Prediction of Stable Morphology of Block Copolymers by using SCF Calculation and Deep Learning

Block copolymers show various microphase separated structure depending on the chain architecture, and miscibility (chi parameter) between different segment type. Various stable phase such as lamellar, double gyroid, hexagonal cylinder and bcc sphere are known, and phase diagram has been studied for simple block copolymers by SCF calculation. However, it is not simple to find equilibrated morphologies even by the calculation, because many metastable morphologies are obtained. Usually, it takes large computational resource to obtain stable structure from many possible metastable structures by real space SCF calculation.
We applied 3D CNN deep learning technique to predict stable morphology from metastable morphology obtained from the SCF calculation without initial constraint. Metastable morphology of diblock copolymer and stable morphologies derived from well-known phase diagram are used for training sample. The optimized deep learning network can predict stable morphology of diblock copolymers of arbitral volume fraction and chi parameter.