Virtual Screening of Hole Transport, Electron Transport, and Host

Article pubs.acs.org/jcim

Cite This: J. Chem. Inf. Model. 2018, 58, 2440−2449

Virtual Screening of Hole Transport, Electron Transport, and Host Layers for Effective OLED Design Shao-Yu Lu,† Sukrit Mukhopadhyay,‡ Robert Froese,‡ and Paul M. Zimmerman*,† †

Department of Chemistry, University of Michigan, 930 North University Avenue, Ann Arbor, Michigan 48109, United States The Dow Chemical Company, Midland, Michigan 48674, United States

‡

J. Chem. Inf. Model. 2018.58:2440-2449. Downloaded from pubs.acs.org by IOWA STATE UNIV on 01/21/19. For personal use only.

S Supporting Information *

ABSTRACT: The alignment of energy levels within an OLED device is paramount for high efficiency performance. In this study, the emissive, electron transport, and hole transport layers are consecutively evolved under the constraint of fixed electrode potentials. This materials development strategy takes into consideration the full multilayer OLED device, rather than just individual components. In addition to introducing this protocol, an evolutionary method, a genetic algorithm (GA), is evaluated in detail to increase its efficiency in searching through a library of 30 million organic compounds. On the basis of the optimization of the variety of GA parameters and selection methods, an exponential ranking selection protocol with a high mutation rate is found to be the preferred method for quickly identifying the top-performing molecules within the large chemical space. This search through OLED materials space shows that the pyridine-based central core with acridine-based fragments are good target host molecules for common electrode materials. Additionally, weak electron-donating groups, such as naphthalene- and xylene-based fragments, appear often in the optimal electron-transport layer materials. Triphenylamine- and acridine-based fragments, due to their strong electron-donating character, were found to be good candidates for the holetransport layer.

■

blue pixels, the fluorescent emitters are still the desired choice due to high stability. More recently, Adachi demonstrated that organic molecules could undergo thermally activated delayed fluorescence (TADF), where small energy gaps (ΔEST) between the lowest-lying singlet (S1) and triplet (T1) excited states of the emitter led to similar quantum efficiencies as that of phosphorescent OLEDs.10,11 Various metal-free TADF materials, based on cyanobenzene, 12 triazine, 13,14 and triazole,15 have been designed to widen the scope of this strategy. Importantly, the energy levels of frontier molecular orbitals (FMOs) can be aligned to tune the concentration of electrons and holes in the EML, which in turn determines the width and position of the recombination zone, resulting in a greatly improved device efficiency and lifetime.16−18 Despite these important conceptual and engineering advances, many challenges remain to improve operating

INTRODUCTION Organic light-emitting diodes (OLEDs) were first described by Tang1 and have attracted significant attention for their utility in displays and energy-efficient lighting sources.2−4 A typical OLED has a multilayer structure, consisting of a series of stacked organic compounds. The key layers are the hole- and electron-transport layers (HTL, ETL) and an emissive layer (EML), which usually consists of host molecules doped with a small percentage of emitters. Specific materials for each layer can be selected to optimize the structural and electronic properties of OLED devices. In order to achieve the most efficient performance, including enabling a wide range of emitting wavelengths and minimal voltage requirements for high-intensity emission, the transport materials must work together as a highly tuned cohort. Advances in OLED performance5−7 have been achieved by improving charge transport within layers, charge injection across layers, and quantum efficiency of the emissive layer.8,9 Traditionally, phosphorescent emitters are used for green- and red-emitting pixels due to efficient harvesting of both singlet and triplet excitons and high internal quantum efficiency. For © 2018 American Chemical Society

Special Issue: Materials Informatics Received: January 25, 2018 Published: June 27, 2018 2440

DOI: 10.1021/acs.jcim.8b00044 J. Chem. Inf. Model. 2018, 58, 2440−2449

Article

Journal of Chemical Information and Modeling

Figure 1. OLED discovery pipeline: Diagram of the high-performance OLED discovery.

were systematically examined to determine the most effective algorithmic choices in this materials’ discovery process. Not only does this study provide a first-principles means to identify improved designs for OLED devices, it also gives useful operating conditions for the use of evolutionary methods in searching the vast chemical space of molecular libraries.

characteristics. Because high-efficiency OLEDs in the highenergy blue spectral range are still lacking, for example, a search through the chemical space of small organic molecules could identify new materials with shorter-wavelength emission.19,20 This chemical space, however, is far too large for exhaustive screening, making advances in the large-scale, computer-driven search for high-performance OLED materials of high importance. Recently, optimization strategies that combine highthroughput virtual screening with computer-aided design have been attracting attention for their potential to accelerate discovery and development of new materials.21−23 Many research groups have performed quantum chemical calculations together with high-throughput prescreening evolutionary methods, such as artificial neural networks and genetic algorithms (GAs).24−27 For instance, Aspuru-Guzik28 reported an algorithm to select potential candidates from a chemical set that included more than 1.6 million molecules and then screened over 400 000 using time-dependent density functional theory (TD-DFT) to identify improved TADF molecules. Shu et al.29 have also implemented a GA approach to generate ideal TADF molecules for emitters in the blue region. This study identified nearly 4000 promising candidates based upon a database of just over 1 million. Although these studies were successful, their focus was on TADF emitters, which is only one piece of the OLED device. A complete design, taking into consideration all layers of the device, is necessary to find the top-performing device architectures for efficient OLED devices (Figure 1). Recently, Kuttipillai et al.30 reported that the energy level alignment between the transporting layer and emissive host affects the charge carrier interface injection barriers and is therefore of great importance to achieving high external quantum efficiencies. While other factors such as the singlet−triplet gap of the emitter,10−15 charge recombination,16−18 and electron mobility affect OLED performance,8,9 our goal in this work is to address the energy alignments of the three key OLED layers, which up to this point have not been designed as an integrated whole. To address this challenge, we present an evolutionary strategy to design multilayer OLED devices by level-matching between the many layers. Three different core building blocks (quinazoline, pyridine, and triazine) along with 29 molecular fragments were used to generate numerous candidate molecules for the host, ETL, and HTL. These candidates were evaluated using the energy alignment between the active layers’ FMOs and the (fixed) Fermi level of electrodes with the top candidates identified using a GA. In all, this search examines a chemical space of 3 × 107 possible organic materials for use in OLED active layers. To improve the efficiency and speed of molecular design, the GA parameters and a variety of evolutionary selection methods

■

COMPUTATIONAL METHODS High-Performance OLED Design and Fitness Function. To begin designing high-performance OLED materials, the overarching principles governing device behavior will first be outlined. For high-efficiency emission, injected electrons and holes must cross device layers with minimal applied bias and high currents. At interfaces between layers, poorly aligned energy levels result in large thermal energy losses as carriers have difficulty passing through.31,32 To minimize the efficiency loss at anode/HTL, HTL/EML, EML/ETL, and ETL/cathode interfaces, the frontier orbital energy levels should be carefully aligned. Leakage can lead to degradation, loss of efficiency, and broadening of the emission spectra, leading to undesired chromaticity. Figure 2 shows the alignments expected for a low-bias, highly efficient device design. Under forward bias, charge

Figure 2. Schematic energy level diagram for OLED devices.

carriers are injected from the cathode and anode, where they are then transported to the emissive layer so as to recombine to generate a photon. For a device with poorly aligned energy levels, however, the charge carriers will need more energy to overcome injection barriers and therefore higher voltage to activate the light generation process. According to these key factors, the entire device should be designed as a cohort, and optimization of individual components alone is not the optimal procedure. One way of approaching this problem is by first focusing on the host molecule in the emissive layer and then choosing transport 2441


Article


Figure 3. Pictured fragment sets define the molecular space explored for the high-performance OLED devices (PLG is defined for the plug position and R and R1 are the socket positions).

layers to match its energy levels. Candidates for a good host will be evaluated through a scoring function 2

Fhost = 1000(e−((hostHOMO−ϕHOMO)

2

+ (hostLUMO −ϕLUMO) )/2β

SB(HTL LUMO) =

)

SB(ETL HOMO) =

(1)

2

2

S(ETL LUMO) = e−(ETL LUMO− hostLUMO)

/β

(1 + e

)

1 (−α(hostHOMO − ETL HOMO))

(1 + e

)

(3)

where α is set as a scaling factor. Two fitness functions combine the S and SB metrics and quantify the effectiveness of ETL and HTL materials

where β is a scaling factor. ϕHOMO and ϕLUMO are defined to be the idealized energy levels for the transport of holes and electrons across the device (−4.75 and −1.72 eV, respectively).33 The factor of 1000 fixes the optimal materials to approach a high score of 1000. By design, high fitness scores for the host correspond to minimal applied bias potentials. For each promising OLED device, fitness functions that optimize the FMO energy level matching between ETL, HTL, and host layers must be chosen.34−36 The score functions for the transport layers are therefore defined as S(HTL HOMO) = e−(HTL HOMO− hostHOMO)

1 (−α(hostLUMO − HTL LUMO))

FHTL = 1000S(HTL HOMO)SB(HTL LUMO) FETL = 1000S(ETL LUMO)SB(ETL HOMO)

(4)

The overall fitness score (FETL and FHTL) functions are dependent on the reference work function of the electrodes (which we assume are fixed) and the frontier orbital energy levels of host material. ETL and HTL can therefore be optimized for particular hosts to reveal specific combinations for best performance. Materials Library and Evolutionary Algorithm. To optimize OLED devices, candidates that maximize the fitness functions (F) must be identified from a huge chemical space. It is impossible, however, to screen astronomical numbers of organic molecules due to the high computational costs involved. Fortunately, an exhaustive exploration of chemical space is not necessary because the (relatively) local area of chemical space with promising molecules is reasonably understood. Specifically, relatively small organic molecules can be tuned for optimized FMO levels by introducing donor and acceptor groups. A wide variety of synthesizable

/β

(2)

where each score depends on the selected host energy levels. In addition to promoting transport of holes in the HTL and electrons in the ETL, the layers must block injection of the complementary charge carriers to prevent leakage. To stop hole transfer in the ETL, the ETL must have a deeper HOMO level compared to the HOMO of the host. Similarly, the HTL must have a high LUMO to stop electron transfer. Therefore, the score function for blocking is set up as 2442


Article


Figure 4. Three-layer molecular structure generating process.

Figure 5. Genetic operator schemes: (a) mutation and (b) crossover.

search, and a GA is therefore used in this work. Starting from a randomly chosen initial population, each potential molecule is built by attaching fragments to scaffolds, as shown in Figure 4. All molecules are brought into three dimensions through the assistance of OpenBabel.38 These molecules were optimized using a GA evolutionary procedure, evaluated for orbital energies, and then ranked based on their fitness score. During the evolution phase of the GA, the selection process for the high-ranking candidates dictates which genes will continue propagating through the population. Therefore, a handful of these methods were examined, including roulette wheel selection,39 tournament selection,40 and rank selection41 (see SI, part 1) to determine which of these techniques was preferred. Complicating this matter, each selection method has a variety of parameters that must be chosen. Thus, it is essential to fine-tune the selection methods and their parameters to provide a high-efficiency search as it is not obvious that a generic GA would be particularly effective for OLED device optimization.42,43 After the top candidates in a population are selected, crossover (at probability Pcross) and mutation (Pmut) operations are performed (Figure 5). The former represents inheritance, where genes are passed from the top candidates to the next generation, and the latter allows exploration of new areas of chemical space. Clearly, some tuning of these parameters is required to optimally cover a large region of chemical space

semiconducting molecules can therefore be envisioned by assembling a set of seed molecules and associated functional groups. To this end, quinazoline, pyridine, and triazine fragments were selected as common cores for the molecules, and several fragments for the functional groups were chosen to generate a library, including electron-withdrawing (e.g., the cyano and halide groups) and electron-donating groups (e.g., the alkyl, phenyl, and carbazolyl groups). The potential molecules are represented as molecular trees, which include (1) 3 different seeds, (2) 15 functional fragments for the first and second molecular segments, and (3) 14 capping fragments to form the third molecular segment (Figure 3). In these molecular building blocks, plug and socket positions are predefined as connection points to restrict the resulting materials to welldefined, synthesizable species. Each molecule therefore is assembled by connecting units at their plug and socket positions, which can be straightforwardly performed computationally. The synthetic accessibility of the candidates can be checked to ensure they are experimentally plausible,37 and the details of this procedure are shown in the SI. Overall, the potential candidates for HTL/host/ETL layers span a chemical space of 3 × 154 × 142 (or 29.7 million) unique candidate molecules. Due to the complexity of the library, maximizing the fitness of OLED devices is best performed using an evolutionary 2443


Article


host is saved after the evolution process to match them together into optimal OLED multilayer devices.

while still maintaining inheritance of good traits during evolution. Evolution Process for High-Performance OLED Search. The full strategy for OLED device design is shown in Figure 6. First, potential host candidates are generated as an

■

RESULTS AND DISCUSSION Fine-Tuning of GA Parameters. In order to realize the potential of the proposed GA methodology to generate highperformance OLEDs, a test set was built from several molecular fragments that have been shown to be highly useful in existing OLEDs. These OLED “active” fragments were chosen based on experimentally tested ETL, HTL, and host materials46−48 and will be used to validate both the GA procedure and the chosen fitness functions, as shown in Figure 7. Accordingly, the fragment library for this validation consists of two carbazole derivates from active host materials, three fragments (pyridine, naphthalene, triazine) based on ETLactive molecules, and three fragments (dimethyl fluorene, triphenylamines) from HTL-active molecules (see SI Figure S3). In these tests, benzene is used as a typical common core unit, and methyl, phenyl, and biphenyl groups were selected as capping fragments.49,50 All of these fragments are mixed in the benchmark library, spanning a chemical space of 1 × 84 × 32 (36 864) unique candidate molecules. The GA is used to individually optimize ETL, HTL, and host layers, resulting in OLED candidate molecules that are optimized within this set and, importantly, a test of whether the employed fitness functions are relevant to experiment. From 20 independent GA calculations of 40 generations each, the molecules in the top 10% of fitness scores were selected to count which “active” fragments appeared in these molecule structures (see SI Figure S4). The calculated results show that when the HTL is optimized, fragments from experimental HTL-active materials are preferentially selected. In other words, given a diversity of molecular fragments, the fitness function for the HTL works well in combination with the GA to recover the experimentally validated chemical motifs for HTL. Similarly, known ETL-active fragments are preferentially selected when optimizing for ELT, and host fragments follow the same trend, though not as strongly as with HTL/ETL optimization. The results shown that the GA fitness evolutionary process rediscovers the experimentally active OLED molecules on a layer-by-layer basis. In total, this provides substantial evidence that the chosen fitness functions for host, ETL, and HTL layers lead to relevant OLED materials.

Figure 6. Process flow of our evolutionary high-performance OLED design.

initial population for the GA process. The frontier orbital energy levels for each candidate need to be precisely calculated to evaluate the fitness function. To this end, Mukhopadhyay et al.44 reported a good correlation of HOMO/LUMO energies from B3LYP to experimental HOMO/LUMO energies from UPS spectroscopy and IPES measurements. In turn, MOPAC45 simulations at the PM6 level of theory are shown in the SI to produce HOMO/LUMO levels that correlate well with DFT values. The correlation shows that average unsigned errors (see SI Figures S1 and S2) between B3LYP and PM6 are 0.104 and 0.021 eV for the HOMO and LUMO, respectively. Using PM6 gives a great advantage for extensive screening, being orders of magnitude faster than DFT, allowing the host candidates to be quickly ranked based on the fitness function, Fhost, and optimized using the GA. For the topperforming host materials, matching ETL and HTL candidates are subsequently optimized. The GA process is repeated until the change in fitness score over one iteration is small (Fscore(i + 1) − Fscore(i) < 1) or a maximum number of 40 generations is reached. Finally, the top performance of each ETL, HTL, and

Figure 7. Process flow of our evolutionary high-performance OLED benchmark test. 2444


Article


Figure 8. Scatter plot of the HOMO and LUMO energy levels for 9450 candidate symmetric molecules computed in this study for the host.

Figure 9. Top 10 performing candidates of the symmetric host.

Figure 10. Best fitness score of the pool as a function of generation for a series of GA optimizations with (a) different mutation rates, (b) different parameter of L-rank used, (c) different parameter of R-rank used, and (d) different selection methods.

2445


Article


Figure 11. Top three performing candidates for designing sandwich OLED material by using the GA evolutionary process.

therefore stored for all candidate molecules, regardless of their presence in the current GA pool. For each parameter test set, 100 independent GA runs of 40 generations were carried out, and the average results are shown in Figure 10. Figure 10a presents the average of the best fitness scores as a function of generation for a series of optimizations with Pmut ranging from 0.05 to 1.0 and Npool equal to 100. The average best fitness scores rapidly converge by 20 generations for Pmut larger than 0.6. Even with forced mutations for duplicate structures, the high mutation probability is needed to increase diversity of ETL molecules during the evolutionary process, giving higher chances that the best structures are found. In Figure 10b,c, the linear ranking method (L-rank) and exponential ranking method (E-rank) are evaluated using parameters ranging from 0.2 to 1.0. Fitness scores improve with decreasing prefactors for both selection methods. The efficiency of these selection methods was then compared with other selection methods such as roulette wheel selection (RWS), stochastic universal sampling (SUS), and tournament selection. The results show that E-rank and RWS methods are optimal for the symmetrical OLED chemical space, consistently outperforming other methods for small numbers of generations. In all, E-rank with Pmut = 0.6 is selected for subsequent GA optimizations as it reaches the highest fitness metrics with fewer iterations than competing selection methods. Prediction of High-Performance OLED Materials. To apply the GA methodology to the larger, asymmetric chemical space of 30 million molecules, the search begins with 40 host generations from the symmetrical chemical space, where the fixed reference work function of electrodes is used to evaluate the fitness. On the basis of the average maximum fitness scores, hundreds of molecules with Fhost above 950 are available. To reach these high scores, the average HOMO and LUMO

Moving on from the experimental benchmark to an increased diversity of fragments, we considered the full set of functional group possibilities shown in Figure 3. Because the complete set including asymmetric structures is huge, symmetrical structures are first considered as a second test set before studying the full chemical space. This set consists of 3 × 152 × 14 (9450 unique molecules) and therefore can be thoroughly examined at relatively low cost to better interrogate the properties of the GA method. The fitness values for the entire set of three-layer symmetric host molecules were computed, and the score distribution for the host (Fhost) is shown in Figure 8. Host structures with fitness score higher than 700 (red points) are located at an energy range from −5.10 to −4.25 eV for the HOMO and an energy range from −2.10 to −1.35 eV for the LUMO orbital energy. This chemical set has about 1200 chemical structures and therefore is too large to use in combined optimization along with the ETL/HTL layers. Therefore, the top 1% of host structures were selected, representing about ∼100 unique chemical structures. The top 10 host structures are shown in Figure 9 from this symmetrical set. The top 1% of host structures (shown within R1 in Figure 8) correspond to an energy range of approximately 0.1 eV around the reference electrode potentials. Before considering the full OLED design, optimization of just the ETL was used to fine-tune GA parameters, including rates of mutation, crossover, and selection methods. In these runs, molecular structures that were evaluated in any prior generation are forced to mutate until a new structure is found. While this behavior is nonstandard for GAs, the rate-limiting step of the GA is evaluation of the FMO energy levels, and repeated computation for identical molecules wastes computational resources. Structures, FMO levels, and fitness scores are 2446


Article


Figure 12. Matrices showing the numbers of molecules for the HTL, host, and ETL with fitness scores > 800. Statistical results contain the population of each substitution fragment shown on the y-axis.

energies in the host are located near −4.75 and −1.72 eV, respectively. This shows that possible candidates for highperformance host materials are found from the large chemical structure library and, furthermore, that these reference HOMO/LUMO values can be used for further optimization of the ETL and HTL (by selecting these as the host properties in the FHTL and FETL fitness scores). Using the same GA strategy as that in host optimization, fitness scores converge very efficiently for the asymmetric molecular library to over 950 for ETL and over 900 for HTL within only 5 GA generations. This rapid convergence is due to the use of the optimized GA parameters (vide supra) based on the symmetric library test set (see Supporting Information Figure S5). On the basis of these results, multilayer OLED devices are constructed by combining the host layer and transporting layers and ranking these by host fitness, with the fittest HTL and ETL shown for each selected host. To identify OLED materials with feasible synthesis, the synthetic accessibility score (SAscore) and synthetic complexity score (SCscore) are taken into consideration (see Supporting Information Table S2). On the basis of the results, molecules with high fitness, low SCscore, and high SAscore can be chosen to be optimal. In Figure 11, the top three OLED multilayer candidates from the top 100 candidates are selected due to their SAscore being higher than the top 100 candidates’ average SAscore (−0.070) and the SCscore is lower than that of the top 100 candidates’ average SCscore (0.756). The results indicate that these top three OLED multilayer candidates are not synthetically complicated and will be much easier to synthesize than other high-performance candidates. According to the results, neutral groups serve as the best candidate fragments for high-performance ETL materials. Additionally, acridine-based and triphenylamine fragments are good candidates as HTL materials due to their strong electron-donating character. The best ETL materials have relatively higher fitness scores than the top HTL materials. This observation is likely due to the lower availability of good HTL molecules in the designated

chemical space: in the symmetric library, 815 of 9450 candidates have FETL > 950, while only 20 HTL candidates have FHTL > 950. Because the asymmetric library is built using the same fragments as the symmetric library, it is highly likely that, in comparison to the ETL material, relatively few of the 30 million compounds are optimal for the HTL. Nonetheless, the top scores for asymmetric HTL materials are still greater than 970, indicating near-perfect GA optimization. In addition to providing predictions for high-performing materials’ combinations, the collection of all individual data points for each layer can give insight into the OLED materials’ landscape. To do so, the effect of specific molecular fragments (i.e., seed, fragments, and caps) on the fitness score is studied by counting the number of fragments appearing in a high fitness score region. Figure 12 shows the population of fragments with fitness scores above 800. The top of Figure 12 indicates that pyridine is suitable as a seed to build up the three-layer structure for HTL/host/ETL as it shows high performance as an OLED material. For the first and second molecular segments, triphenylamine and acridine-based derivatives are well-known as good electron donors51−53 to better raise HOMO energy levels to generate high-performance HTL materials, as shown in the central left of Figure 12b. The neutral fragments (e.g., xylene- and naphthalene-based fragments) were shown to not only decrease the HOMO energy level but also to increase the LUMO and, therefore, have a good energy alignment of the frontier orbital energy level between the ETL and host material. This good energy alignment leads these fragments to attain a high fitness score. For the caps (3rd layer) shown on the right of Figure 12, we found that strong electron-donor fragments (e.g., triphenylamine and carbazole-based) can slightly enhance the fitness score for the HTL, but there is not a strong correlation to the caps on the fitness score for the ETL. 2447


