-RSS search60. We use such very-small simulation cells to represent large structural diversity in machine learning potential fitting where computational cost is at a premium. c, Quality-of-fit for the difference model, shown in the form of a scatter plot for the training data (blue) and a separate test set (green) of the machine learning prediction (vertical axis) against the ‘ground truth’ to be learned (horizontal axis). The distribution of the target values, σ, is given at the top left, alongside the root mean square error (RMSE) measures for training and testing set. d, Snapshots from a compression simulation using the same starting structure and protocol as for the GAP-18 (Fig. 2a–e) and SCAN (Extended Data Fig. 4) results, but now using the GAP-18 + Δ-GAP(RPA–DFT) difference machine learning potential. The collapse into VHDA is clearly reproduced, as is the subsequent nucleation of crystallites; the result at 20 GPa is a poly-crystalline β-Sn-like phase (compare with Extended Data Fig. 4a).

Extended Data Fig. 6 Describing VHDA formation and crystallization requires quantum-accurate simulations.

a, The results of our machine learning-driven simulation, with the collapse to VHDA between 12 and 13 GPa, and the crystallization between 15 and 20 GPa (presented in Fig. 2 and shown here for comparison). b, As in a, now using the empirical Stillinger–Weber (SW) potential41, which had been the state of the art for 100,000-atom simulations of silicon so far. Here, neither VHDA formation nor the subsequent crystallization are observed.

Extended Data Fig. 7 The enthalpy landscape of metastable disordered forms of silicon.

a, Computed enthalpy of 100,000-atom systems, ΔH, given relative to the respective most stable crystalline form at any given pressure (diamond-type → β-Sn-type → sh); see Methods. The red line shows the result for snapshots along the 500 K compression trajectory. Square symbols indicate results for snapshots, which have been frozen-in by rapid molecular-dynamics quenching (over 1 ps) and subsequently relaxed with a conjugate-gradient algorithm, all at the given external pressure (relevant structures are visualized below). The shaded area is a guide to the eye and corresponds to the enthalpy difference between the 500 K and fully relaxed a-Si structures at 0 GPa. Relevant structures are shown: note the near-perfect ordering of layers in the polycrystalline (‘pc’) sample. est., estimated; at−1, per atom. b, Enthalpy changes associated with the structural changes during compression. Copies of the 10 GPa structure (LDA/HDA polyamorph) were relaxed with increased external pressure (open symbols); this direct relaxation fixes the structure in place and does not allow it to transform to VHDA. A direct comparison between two competing phases at 13 GPa is therefore possible (labelled as ΔH1) and indicates the preference for VHDA formation. The enthalpy is lowered much further upon crystallization (ΔH2). c, Relaxation of copies of the pc-sh structure with decreased external pressure, mirroring decompression of a sample in experiment. The relative enthalpic stability over a relatively wide pressure range is qualitatively consistent with the observation of a hysteresis upon decompression: for example, in a previous work10, the LDA phase was fully recovered only after decompression to about 4 GPa. A dashed vertical line in b and c emphasizes the change in the crystalline reference, from diamond-type (dia) to β-Sn-type silicon.

Extended Data Fig. 8 Computed structure factors.

a, The static structure factor, S(q), as a probe for medium-range structural order, has been evaluated for the fully relaxed amorphous system. The computed result, including the height of the first sharp diffraction peak (FSDP), is in excellent agreement with previous experimental data90. The inset shows a radial distribution function, g(r), for the same structure, indicating long-range correlations beyond the first nanometre, which our machine learning-driven simulations can access. A dashed line at approximately 11 Å illustrates the limit of DFT modelling (half the cell length of the smallest system sketched in Extended Data Fig. 1). Expt., experimental. b, Computed structure factors during quenching. Left, the evolution of simulated structure factors through the relevant part of the liquid-quenching trajectory in the vicinity of the glass transition, plotted in 1-K temperature increments. The emergence of the FSDP (between 1.5 and 2.0 Å−1), as well as the structuring of the third peak (between 5 and 6 Å−1), are clearly visible. Right, detail view for 1.0 Å−1 ≤ q ≤ 4.0 Å−1 of the evolution of the FSDP with decreasing temperature, using the same colour scale as on the left-hand side.

Extended Data Fig. 9 Tight-binding DOS for an ultralarge system.

Supplementing the tight-binding electronic-structure computations in Fig. 3a–d, this figure shows the electronic DOS computed with the same approach but for a diamond-type crystalline silicon supercell at atmospheric pressure, containing >2 million atoms (see details in Methods). The energy scale is set by ref. 70.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Deringer, V.L., Bernstein, N., Csányi, G. et al. Origins of structural and electronic transitions in disordered silicon. Nature 589, 59–64 (2021). https://doi.org/10.1038/s41586-020-03072-z

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41586-020-03072-z

This article is cited by

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing