Final Report Summary  IM3OLED (Integrated Multidisciplinary & Multiscale Modeling for Organic LightEmitting Diodes)
EU FP7 project IM3OLED is a softwarefocused research project on multiscale, multidisciplinary and multiphysics modelling of organic lightemitting diodes. IM3OLED is a collaborative project with experts from the European Union and the Russian Federation. Funded through the European Union’s Seventh Framework Programme (FP7NMP2011EURUSSIA), the EU portion of the IM³OLED project officially launched in October 2011 and ran for 30 months. The project website can be found at: http://www.im3oled.eu/
Because of their versatility, low energy consumption, lack of hazardous metals and potentially low cost, organic lightemitting diodes (OLEDs) are attractive for application in displays and general lighting. OLEDs are ideally suited to replace existing lighting technologies. in contrast to inorganic LEDs, OLEDs by design are flat and diffuse luminaires. Their unique planar and ultrathin design make it possible for architects and designers to integrate these remarkable lighting devices into structures of any shape and size. For OLEDs to be competitive with existing lighting technologies, however, high efficiencies and lifetimes are crucial. However, improving this requires an integral approach of all variables/processes involved, as these are all interrelated and influencing each other.. Achieving improvements on all involved topics, from chemistry to device physics, by carrying out experiments is possible, but expensive and timeconsuming and requires expertise in all disciplines. A predictive simulation tool would significantly simplify this task.
The overall objective of the IM³OLED research project was therefore to find out how OLEDs can become more efficient, by further improving and developing existing and novel models. The project included molecular calculations, electrical and optical simulation, 3D OLED optics and scaling / integration effects.
To achieve our objectives, the IM³OLED project combined partners from Europe and the Russian Federation, including OLED manufacturer Philips and research institute Holst Centre / TNO, modelling partners Fluxim, JCMwave, Kintech Laboratory and leading university groups on the topic of computational physics and multiscale modelling at the atomistic and molecular length scales (ZHAW, PCCRAS and MEPhI):
Project Context and Objectives:
The performance of an OLED depends on the properties that are characterized by different length scales: from the macro scale of the design of the OLED module down to the nanometre scale of the organic materials that make up the functional organic layers of the OLED. The main challenge in developing a predictive modelling tool is the coupling of the properties at the different length scales.
Multidisciplinary modelling is crucial to access a sufficiently wide parameter space to properly address a multitude of issues that are in fact correlated. The OLED efficiency, for instance, is directly related to the effectiveness of the separate layers in the OLED structure to fulfil their role properly, defined by their molecular design and its environmental conditions. The complexity of the OLED and the many parameters that need to be optimized simultaneously requires predictive modelling on various length scales and includes:
• Characteristics of molecules and layers
• Charge transport and injection through layers and past interfaces
• Exciton formation, diffusion and radiative decay (leading to emission)
• Thin film optics for propagation through thin film stacks
• Geometric optics when it concerns large length scales
• Lateral current distribution in large area OLEDs
• Resistive heating in large area OLEDs
The overall objective of the IM³OLED research project was therefore to find out how OLEDs can become more efficient, by further improving and developing existing and novel models. The project included molecular calculations, electrical and optical simulation, 3D OLED optics and scaling / integration effects. At the end of the project, three tools were realized or extended by SME’s Fluxim (Ch), JCMWave (De) and Kintech (Ru), thereby providing useful software on the topic to the OLED industry.
At the time of writing, multiscale modelling of OLEDs is split into sections, for which the connections are carefully explored by experts before involving another complexity. Stacking of individual molecules and the impact of stacking on the charge transport characteristics can be described by combining molecular dynamics and 3D Monte Carlo techniques. The resulting excited states and energy levels, formed by recombination of charge carriers, can be visualized and assessed by computational quantum mechanical modelling. Such ab initio modelling is more commonly used to understand and predict charge transport in OLEDs, albeit mostly for individual layers. The reliability of predictive modelling of materials and layers and subsequent charge transport of OLEDs relies heavily on the accuracy of the modelled parameters. OLEDs based on small molecules may contain a few layers, but can contain dozens of layers. If the modelled parameters differ for each layer to some degree, even within reasonable limits, a predictive model of the entire stack may result in unrealistic behaviour.
On the complete other side of the spectrum, in the scale of centimeters to eventually meters, the actual fabricated large area OLEDs show a clear macroscopic behaviour when in operation. Understanding the impact of the current distribution and heat dissipation is clearly valuable for the design and production of OLEDs as these affect homogeneity of emission, overall device efficiency and operational lifetime. As a separate simulation, FEM modelling allows optimization of the current distribution and heat dissipation. However, to fully make use of its benefits, it is ideal to link modelling of scaling effects to calculations of the charge transport. Within most large area simulations, the OLED stack is considered a single layered entity. It’s light and current output is implemented through the measured currentvoltageluminance (IVL) characteristics. A distinction between the several components and their micro and macroscopic behavior is not deemed relevant.
In the middle of the length scales, modelling of the optics of OLEDs faces the challenge of linking the microscale to the macroscale. Light emission occurs from a multitude of nanometer sized dipoles in a stack of a few hundred nanometers. Light interacts over various length scales, which therefore have to be taken into account. Surface roughness, scatter particles, photonic crystals and other multiscale optical features may be required to create efficient OLEDs. Efficient methods to solve the computational challenges and enable linking the various length scales are clearly a necessity. The members of the IM3OLED consortium have explored these challenges and found opportunities to implement these into commercial software, thereby providing access to our findings to the OLED community.
Uniqueness of IM3OLED
Several FP7 projects are in the stage of execution or have been completed on the topic of organic devices, multiscale modelling or the combination of the two. Multiscale modelling projects MINOTOR and MONAMI involve simulation of properties of materials for application in organic electronics, mostly OPV and OLED. MMM@HPC has a wider aim in that it also targets establishing an infrastructure by which access from industry, institutes and academia is enabled. TRANSSIM aims at developing innovative multiscale computational tools for tackling charge transport in organics materials (organic crystals, organic thin films, etc.) based on firstprinciple calculations. AEVIOM was another strong modelling project that completed a few years ago. It included electrical modelling of OLEDs and advanced charge driftdiffusion models to include energetic disorder. By evaluating material parameters and their impact on the electrical properties of OLEDs, AEVIOM made important improvements to available OLED device modelling tools commercialized by Fluxim (Ch) and Sim4Tec (De). In the Russian Federation, the SME Kintech specialized in multiscale modelling and has intimate ties to the Russian Academy of Sciences. In 2013, their collaborative project with the EU partners of IM3OLED completed the development of a multiscale tool for predictive modelling of OLEDs. Their tool and the improvements made to software tool SETFOS, by EU partner Fluxim and multiscale optics tool JCMsuite by EU partner JCMwave , will be described further on in this report.
The EU part of the collaborative project (IM3OLED) had a strong focus on the electrical and optical modelling of small and large area OLEDs (~ mm2 to > 100 cm2), wherein the presence of multiple multiscale features modified the propagation of light through the device. The presence of such features was explored experimentally by Philips and TNO and theoretically by Philips, JCMwave. ZHAW and Fluxim. The advances in fullwave modelling of such features was implemented in commercial software tools Setfos and JCMsuite. The novel modelling approaches implemented in JCMsuite that enable multiscale optics in OLEDs were disseminated by JCMwave and Philips in several proceedings (2013,2014) and presentations at SPIE. The direct implementation of the results obtained within the project into commercial tools, and the relevance of achieving the project’s objectives on today’s topics in optimizing OLED, directly impacts the research and development of OLEDs for lighting application.
Defined objectives:
In the DOW of the IM3OLED project (grant number 295368, dated 20110929), the following S/T objectives were defined:
• The overall goal of IM3OLED is the development, evaluation and validation of a predictive multiscale and multidisciplinary modelling tool that will accelerate research and development of organic lightemitting diodes for lighting applications.
• Implementation of validated theoretical models into software tools including the commercially available Europeanmade Setfos™ simulation package for organic semiconductor devices and a multiscale modelling tool developed by the Russian software developer Kintech Laboratories.
