Materials Engineers

DeepMind's GNoME (November 2023) produced 2.2 million stable crystal structure predictions in a single computational run — 45x the total prior human discovery record accumulated over centuries — using a graph neural network trained on the Materials Project database. Microsoft's MatterGen (2024) extends this by generating novel crystal structures with targeted properties on demand, not just screening known chemical spaces. These platforms are being integrated with automated synthesis pipelines (Lawrence Berkeley's A-Lab) to close the loop from prediction to physical validation without human-mediated design iteration.

A new generation of universal ML interatomic potentials — MACE-MP-0 (Cambridge, released 2023), CHGNet (Berkeley, 2023), M3GNet (2022), SevenNet (2024) — are trained across the entire periodic table and achieve near-DFT accuracy for energy, force, and stress predictions at computational costs 1,000-100,000x lower than DFT. These models enable molecular dynamics simulations at timescales and system sizes previously impossible (microseconds, millions of atoms), and enable property screening at a scale that collapses the traditional experimental iteration cycle. MACE-MP-0 in particular is already being used as a drop-in DFT replacement for high-throughput screening campaigns in academic and industrial labs.