Bypassing the Single Junction Limit with Advanced Photovoltaic Architectures.

Multijunction devices and photon up- and down-conversion are prominent concepts aimed at increasing photovoltaic efficiencies beyond the single junction limit. Integrating these concepts into advanced architectures may address long-standing issues such as processing complexity, microstructure control, and resilience against spectral changes of the incoming radiation. However, so far, no models have been established to predict the performance of such integrated architectures. Here, a simulation environment based on Bayesian optimization is presented, that can predict and virtually optimize the electrical performance of multi-junction architectures, both vertical and lateral, in combination with up- and down-conversion materials. Microstructure effects on performance are explicitly considered using machine-learned predictive models from high throughput experimentation on simpler architectures. Two architectures that would surpass the single junction limit of photovoltaic energy conversion at reasonable complexity are identified: a vertical "staggered half octave system," where selective absorption allows the use of 6 different bandgaps, and the lateral "overlapping rainbow system" where selective irradiation allows the use of a narrowband energy acceptor with reduced voltage losses, according to the energy gap law. Both architectures would be highly resilient against spectral changes, in contrast with two terminal multi-junction architectures which are limited by Kirchhoff's law.

Understanding Causalities in Organic Photovoltaics Device Degradation in a Machine-Learning-Driven High-Throughput Platform.

Organic solar cells (OSCs) now approach power conversion efficiencies of 20%. However, in order to enter mass markets, problems in upscaling and operational lifetime have to be solved, both concerning the connection between processing conditions and active layer morphology. Morphological studies supporting the development of structure-process-property relations are time-consuming, complex, and expensive to undergo and for which statistics, needed to assess significance, are difficult to be collected. This work demonstrates that causal relationships between processing conditions, morphology, and stability can be obtained in a high-throughput method by combining low-cost automated experiments with data-driven analysis methods. An automatic spectral modeling feeds parametrized absorption data into a feature selection technique that is combined with Gaussian process regression to quantify deterministic relationships linking morphological features and processing conditions with device functionality. The effect of the active layer thickness and the morphological order is further modeled by drift-diffusion simulations and returns valuable insight into the underlying mechanisms for improving device stability by tuning the microstructure morphology with versatile approaches. Predicting microstructural features as a function of processing parameters is decisive know-how for the large-scale production of OSCs.