Fig. 1

Elements of graph-based microstructure design framework. a The concept of representing morphology as an equivalent graph allows construction of fast graph based surrogates to full-physics based device performance indicators; b Scheme of population based incremental learning: An initial (or prior) distribution is sampled to generate candidate graphs (i.e., microstructures), which are rapidly evaluated and ranked. This ranked information is used to update the distribution. Iterating this process evolves the distribution towards graphs exhibiting better performance. Such an approach naturally allows incorporation of expert information into the prior