Article


■

Emission from Organic Electroluminescent Devices. Nature 1998, 395, 151−154. (6) Adachi, C.; Baldo, M. A.; Thompson, M. E.; Forrest, S. R. Nearly 100% Internal Phosphorescence Efficiency in An Organic Light Emitting Device. J. Appl. Phys. 2001, 90, 5048−5051. (7) Yu, T.; Liu, L.; Xie, Z.; Ma, Y. Progress in Small-Molecule Luminescent Materials for Organic Light-Emitting Diodes. Sci. China: Chem. 2015, 58, 907−915. (8) Huang, C. W.; Peng, K. Y.; Liu, C. Y.; Jen, T. H.; Yang, N. J.; Chen, S. A. Creating a Molecular-scale Graded Electronic Profile in a Single Polymer to Facilitate Hole Injection for Efficient Blue Electroluminescence. Adv. Mater. 2008, 20, 3709−3716. (9) Zhang, Y.; Lai, S. L.; Tong, Q. S.; Lo, M. F.; Ng, T. W.; Chan, M. Y.; Wen, Z. C.; He, J.; Jeff, K. S.; Tang, X. L.; Liu, W. M.; Ko, C. C.; Wang, P. F.; Lee, C. S. High Efficiency Nondoped Deep-Blue Organic Light Emitting Devices Based on Imidazole-π-triphenylamine Derivatives. Chem. Mater. 2012, 24, 61−70. (10) Endo, A.; Sato, K.; Yoshimura, K.; Kai, T.; Kawada, A.; Miyazaki, H.; Adachi, C. Efficient Up-Conversion of Triplet Excitons into a Singlet State and Its Application for Organic Light Emitting Diodes. Appl. Phys. Lett. 2011, 98, 083302. (11) Goushi, K.; Yoshida, K.; Sato, K.; Adachi, C. Organic LightEmitting Diodes Employing Efficient Reverse Intersystem Crossing for Triplet-To-Singlet State Conversion. Nat. Photonics 2012, 6, 253− 258. (12) Lee, G. H.; Kwon, D. Y.; Kim, Y. S. Study of Cyanobenzene Derivatives for Thermally Activated Delayed Fluorescence Emitters. J. Nanosci. Nanotechnol. 2016, 16, 11453−11453. (13) Youn Lee, S.; Yasuda, T.; Nomura, H.; Adachi, C. High Efficiency Organic Light-Emitting Diodes Utilizing Thermally Activated Delayed Fluorescence from Triazine-Based Donor−Acceptor Hybrid Molecules. Appl. Phys. Lett. 2012, 101 (9), 093306. (14) Lin, T. A.; Chatterjee, T.; Tsai, W. L.; Lee, W. K.; Wu, M. J.; Jiao, M.; Pan, K. C.; Yi, C. L.; Chung, C. L.; Wong, K. T.; Wu, C. C. Sky-Blue Organic Light Emitting Diode with 37% External Quantum Efficiency Using Thermally Activated Delayed Fluorescence from Spiroacridine-Triazine Hybrid. Adv. Mater. 2016, 28, 6976−6983. (15) Lee, J.; Shizu, K.; Tanaka, H.; Nomura, H.; Yasuda, T.; Adachi, C. Oxadiazole- and triazole-based highly-efficient thermally activated delayed fluorescence emitters for organic light-emitting diodes. J. Mater. Chem. C 2013, 1 (30), 4599−4604. (16) Oehzelt, M.; Akaike, K.; Koch, N.; Heimel, G. Energy-Level Alignment at Organic Heterointerfaces. Sci. Adv. 2015, 1 (10), e1501127. (17) Lee, H.; Cho, S. W.; Yi, Y. Interfacial Electronic Structure for High Performance Organic Devices. Curr. Appl. Phys. 2016, 16, 1533−1549. (18) Ihn, S. G.; Lee, N.; Jeon, S. O.; Sim, M.; Kang, H.; Jung, Y.; Huh, D. H.; Son, Y. M.; Lee, S. Y.; Numata, M.; Miyazaki, H.; Gómez-Bombarelli, R.; Aguilera-Iparraguirre, J.; Hirzel, T.; AspuruGuzik, A.; Kim, S.; Lee, S. An Alternative Host Material for LongLifespan Blue Organic Light-Emitting Diodes Using Thermally Activated Delayed Fluorescence. Adv. Sci. 2017, 4, 1600502. (19) Chen, W.; Yuan, Y.; Ni, S.; Zhu, Z.; Zhang, J.; Jiang, Z.; Liao, L.; Wong, F.; Lee, C. Highly Efficient Deep-Blue Electroluminescence from a Charge Transfer Emitter with Stable Donor Skeleton. ACS Appl. Mater. Interfaces 2017, 9, 7331−7338. (20) Pilania, G.; Wang, C.; Jiang, X.; Rajasekaran, S.; Ramprasad, R. Accelerating Materials Property Predictions using Machine Learning. Sci. Rep. 2013, 3, 2810. (21) Pereira, F.; Xiao, K. X.; Latino, D. A. R. S.; Wu, C. C.; Zhang, Q. Y.; Aires-de-Sousa, J. Machine Learning Methods To Predict Density Functional Theory B3LYP Energies Of HOMO And LUMO Orbitals. J. Chem. Inf. Model. 2017, 57, 11−21. (22) Supady, A.; Blum, V.; Baldauf, C. First-Principles Molecular Structure Search with a Genetic Algorithm. J. Chem. Inf. Model. 2015, 55, 2338−2348. (23) Vargas, J. A.; Buendía, F.; Beltrán, M. R. New AuN (N = 27− 30) Lowest Energy Clusters Obtained by Means of an Improved

CONCLUSIONS GA optimization is herein shown to be useful for suggesting combinations of molecular materials for high-performance multilayer OLED device designs. Using a large library of over 30 million organic species constructed from 32 molecular fragments, this strategy matches electron/hole-transport energy levels among the device layers to ensure compatibility between the device layers. Ultimately, the method was able to optimize these alignments to near perfectionat least to within the limits of the postulated fitness functions and accuracy of the quantum chemical computations. Achieving this success required systematic evaluation of the performance of the GA parameters, including the rates of mutation/crossover and the selection algorithm to best navigate the large chemical space. Furthermore, this search through OLED materials’ space shows that the pyridine-based central core with acridine-based fragments are probable molecules for host molecules that have excellent energy alignment with common electrode materials. Additionally, weak electron-donating groups, such as naphthalene- and xylene-based fragments, appear often in the optimal ETL materials, and triphenylamine- and acridine-based fragments, due to their strong electron-donating character, are good suggested candidates for the HTL layer.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jcim.8b00044. Selection method overview; energy level rescaling; experimental benchmark test set; GA evolution for the ETL and HTL; and fitness scores for the top 100 candidates for the ETL, HTL and host (PDF)

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. ORCID

Shao-Yu Lu: 0000-0001-7304-2322 Paul M. Zimmerman: 0000-0002-7444-1314 Notes

The authors declare no competing financial interest.

