The inverse design of superhard materials is often hindered by the vastness of the configurational space. In this work, we employ our recently developed ${\mathrm{D}}^3\mathrm{REAM}$ framework—an approach that integrates machine-learning potentials with optimization strategies—to enable automated and efficient discovery of superhard materials. Using ${\mathrm{D}}^3\mathrm{REAM}$, we identified carbon allotropes, ${\mathrm{C}}_8\left(\mathrm{C}2/\mathrm{m}\right)$ and ${\mathrm{C}}_8\left(\mathrm{I}{2}_1{2}_1{2}_1\right)$, with bulk moduli approaching 400 GPa. Furthermore, we found four carbon allotropes, ${\mathrm{C}}_4\left(\mathrm{I}{2}_1{2}_1{2}_1\right), {\mathrm{C}}_8\left(\mathrm{I}{2}_1{2}_1{2}_1\right), {\mathrm{C}}_8\left(\mathrm{P}\overline{1}\right)$, and ${\mathrm{C}}_8\left(\mathrm{C}2/\mathrm{m}\right)$, exhibiting uniaxial elastic moduli surpassing that of diamond. Extending the search to the complex B-C-N system, active learning revealed superhard phases ${\mathrm{B}}_2{\mathrm{C}}_5\mathrm{N}\left(\mathrm{Pmm}2\right), {\mathrm{B}}_2{\mathrm{C}}_3\mathrm{N}\left(\mathrm{Cm}\right)$, and ${\mathrm{B}}_2{\mathrm{C}}_{16}{\mathrm{N}}_4\left(\mathrm{P}1\right)$. Analysis of the B-C-N search space further indicates that boron-nitrogen incorporation up to 40% enables the design of superhard materials with enriched structural and mechanical diversity.

Close

PrevNext

View figure in article

• Hardness
• Carbon-based materials
• Data mining