Toward Predicting Efficiency of Organic Solar Cells via Machine Learning and Improved Descriptors

Abstract

include the strong electron-electron interactions and strong electron-phonon couplings as well as the complicated donor/ acceptor (D/A) interface morphology, which are fundamentally different from inorganic semiconductors. [23,24] Therefore, the accurate simulating of OPVs requires high-level theoretical methods in quantum chemistry, quantum dynamics and statistical mechanics, and in recent years there have been substantial progress for the theoretical understanding of many microscopic processes such as charge transport, [25] exciton dissociation, [26,27] singlet fission, [28][29][30] etc. These methods are useful for few benchmark systems but cannot be used to explore the chemical space, i.e., screen a large number of candidate materials.The predictive power of a theoretical model by high-throughput virtual screening has been explored in the recent years. [31][32][33][34][35][36][37][38][39][40] For example, in the Harvard Clean Energy Project (CEP), [41] Aspuru-Guzik and co-workers have screened ≈2.3 million compounds by employing the Scharber model [42,43] to find out efficient new donor molecules for OPVs. Here, the computed energies of the highest occupied molecular orbital (HOMO)/the lowest unoccupied molecular orbital (LUMO) of organic molecules are calibrated to experimentally determined values, and then employing an averaging scheme, energies of HOMO/LUMO are estimated to use them as input in the Scharber model. However, the prediction of PCE of an OPV is much more challenging compared to the energies of orbitals. Very recently, the same research team [44] noticed that the Scharber model, widely used in these virtual screening works, is not accurate enough for the prediction of PCE, and calibration of PCE by considering molecular similarity, molecular weight and band gap energy leads to improvement in correlation between experimental (49 OPVs) and calculated PCE (Pearson's coefficient (r) is increased from 0.30 to 0.43). Further, the Scharber model is only optimized for the PC 61 BM acceptor [42] and a more general model is desired to take account of nonfullerene molecules.In an OPV, energy conversion is accomplished by four consecutive steps: i) absorption of photons and exciton formation, ii) exciton diffusion to D/A interface, iii) exciton dissociation, and iv) transport of holes/electrons to the respective electrode. Taking account of all these processes, the efficiency of a device depends on many microscopic properties of the organic material such as optical gap, charge-carrier mobility, ionization To design efficient materials for organic photovoltaics (OPVs), it is essential to identify the largest number of parameters that control their properties and build a model using these parameters (known as descriptors) for the prediction of the power conversion efficiency (PCE). By constructing a dataset for 280 small molecule OPV systems, it is found that for all high-performing devices, frontier molecular orbitals of donor molecules are nearly degenerated and in such cases, orbitals other than just high...

A Critical Review of Machine Learning of Energy Materials

“…For example, the coefficients in simple linear regression model as well as the logistic regression can provide an indication of the relative importance of different features. Feature importance analysis in tree‐based models has been routinely used to assess the important physical parameters that are related to the predictive targets . For deep learning‐based models, visualizing the hidden layer activations has been found to give interpretable chemical intuition and accurate mapping of structural space .…”

“…While the Scharber model is widely used for computing SCE, Sahu et al have realized, from a data set of 280 small molecule OPV systems, that for all high‐performance devices, frontier orbitals of donor molecules are nearly degenerated and therefore orbitals other than just HOMO and LUMO should be considered. The authors considered 13 quantum‐mechanical descriptors including 1) number of unsaturated atoms in the main conjugation path of donor molecules(), 2) polarizability of donor molecules, 3) the energetic differences of LUMO and LUMO+1 of donor molecules (Δ L ), 4) the energetic differences of HOMO and HOMO−1 of donor molecules(Δ H ), 5) vertical ionization potential of donor molecules (IP(ν)), 6) reorganization energy for holes in donor molecules (λ h ), 7) hole–electron binding energy in donor molecules( E bind ), 8) the energetic difference of LUMO of donor and LUMO of acceptor(), 9) the energetic difference of HOMO of donor and LUMO of acceptor(), 10) energy of the electronic transition to a singlet excited state with the largest oscillator strength( E g ), 11) change in dipole moment in going from the ground state to the first excited state for donor molecules(Δ ge ), 12) energy of the electronic transition to the lowest‐lying triplet state(), and 13) the energetic difference of LUMO and LUMO+1 of acceptors ().…”

Performance and Matching Band Structure Analysis of Tandem Organic Solar Cells Using Machine Learning Approaches

“…Recently, machine‐learning technologies have revolutionized the solar energy fields and played an important role in designing the optoelectronic materials and predicting OSC performances. Previous studies demonstrated that the machine‐learning approaches (e.g., Gaussian processes regression, gradient‐boosting regression tree, artificial neural network, and Random Forest regression) can be used to investigate the relationship between machine‐learning predicted and experimental measured (or density functional theory calculated) PCE values for electronic and device properties of OSCs . To discover efficient photovoltaic materials, machine‐learning‐assisted technology has been used for designing promising molecules and screening conjugated polymers for OSCs .…”

“…Previous studies demonstrated that the machine-learning approaches (e.g., Gaussian processes regression, gradientboosting regression tree, artificial neural network, and Random Forest regression) can be used to investigate the relationship between machine-learning predicted and experimental measured (or density functional theory calculated) PCE values for electronic and device properties of OSCs. [18][19][20][21] To discover…”

Machine Learning for Understanding the Relationship between the Charge Transport Mobility and Electronic Energy Levels for n‐Type Organic Field‐Effect Transistors

“…The RMSE of Random Forest is 2.13, for cross-validated mean, 1.91, for test set. [27] As discussed in the principles of charge transportation in OFETs, [30] molecular frontier energies of n-type materials in the n-type OFETs play a crucial role in effective electrons injection, as the electron mobility can be optimized by reducing small electron injection barrier height. The Gradient Boosting model also has performance with RMSE of 2.36 and 2.33 for cross-validated train set and test set, which is insufficient for practical use.…”