■ ■

ACKNOWLEDGMENTS S.-Y.L. and P.M.Z. thank the Dow Chemical Company for support of this project. REFERENCES

(1) Tang, C. W.; VanSlyke, S. A. Organic Electroluminescent Diodes. Appl. Phys. Lett. 1987, 51, 913−915. (2) Schwartz, G.; Reineke, S.; Rosenow, T. C.; Walzer, K.; Leo, K. Triplet Harvesting in Hybrid White Organic Light-Emitting Diodes. Adv. Funct. Mater. 2009, 19, 1319−1333. (3) Kim, S.; Kwon, H.-J.; Lee, S.; Shim, H.; Chun, Y.; Choi, W.; Kwack, J.; Han, D.; Song, M. S.; Kim, S.; Mohammadi, S.; Kee, I. S.; Lee, S. Y. Low-Power Flexible Organic Light-Emitting Diode Display Device. Adv. Mater. 2011, 23, 3511−3516. (4) Fan, Y.; Zhang, H. M.; Chen, J. S.; Ma, D. G. Three-Peak TopEmitting White Organic Emitting Diodes with Wide Color Gamut for Display Application. Org. Electron. 2013, 14, 1898−1902. (5) Baldo, M. A.; O’Brien, D. F.; You, Y.; Shoustikov, A.; Sibley, S.; Thompson, M. E.; Forrest, S. R. Highly Efficient Phosphorescent 2448


Article

Journal of Chemical Information and Modeling DFT−Genetic Algorithm Methodology. J. Phys. Chem. C 2017, 121 (20), 10982−10991. (24) Venkatraman, V.; Foscato, M.; Jensen, V. R.; Alsberg, B. K. Evolutionary De Novo Design of Phenothiazine Derivatives for DyeSensitized Solar Cells. J. Mater. Chem. A 2015, 3, 9851−9860. (25) Lysgaard, S.; Mýrdal, J. S. G.; Hansen, H. A.; Vegge, T. A DFTBased Genetic Algorithm Search for AuCu Nanoalloy Electrocatalysts for CO2 Reduction. Phys. Chem. Chem. Phys. 2015, 17, 28270−28276. (26) Chu, Y.; Heyndrickx, W.; Occhipinti, G.; Jensen, V. R.; Alsberg, B. K. An Evolutionary Algorithm for De Novo Optimization of Functional Transition Metal Compounds. J. Am. Chem. Soc. 2012, 134, 8885−8895. (27) Abburu, S.; Venkatraman, V.; Alsberg, B. K. TD-DFT Based Fine-Tuning of Molecular Excitation Energies Using Evolutionary Algorithms. RSC Adv. 2016, 6 (2016), 3661. (28) Gomez-Bombarelli, R.; Aguilera-Iparraguirre, J.; Hirzel, T. D.; Duvenaud, D.; Maclaurin, D.; Blood-Forsythe, M. A.; Chae, H. S.; Einzinger, M.; Ha, D. G.; Wu, T.; Markopoulos, G.; Jeon, S.; Kang, H.; Miyazaki, H.; Numata, M.; Kim, S.; Huang, W.; Hong, S. I.; Baldo, M.; Adams, R. P.; Aspuru-Guzik, A. Design of Efficient Molecular Organic Light Emitting Diodes by a High-Throughput Virtual Screening and Experimental Approach. Nat. Mater. 2016, 15, 1120−1127. (29) Shu, Y. N.; Levine, B. G. Reducing the Propensity for Unphysical Wavefunction Symmetry Breaking in Multireference Calculations of The Excited States of Semiconductor Clusters. J. Chem. Phys. 2015, 142, 104104. (30) Kuttipillai, P. S.; Mukhopadhyay, S.; Sokolov, A.; Kearns, K. L.; Woodward, W.J. W.; Lunt, R. R. Mapping Recombination Profiles in Single-, Dual-, and Mixed-Host Phosphorescent Organic Light Emitting Diodes. Org. Electron. 2018, 57, 28−33. (31) Sasabe, H.; Kido, J. Development of High Performance OLEDs for General Lighting. J. Mater. Chem. C 2013, 1, 1699. (32) Wang, Y. J.; Lu, J. G.; Shieh, H. P. D. Efficiency Enhancement of Organic Light-Emitting Diodes on Flexible Substrate with Patterned Inverted Conical Structure. IEEE Photonics J. 2016, 8, 1. (33) Duan, L.; Xie, K.; Qiu, Y. Review Paper: Progress on Efficient Cathodes for Organic Light-Emitting Diodes. J. Soc. Inf. Disp. 2011, 19, 453−461. (34) Chen, C. T.; Chao, W. S.; Liu, H. W.; Wei, Y.; Jou, J. H.; Kumar, S. Spirally Configured Cis-Stilbene/Fluorene Hybrids as Ambipolar, Fluorescent Materials for Organic Light Emitting Diode Applications. RSC Adv. 2013, 3, 9381−9390. (35) Hung, W.-Y.; Chiang, P.-Y.; Lin, S.-W.; Tang, W.-C.; Chen, Y.T.; Liu, S.-H.; Chou, P.-T.; Hung, Y.-T.; Wong, K.-T. Balance the Carrier Mobility to Achieve High Performance Exciplex OLED Using a Triazine-Based Acceptor. ACS Appl. Mater. Interfaces 2016, 8, 4811−4818. (36) Zhang, D.; Cai, M.; Bin, Z.; Zhang, Y.; Zhang, D.; Duan, L. Highly Efficient Blue Thermally Activated Delayed Fluorescent Oleds with Record-Low Driving Voltages Utilizing High Triplet Energy Hosts with Small Singlet−Triplet Splittings. Chem. Sci. 2016, 7, 3355−3363. (37) Ertl, P.; Schuffenhauer, A. Estimation of Synthetic Accessibility Score of Drug-Like Molecules Based on Molecular Complexity and Fragment Contributions. J. Cheminf. 2009, 1, 8. (38) O’Boyle, N. M.; Banck, M.; James, C. A.; Morley, C.; Vandermeersch, T.; Hutchison, G. R. Open Babel: An Open Chemical Toolbox. J. Cheminf. 2011, 3, 33. (39) Goldberg, D. E. Genetic Algorithms in Search, Optimization, and Machine Learning; Addison-Wesley Publishing Company: Reading, MA, 1989. (40) Miller, B. L.; Goldberg, D. E. Genetic Algorithms, Selection Schemes, and the Varying Effects of Noise. Evol. Comput. 1996, 4, 113−131. (41) Yang, X. S. Engineering Optimization: An Introduction with Metaheuristic Metaheuristic Applications; John Wiley & Sons, 2010.

