Interpretable deep learning for guided microstructure-property explorations in photovoltaics

Abstract

The microstructure determines the photovoltaic performance of a thin film organic semiconductor film. The relationship between microstructure and performance is usually highly non-linear and expensive to evaluate, thus making microstructure optimization challenging. Here, we show a data-driven approach for mapping the microstructure to photovoltaic performance using deep convolutional neural networks. We characterize this approach in terms of two critical metrics, its generalizability (has it learnt a reasonable map?), and its intepretability (can it produce meaningful microstructure characteristics that influence its prediction?). A surrogate model that exhibits these two features of generalizability and intepretability is particularly useful for subsequent design exploration. We illustrate this by using the surrogate model for both manual exploration (that verifies known domain insight) as well as automated microstructure optimization. We envision such approaches to be widely applicable to a wide variety of microstructure-sensitive design problems.

Better, Faster, and Less Biased Machine Learning: Electromechanical Switching in Ferroelectric Thin Films

“…This mentality often leads to a challenging interpretation of outputs, which often results into incorporation of noise, and description of spurious contributions as physical phenomena. [33][34][35] This "interpretability" challenge has become a central focus of many recent publications, [36][37][38] and is directly correlated with the ML techniques' strictly mathematical nature: They inherently provide results lacking a physical basis. Traditional approaches to materials science indeed are based on interpretation of correlated datasets through the lens of physical and/or chemical behaviors-aspects fundamentally absent in ML methods.…”

A Deep Learning Framework for Design and Analysis of Surgical Bioprosthetic Heart Valves

“…For example, researchers have modified the loss functions to ensure some physical constraints are satisfied 41–45 . There has also been work on interpreting the predictions of the deep learning model based on physical conditions 22,23 . Incorporating partial differential equations ( PDEs ) in deep learning models : The key idea is to use the underlying governing equations such as Berger’s equation, Navier-Stokes equation, Cahn-Hillard’s equation, etc. to compute the residual for the sample.…”

“…On the contrary, the inverse problem, of defining the displacement of a given geometry and predicting the set of physics conditions is often ill-posed and could be consistently modeled as a generative model. There are several works showing the capability of Deep Learning methods to act as a surrogate 23,41,49,51 . These surrogates are modeled as distinctive (non-generative) networks, since the physics is deterministic and the problem is well-posed.…”

“…For example, researchers have modified the loss functions to ensure some physical constraints are satisfied 41–45 . There has also been work on interpreting the predictions of the deep learning model based on physical conditions 22,23 .…”

“…Deep learning has emerged as a major machine learning paradigm that has demonstrated transformative potential in many areas of science and engineering 18 . In sciences, the applications range from astronomy 19 , high-energy physics 20 and material science 21–23 to medical diagnostics 24 and plant sciences 25 . In the domain of engineering, deep learning is the key enabler of the recent autonomous driving revolution 26 along with other significant progresses in robotics 27 , design and manufacturing 28,29 and prognostics and health management of engineered systems 30 .…”