{"id":52771,"date":"2025-11-18T15:19:53","date_gmt":"2025-11-18T20:19:53","guid":{"rendered":"https:\/\/engineering.jhu.edu\/materials\/?post_type=news&p=52771"},"modified":"2025-11-19T12:45:02","modified_gmt":"2025-11-19T17:45:02","slug":"revolutionizing-electron-mapping-with-ai","status":"publish","type":"news","link":"https:\/\/engineering.jhu.edu\/materials\/news\/revolutionizing-electron-mapping-with-ai\/","title":{"rendered":"Revolutionizing Electron Mapping with AI"},"content":{"rendered":"

An assistant professor of\u00a0<\/span>materials science and\u00a0engineering<\/span><\/a>\u00a0at\u00a0Johns\u00a0Hopkins\u00a0has\u00a0developed\u00a0an\u00a0artificial intelligence framework\u00a0that can\u00a0map the electrons\u00a0for\u00a065\u00a0elements\u00a0on the periodic table\u00a0and their\u00a0combinations, accelerating the development of better\u00a0computer\u00a0chips, solar\u00a0cells, LED lights, and more.\u00a0How his creation, called\u00a0SlaKoNet,\u00a0works\u00a0is\u00a0explained\u00a0in\u00a0<\/span>The Journal of Physical\u00a0Chemistry Letters<\/span><\/i><\/a>.<\/span>\u00a0<\/span><\/p>\n

\u201cWe needed\u00a0faster\u00a0predictions for electronic\u00a0movement in materials, which will\u00a0determine\u00a0how they move in small chips\u00a0and ultimately improve how\u00a0chips\u00a0are made,\u201d\u00a0says\u00a0<\/span>Kamal\u00a0Choudhary<\/span><\/a>, who holds a joint appointment in the\u00a0<\/span>Department of Electrical and Computer Engineering.<\/span><\/a>\u00a0\u201cTechnology\u00a0chips\u00a0include many layers of materials, and we need to know how the electrons\u00a0will flow\u00a0and\u00a0if adding an element to the chip would make them flow faster.\u00a0This model can accurately predict that for those 65 elements.\u201d<\/span>\u00a0<\/span><\/p>\n

Instead of creating something new, Choudhary used AI\u00a0<\/span>to upgrade the long-standing Slater-Koster tight-binding formalism, a method for predicting electronic band structures.\u00a0These\u00a0structures\u00a0control electrical behavior by allowing energy from electrons to flow within a material.<\/span>\u00a0<\/span><\/p>\n

\u201cThe\u00a0Slater-Koster tight-binding formalism\u00a0requires\u00a0manually\u00a0set parameters\u00a0to predict\u00a0how\u00a0electricity will flow through a material, which\u00a0is\u00a0tedious and prone to error,\u201d\u00a0says Choudhary.\u00a0\u201cSlaKoNet\u00a0is a framework that modernizes the classical method,\u00a0modeling\u00a0combinations of electrons\u00a0to discover\u00a0which\u00a0blend\u00a0performs\u00a0the best.\u201d<\/span>\u00a0<\/span><\/p>\n

The model\u00a0retains\u00a0its basis in physics but\u00a0uses\u00a0a neural network\u00a0framework\u2014a type of artificial intelligence that\u00a0recognizes patterns and makes decisions\u00a0to\u00a0optimize\u00a0parameters of\u00a0prediction equations.\u00a0SlaKoNet\u00a0learns how electrons behave in many metals,\u00a0semiconductors, and insulators,\u00a0easily\u00a0adapts\u00a0to new datasets, and\u00a0is\u00a0eight times faster than the standard\u00a0central processing unit (CPU)\u00a0calculation method.<\/span>\u00a0<\/span><\/p>\n

\u201cIt\u00a0was trained on large datasets,\u00a0<\/span>particularly the\u00a0JARVIS-DFT\u00a0TBmBJ\u00a0dataset from the National Institute of Standards and\u00a0Technology,\u00a0because it provides a highly accurate foundation for learning,\u201d he says. \u201cIn total, the model was trained\u00a0on\u00a0roughly 20,000\u00a0material combinations across 65 elements, including oxides, carbides, nitrides, halides, and\u00a0intermetallics. This breadth ensures that its predictions\u00a0remain\u00a0reliable across a wide range of materials.\u201d<\/span>\u00a0<\/span><\/p>\n

SlaKoNet\u00a0currently includes most materials that are used\u00a0in quantum and semiconductor technologies, but\u00a0Choudhary plans to extend it to the full periodic table.\u00a0\u201cThe long-term goal is to make a universal model for all elements and their combinations,\u201d\u00a0says Choudhary.<\/span>\u00a0<\/span><\/p>\n

Now, the\u00a0model\u00a0is live on\u00a0Choudhary\u2019s ChatGPT platform for materials scientists called\u00a0atomgpt.org. \u201cMy\u00a0ultimate goal\u00a0is to make\u00a0everything\u00a0open access for\u00a0research, so\u00a0SlaKoNet\u00a0is\u00a0freely\u00a0available for scientists in academic and industry\u00a0environments,\u201d he says.<\/span><\/span><\/p>\n