Author Affiliations
1School of Optoelectronic Science and Engineering, Soochow University, Suzhou 215006, Jiangsu , China2Key Lab of Advanced Optical Manufacturing Technologies of Jiangsu Province & Key Lab of Modern Optical Technologies of Education Ministry of China, Suzhou 215006, Jiangsu , Chinashow less
Fig. 1. Combination of intelligent algorithms such as machine learning and multi-physical field design of photovoltaic devices. Neural networks are not only used for forward prediction (blue circles), but also for reverse design (red circles), achieving a comprehensive design from the material, structure, and other parameters to the photoelectric response
Fig. 2. Structure and energy level diagram of perovskite solar cells. (a) Structural diagram; (b) energy level diagram
Fig. 3. Influence of the parameters such as the thickness of each layer of PSCs on their performance. (a) Absorption contour map within the working band, as the thickness of the perovskite layer changes from 450 nm to 900 nm; (b) plot of the percentage stacking area of absorption and reflection of each layer under illumination; (c) contour map of the macroscopic changes in the current-voltage curve, as the electronic lifetime of the perovskite layer increases from 0.0005 to 20; (d) effect of the electronic lifetime of the perovskite layer on the device temperature
Fig. 4. Machine learning models and their predictive effects. (a) Importance ranking of the input features in the decision tree model; (b) underlying decision tree; (c) cross validation results of 50 random forests; (d) learning curves of the random forests during 200 iteration times
Fig. 5. Neural network model and its prediction results. (a) Neural network structure diagram; (b) loss value measured by MSE during the training and testing process; (c) comparison between the predicted current-voltage curves and those calculated by the optoelectronic coupling model; (d) comparison of current-voltage curves in the literature and current-voltage curves predicted by machine learning models
Fig. 6. Regression results between predicted and true values of four important parameters of perovskite solar cells. (a) Short circuit current; (b) open circuit voltage; (c) fill factor; (d) conversion efficiency
Fig. 7. Comparison between the current-voltage curve fitted by the microscopic parameters of the optimized algorithm design and the actual curve
Parameter | Symbol | Unit | Perovskite | ETL | HTL |
---|
Electron affinity | χ | eV | 3.93 | 4 | 2.5 | Band gap | Eg | eV | 1.55 | 3.2 | 2.7 | Relative permittivity | εr | | 6.5 | 9 | 3 | Effective DOS for electron | NC | cm-3 | 0.23m0 | 9m0 | 1×1019 | Effective DOS for hole | NV | cm-3 | 0.29m0 | 3m0 | 1×1019 | Radiative recombination coefficient | C | cm3·s-1 | 3.27×10-11 | 0 | 0 |
|
Table 1. Key material parameters for the optoelectronic coupling simulation
Parameter | Layer | Input | Inverse design output |
---|
Thickness /nm | Glass | 97 | 93 | FTO | 342 | 319 | ETL | 34 | 39 | Perovskite | 564 | 572 | HTL | 153 | 142 | Electron mobility | ETL | 0.458 | 0.452 | Perovskite | 1.6 | 11.81 | HTL | 0.985 | 0.93 | Electronic lifetime /μs | ETL | 0.0005 | 0.0005 | Perovskite | 10.1 | 15.76 | HTL | 0.0075 | 0.01 | Doping concentration | ETL | 6.79×1012 | 6.83×1012 | Perovskite | 1.71×1016 | 1.99×1016 | HTL | 6.5×1018 | 6.12×1018 |
|
Table 2. Comparison between reverse deduction and true values of five structural parameters and nine electrical parameters