Multiscale bridge between externally observed macroscopic OLED efficiency and nanoscale molecular behaviour:
• Abinitio modelling of material properties of small molecule systems by means of quantum chemistry, molecule dynamics, extending into the mesoscale by Monte Carlo and validation thereof
• Abinitio modelling of charge transport and exciton physics and improved predictability of 3D Monte Carlo modelling
• Abinitio modelling of light transmission from within a multilayer stack across planar interfaces, corrugations and taking into account the possible presence of a multiplicity of multiscale heterogeneities and optical effects by birefringence and (in)coherence of light.
• Development of novel fullwave modelling and validation with various established models that address optical behaviour of OLEDs.
• Application of existing optical models by theoretical screening combinations of light extraction techniques that enable beyond stateoftheart light extraction efficiencies.
• Modelling of OLED size scaling, the impact on the lateral current distribution and associated heat dissipation, which can be considered to be input for modelling of material properties, charge transport and exciton physics, etc.
Our final objective concerns experimental validation of models and implementation thereof. A program that enabled multiscale modelling of OLEDs should allow the calculation of the following properties:
• Energy levels of molecules and their excited states, meaning HOMO, LUMO and optical band gap
• Packing density of a layer and molecular interactions between guest and host molecules
• Charge transport, recombination and exciton diffusion
• Thin film optics with emission from within the thin film stack near a metallic surface
• Propagation of light while interacting with interfaces, particles, (periodic) structures and materials of various natures
• Distribution of current over lateral distances that can be considered “large area” for OLEDs, which is above a size of ~1 cm2
• Generation and distribution of heat over large area OLEDs
• The integrated properties of the devices when including all these aspects using a dynamic feedback
Two partners, TNO/Holst Centre and Philips, two of the European technological leaders in OLED
development, routinely make high performance OLEDs and will utilize their knowledge in fabricating appropriate OLED device architectures for verification of the modelling results.
• Theoretical and experimental validation of models and modules using existing and new data obtained on relevant OLED model devices.
Project Results:
Main S&T results/foregrounds:
Multidisciplinary modelling is crucial to access a sufficiently wide parameter space to properly address a multitude of issues that are in fact correlated. The OLED efficiency, for instance, is directly related to the effectiveness of the separate layers in the OLED structure to fulfil their role properly, defined by their molecular design and its environmental conditions. The complexity of the OLED and the many parameters that need to be optimized simultaneously requires predictive modelling on various length scales.
Characteristics of materials and layers in OLEDs [topic of WP2]
In Im3OLED the materials were investigated using computational methods by Photochemistry Center of the Russian Academy of Sciences. Experimental data was provided by TNO and Philips.
Among the materials that were investigated are:
1. TPBi = 1,3,5Tris(1phenyl1Hbenzimidazol2yl)benzene
2. BAlq = Bis(2methyl8quinolinolatoN1,O8)(1,1'Biphenyl4olato) aluminum
3. αNPD = N,N'di(naphthalen2yl)N,N'diphenylbenzidine
4. Ir(MDQ)2acac = iridium(III)bis(2methyldibenzo[f,h]quinoxaline)(acetylacetonate)
5. mMTDATA = 4,4',4''Tris(N3methylphenylNphenylamino)triphenylamine
The structure of a nanomaterial exhibits several levels of organization, namely atomic/molecular (1–2 nm), supramolecular (2–10 nm), nanosized (10–100 nm) and microscopic (100–1000 and more nm). Methods to calculate these properties are
1) quantum chemistry for molecular structure and properties (atomic/molecular and supramolecular levels);
2) molecular dynamics and Monte Carlo for microstructure (nanosized level)
3) band structure (planewave) slab calculations for interface structures and properties (nanosized level)
The packing densities of these materials were modelled with Monte Carlo techniques using two approaches, one consisting of increasing the density of molecules in a certain volume at elevated temperature (coinciding with the gas phase), followed by decreasing the temperature to room temperature. The other method consisted of a simulated sputtering on a substrate. All investigated materials were found to have a slightly lower density than was determined experimentally, but the trends were found to be consistent with the experimental values.
