Fig. 3: The workflow for inverse design of direct narrow-gap semiconductors targeting optoelectronic applications. | Nature Communications

Fig. 3: The workflow for inverse design of direct narrow-gap semiconductors targeting optoelectronic applications.

From: An invertible, invariant crystal representation for inverse design of solid-state materials using generative deep learning

Fig. 3

The inverse design workflow started from training a general recurrent neural network (RNN) on the Materials Project database to learn the syntax of SLICES, followed by training a specialized RNN by tuning the general RNN using a dataset of direct narrow-gap semiconductors. Then, the specialized RNN was used to generate ~10 million SLICES strings, which were reconstructed into ~3.4 million crystal structures. These crystal structures were filtered to identify new direct narrow-gap semiconductors.

Back to article page