Powder x-ray diffraction analysis with machine learning for organic-semiconductor crystal-structure determination.

This article focuses on a novel machine-learning (neural-network) method to determine the crystal structures of organic semiconductors from powder x-ray diffraction (PXRD) patterns, a task traditionally challenging due to peak overlap and noise. The method was applied to two high-mobility organic semiconductors, 3,9-dihexyldinaphtho[2,3-b:2′,3′-d]thiophene (C6-DNT-VW) and 2,7-dioctyl[1]benzothieno[3,2-b][1]benzothiophene (C8-BTBT), successfully identifying peak positions and enabling crystal-structure determination via Rietveld analysis without requiring single-crystal samples. The approach leverages a neural network trained on simulated PXRD data from over 2,000 organic crystal structures and demonstrated accuracy comparable to single-crystal x-ray diffraction results. This technique promises to accelerate the development of organic semiconductor materials by facilitating rapid crystal-structure analysis from powder samples and may be extendable to other complex crystalline materials.