Recent advances in machine learning (ML) have enabled predictive programs for photovoltaic characterization, optimization, and materials discovery. Despite these advances, the standard photovoltaic materials development workflow still involves manually performing multiple characterization techniques on every new device, requiring significant time and expenditures. One barrier to ML implementation is that most models reported to date are trained on computer simulated data, due to the difficulty in experimentally collecting the massive data sets needed for model training, limiting the ability to assess the limitations and validity of these methods, as well as to access new potential physical mechanisms absent in simulations. Herein, several neural networks trained on experimental data from PbS colloidal quantum dot thin‐film solar cells are introduced. These models predict multiple, complex materials properties, including carrier mobility, relative photoluminescence intensity, and electronic trap‐state density, from a single, simple measurement: illuminated current–voltage curves. The measurement system considers the spatial distribution of the materials parameters to gather and predict large amounts of data by treating an inhomogeneous device as a series of thousands of micro‐devices, a novel feature compared to existing solutions. This model can be extended to other materials and devices, accelerating development times for new optoelectronic technologies.
- Award ID(s):
- 1807342
- PAR ID:
- 10412778
- Date Published:
- Journal Name:
- 2022 IEEE 49th Photovoltaics Specialists Conference (PVSC)
- Page Range / eLocation ID:
- 0862 to 0866
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
-
more » « less
Abstract Machine learning (ML) has become a valuable tool to assist and improve materials characterization, enabling automated interpretation of experimental results with techniques such as X-ray diffraction (XRD) and electron microscopy. Because ML models are fast once trained, there is a key opportunity to bring interpretation in-line with experiments and make on-the-fly decisions to achieve optimal measurement effectiveness, which creates broad opportunities for rapid learning and information extraction from experiments. Here, we demonstrate such a capability with the development of autonomous and adaptive XRD. By coupling an ML algorithm with a physical diffractometer, this method integrates diffraction and analysis such that early experimental information is leveraged to steer measurements toward features that improve the confidence of a model trained to identify crystalline phases. We validate the effectiveness of an adaptive approach by showing that ML-driven XRD can accurately detect trace amounts of materials in multi-phase mixtures with short measurement times. The improved speed of phase detection also enables in situ identification of short-lived intermediate phases formed during solid-state reactions using a standard in-house diffractometer. Our findings showcase the advantages of in-line ML for materials characterization and point to the possibility of more general approaches for adaptive experimentation.
-
more » « less
Abstract Significant work has been directed at measuring the exciton diffusion length (
L D) in organic semiconductors due to its significance in determining the performance of photovoltaic cells. Several techniques have been developed to measureL D, often probing photoluminescence or charge carrier generation. Interestingly, in this study it is shown that when different techniques are compared, both the diffusive behavior of the exciton and active carrier recombination loss pathways can be decoupled. Here, a planar heterojunction device based on the donor–acceptor pairing of boron subphthalocyanine chloride‐C60is examined using photoluminescence quenching, photovoltage‐, and photocurrent‐basedL Dmeasurement techniques. Photovoltage yields the device relevantL Dof both active materials as a function of forward bias subject to geminate recombination losses. These values are used to accurately predict the photocurrent as a function of voltage, suggesting geminate recombination is the dominant mechanism responsible for photocurrent loss. By combining these measurements with photocurrent and photoluminescence quenching, the intrinsicL D, as well as the voltage‐dependent charge transfer state dissociation and charge collection efficiencies are quantitatively determined. The results of this work provide a method to decouple all relevant loss pathways during photoconversion, and establish the factors that can limit the performance of excitonic photovoltaic cells. -
more » « less
Abstract Mobilities and lifetimes of photogenerated charge carriers are core properties of photovoltaic materials and can both be characterized by contactless terahertz or microwave measurements. Here, the expertise from fifteen laboratories is combined to quantitatively model the current‐voltage characteristics of a solar cell from such measurements. To this end, the impact of measurement conditions, alternate interpretations, and experimental inter‐laboratory variations are discussed using a (Cs,FA,MA)Pb(I,Br)3halide perovskite thin‐film as a case study. At 1 sun equivalent excitation, neither transport nor recombination is significantly affected by exciton formation or trapping. Terahertz, microwave, and photoluminescence transients for the neat material yield consistent effective lifetimes implying a resistance‐free JV‐curve with a potential power conversion efficiency of 24.6 %. For grainsizes above ≈20 nm, intra‐grain charge transport is characterized by terahertz sum mobilities of ≈32 cm2V−1s−1. Drift‐diffusion simulations indicate that these intra‐grain mobilities can slightly reduce the fill factor of perovskite solar cells to 0.82, in accordance with the best‐realized devices in the literature. Beyond perovskites, this work can guide a highly predictive characterization of any emerging semiconductor for photovoltaic or photoelectrochemical energy conversion. A best practice for the interpretation of terahertz and microwave measurements on photovoltaic materials is presented.
-
Abstract Modern machine learning (ML) and deep learning (DL) techniques using high-dimensional data representations have helped accelerate the materials discovery process by efficiently detecting hidden patterns in existing datasets and linking input representations to output properties for a better understanding of the scientific phenomenon. While a deep neural network comprised of fully connected layers has been widely used for materials property prediction, simply creating a deeper model with a large number of layers often faces with vanishing gradient problem, causing a degradation in the performance, thereby limiting usage. In this paper, we study and propose architectural principles to address the question of improving the performance of model training and inference under fixed parametric constraints. Here, we present a general deep-learning framework based on branched residual learning (BRNet) with fully connected layers that can work with any numerical vector-based representation as input to build accurate models to predict materials properties. We perform model training for materials properties using numerical vectors representing different composition-based attributes of the respective materials and compare the performance of the proposed models against traditional ML and existing DL architectures. We find that the proposed models are significantly more accurate than the ML/DL models for all data sizes by using different composition-based attributes as input. Further, branched learning requires fewer parameters and results in faster model training due to better convergence during the training phase than existing neural networks, thereby efficiently building accurate models for predicting materials properties.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/PVSC48317.2022.9938867
