Jueves, 27 de agosto de 2026

Aube.

Las noticias del progreso
LaboratorioFuente única

MIT framework reaches 68% simulated stability in crystal designs

Idiomas de este artículo
Original · ENFR

Texto original en inglés. 2 idiomas disponibles, el tuyo se añade con un clic.

A materials model can produce millions of designs in minutes. The harder part comes afterward: checking whether any of them can exist without falling apart. MIT researchers have built CrysVCD, a framework that applies chemical constraints before an AI generates the crystal structure, and their computational tests reached 68% mechanical stability.

The bottleneck is expensive. Stability validation can account for 90% of the computational cost of creating usable materials and take weeks or months, according to Mouyang Cheng, one of the MIT researchers. That burden favors companies with large computing budgets, while many smaller companies and research labs cannot afford it, potentially limiting innovation.

CrysVCD splits the work into two stages. A language model first produces chemically valid formulas, using rules about the electrons around a material’s atoms. A diffusion model — an AI system that builds an output through successive steps — then generates the crystal’s atomic structure from that formula. The constraint arrives at the beginning, rather than serving as a filter at the end.

The researchers say the framework worked with several material-generation models and produced stable materials an order of magnitude more efficiently than approaches that generate first and screen afterward. In their tests, it achieved 68% mechanical stability and 85% metastability; nearly 70% of computational generations passed the study’s stringent lattice-dynamics stability test.

So what changes in practice? Research teams could spend less computing time rejecting chemically invalid candidates and more time examining materials aimed at specific jobs, including high thermal conductivity or a high dielectric constant for computer chips and data centers. The gain is most direct for groups with limited computing resources, although the reported results are still computational: the study does not establish that these candidates have been manufactured or tested in products.

68%Crystalline materials achieving mechanical stability in computational tests

Fuentes — leer los originales(hora de París)

Phys.org — TechnologyEN
0000

Para leer a continuación

Comentarios

Cargando el hilo…

Inicia sesión para escribir un comentario. Iniciar sesión