(42) Brain, Z. E.; Addicoat, M. A. Optimization of a Genetic Algorithm for Searching Molecular Conformer Space. J. Chem. Phys. 2011, 135, 174106. (43) Johnston, R. L. Evolving Better Nanoparticles: Genetic Algorithms for Optimising Cluster Geometries. Dalton Trans. 2003, 22, 4193−4207. (44) Bell, B. M.; Clark, T. P.; De Vries, T. S.; Lai, Y.; Laitar, D. S.; Gallagher, T. J.; Jeon, J.-H.; Kearns, K. L.; McIntire, T.; Mukhopadhyay, S.; Na, H.-Y.; Paine, T. D.; Rachford, A. A. BoronBased Tadf Emitters with Improved OLED Device Efficiency Roll-Off and Long Lifetime. Dyes Pigm. 2017, 141, 83−92. (45) Stewart, J. J. P. Optimization of parameters for semiempirical methods I. Method. J. Comput. Chem. 1989, 10 (2), 209−220. (46) Kulkarni, A. P.; Tonzola, C. J.; Babel, A.; Jenekhe, S. A. Electron Transport Materials for Organic Light-Emitting Diodes. Chem. Mater. 2004, 16, 4556−4573. (47) Ho, M. D.; Kim, D.; Kim, N.; Cho, S. M.; Chae, H. Polymer and Small Molecule Mixture for Organic Hole Transport Layers in Quantum Dot Light-Emitting Diodes. ACS Appl. Mater. Interfaces 2013, 5, 12369−12374. (48) Brunner, K.; van Dijken, A.; Börner, H.; Bastiaansen, J. J. A. M.; Kiggen, N. M. M.; Langeveld, B. M. W. Carbazole Compounds as Host Materials for Triplet Emitters in Organic Light-Emitting Diodes: Tuning the HOMO Level without Influencing the Triplet Energy in Small Molecules. J. Am. Chem. Soc. 2004, 126, 6035−6042. (49) Chen, D.; Su, S.-J.; Cao, Y. Nitrogen Heterocycle-Containing Materials for Highly Efficient Phosphorescent OLEDs with Low Operating Voltage. J. Mater. Chem. C 2014, 2, 9565. (50) Sun, F.; Jin, R. DFT and TD-DFT Study on the Optical and Electronic Properties of Derivatives of 1,4-bis(2-Substituted-1,3,4Oxadiazole)benzene. Arabian J. Chem. 2017, 10, S2988. (51) Lee, H. K.; Ahn, H.; Lee, C. H. Device Characteristics of Blue Phosphorescent Organic Light-Emitting Diodes Depending on The Electron Transport Materials. J. Inf. Disp. 2011, 12, 219−222. (52) Cias, P.; Slugovc, C.; Gescheidt, G. Hole Transport in Triphenylamine Based OLED Devices: from Theoretical Modeling to Properties Prediction. J. Phys. Chem. A 2011, 115, 14519−14525. (53) Yang, C.-H.; Liao, S.-H.; Sun, Y.-K.; Chuang, Y.-Y.; Wang, T.L.; Shieh, Y.-T.; Lin, W.-C. Optimization of Multiple Electron Donor and Acceptor In Carbazole-Triphenylamine-Based Molecules for Application of Dye-Sensitized Solar Cells. J. Phys. Chem. C 2010, 114, 21786.

2449


Virtual Screening of Hole Transport, Electron Transport, and Host

Recommend Documents