Also the highest occupied molecular orbitals (HOMO) and lowest unoccupied molecular orbitals (LUMO) energy levels were calculated and compared to experimental results. For TPBI, MD simulations were performed at 900 K for 500 ps using 144 molecules that were placed randomly in a 6х6х6 nm cell. The cell side was reduced steadily until the density of the material was identical to the experimental result. At the last stage of reducing temperature (298 K), 10 independent calculations were performed according to this scheme which resulted in HOMO/LUMO energy levels of 6.2 eV / 1.9 eV. The HOMO level is in excellent agreement to the experimental result of 6.3 eV, but we noticed that the LUMO energy level deviated by a few tens of an eV. The energy levels of the emissive layer αNPD:Ir(MDQ)2acac were also found to deviate from the experimental values. First it was determined that each αNPD molecule has approximately 21 nearest neighbours within 5 Å and Ir(MDQ)2acac approximately 15 neighbours within 5 Å. Using the calculated microstructure and the polarizable continuum model (PCM) with εinf of 3, the HOMO levels of αNPD and Ir(MDQ)2acac were 5.2 eV (exp: 5.5 eV) and 5.5 eV (exp: 5.2 eV). The optical band gaps were found to be 3.2 eV and 2.6 eV (exp: 3.1 eV and 2.2 eV). The calculated emission spectrum and radiant rate constants were close to experimental data, but not exact (calculated emission maximum of Ir(MDQ)2acac at 17513 cm1, exp: 16750 cm1; radiative decay rate 1.9 s, exp: 4.2s).
While these calculations, for a few of the materials that were explored, provide good insight into the properties of molecules and layers, we noticed that it can be challenging to find perfect agreement with experimentally obtained values.
Charge transport and exciton dynamics [topic of WP3]
In the EU project, the following tasks were defined:
Task 3.1: Development of advanced charge transport models (Fluxim, ZHAW)
Analytical models for charge and exciton transport suitable for implementation in 1D solvers are being developed.
Particular emphasis will is put on:
• Charge doping in organic semiconductors.
• Trap state modelling: flexibility, allowing for different trap distributions and trapping behaviour of dye dopants.
• Multiple trapping and release model for dynamic description of dispersive transport.
• Dynamics of excitons including excitonexciton and excitoncharge interactions
• Exciton transport across layer interfaces.
Task 3.2: Development of fast numerical methods for device characterization.
Task 3.3: Parameter extraction
Task 3.4: Adopting 3D advanced material, charge transport and exciton dynamics models for implementation in Setfos.
Task 3.1: Development of advanced charge transport models
Charge doping in organic semiconductors
A model was implemented in Setfos to include doping concentrations in organic OLED layers. Due to tailored photoelectric properties organic semiconducting materials have received much attention in recent years. Doping of organic semiconductors, by adding electron or holedonors in the organic material, offers a particularly effective way of improving device properties and performance. Charge doping can be considered by including a background charge in the Poisson equation. This method is also applicable in the case of socalled deep traps. The trapped carriers act as a stationary background charge and their treatment is completely analogous to doping.
Trap state modelling
Trap state modelling: flexibility, allowing for different trap distributions and trapping behaviour of dye dopants. Different densities of states of traps are frequently used when simulating organic materials. The simplest choice is a discrete trap level that is equivalent with a  shaped density of states. Other options are an exponential or a Gaussian shaped density of states. The effect of trap DOS on the trapped density of states is illustrated in Figure 1a. The device used in the simulation (Figure 1b) is a 140 nm thick, electrononly device where the electron mobility is modelled by the EGDM model. For the discrete level of trapping of electrons, a density of 1024 m3 is chosen and the trap depth in energy is 0.2 eV below the LUMO level. The parameters for the Gaussian distribution for electron trapping are 1024 m3 for the density, the width of the Gaussian shape is 0.13 eV and the energy offset is also 0.2 eV below the LUMO level as in the discrete case.
