Proceedings of International Conference on Hybrid and Organic Photovoltaics (HOPV23)
DOI: https://doi.org/10.29363/nanoge.hopv.2023.201
Publication date: 30th March 2023
Abstract
The recent successes of emerging photovoltaics such as organic and perovskite solar cells are largely driven by innovations in material science. However, closing the gap to commercialization still requires significant innovation to match contradicting requirements such as performance, longevity and recyclability in one and the same material. In this perspective, we start with the notion that the learning curve in innovation must be substantially increased to achieve commercialization in the presence of a mature competitor. Then, we show that the learning curve, as of today, is limited by a lack of design principles linking detailed chemical structure to microscopic structure, and by an incapacity to experimentally access microscopic structure from investigating macroscopic devices. Both limitations in turn are caused by an individualist approach to learning, not being able to produce datasets large enough to find patterns leading us to breakthrough innovations. In this Perspective, we propose a layout of a Digital Twin For PV Materials able to remove both limitations.
Microscopic design principles are required to get handles for a true multi-objective optimization leading to PV devices with unseen properties. We argue that such microscopic design principles can principally not be achieved by either a pure knowledge based or a pure data-driven approach. To solve the problem, our digital twin layout combines ideas from acceleration platforms in material science with digital twin concepts from the engineering world. To tackle the specific challenges in material science, we propose to use ad-hoc trained local surrogates of complex solid state models to achieve self-calibration of simple proxy experiments onto the underlying physics. This allows to build fast but still detailed predictive models across scales, so that inverse design, from the desired performance of a working device to the chemical structure, becomes possible. We show the building blocks that are already available and comment on active research closing the remaining gaps. Finally we propose to overcome the individualist learning approach by adopting the novel paradigm of a federated learning approach, able to protect investments and IP aspects while at the same time maximizing the learning rate for all stakeholders.
We highly acknowledge funding from FAU Solar, the Erlangen Research Cluster on accelerating innovation in Carbon-based solar energies