Figure 1: a) Influence of trap DOS shape on the trapped charge carrier density, b) demonstration of the effect on currentvoltage curves for different trap DOS (discrete level and Gaussian shaped) in Setfos
Multiple trapping and release
The EGDM or ECDM models are successfully applied to the simulation of steadystate currentvoltage curves. However, in timedependent experiments such as dark injection transient currents and impedance spectroscopy the relaxation of charge carriers plays an important role and is not considered in the EGDM or ECDM. We therefore resort to the multiple trapping and release model that accounts for the energetic relaxation. In the following we show the effect of slow and fast trapping on the frequencydependent capacitance in comparison with the trapfree case. In the case of slow trapping (small capture coefficient) a rise at low frequency is observed. In the case of fast trapping the maximum of the negative differential susceptance B is shifted to lower frequencies which results in a lower effective mobility due to the permanent capture and release process. We have demonstrated a simulation of a smallsignal perturbation experiment (impedance spectroscopy) by linearizing the model that accounts for trapping. Depending on the capture coefficient two different regimes with characteristic features are obtained.
Figure 2: a) Top: Frequencydependent capacitance for different trap types and a trapfree material. Bottom: Frequencydependent negative differential susceptance that allows the determination of the effective mobility value. With the multiple trapping and release model for slow trapping the low frequency rise as often observed in measurements can be reproduced.
Dynamics of excitons
In the driftdiffusion solver in Setfos, we have implemented the exciton dynamics by including a position dependent decay rate. The position dependent decay rate of excitons is of importance because the optical environment in which the excitons reside can enhance or suppress the radiative decay. The comprehensive optical model uses the information of the exciton lifetime to determine where the decay of the excitons increases and thus the density of the exciton decreases. This model is used in case of opticsonly simulations. We demonstrated the impact of exciton dynamics on the resulting exciton profile inside the OLED by the comprehensive model implemented in Setfos and have performed an experimental study in collaboration with the Holst Centre in order to extract the emission profile from measurements of OLED emission spectra. These results (summarized in D3.1) were published in a joint publication.
Task 3.2: Development of fast numerical methods for device characterization
Multiple trapping and release model for large signal perturbation
In recent years it has become clear that charge traps can be revealed by performing impedance spectroscopy which is a smallsignal perturbation technique. After gaining experience with slow and fast trap states in impedance modelling, we have recently extended our focus on large signal perturbation such as the dynamic current response to a step voltage. We study the large signal perturbation with an implementation in our academic code that uses an adaptive timestepping algorithm (implicit Euler and TRBDF2).
We employ the concept of slow and fast trap states and can find two specific regimes. To demonstrate the practical impact of these regimes, we apply our numerical model to the DITC (dark injection transient currents) that have commonly been used to determine the charge mobility in organic semiconductor devices. The obtained results from DITC studies strongly depend on the measurement conditions. As an application we analyze the measurements of Esward et al., which were reproduced with a numerical model in Setfos. The simulations are able to explain the experimental observations with the help of relaxation effects due to shallow traps. See D3.1/D3.2 for more information.
Implementation of fast multithreading algorithm in SETFOS
The development of fast numerical methods for electrical modeling of OLEDs is crucial for this project and the community as a whole. Apart from purely numerical tricks, a suitable approach to this challenge is advanced software engineering. Fluxim has developed a concept for exploiting multicore PC architectures by use of a multithreading approach. It is expected that this approach satisfies any user of SETFOS who, in fact, is typically performing calculations on a desktop PC. The main idea is to parallelize the sweep command which is present in most simulation tasks of SETFOS and allow the user to customize the settings. SETFOS can now exploit multicore PC architectures thanks to the implemented multithreading algorithm. Most SETFOS applications will benefit from this.
Implementation of small signal perturbation technique in SETFOS for admittance spectroscopy (Task 3.2)
The SETFOS Driftdiffusion module is one of three simulation software modules that Fluxim is licensing to its international users. It was originally introduced to the market in 2007 and was based on a transient algorithm for the integration of the underlying charge continuity and Poisson equations. Though that method provides the advantages of relatively easy implementation and timedependent simulation results, it is not computationally efficient and provides more results than usually needed for steadystate analysis of currentvoltage curves of OLEDs.
An advanced experimental characterization technique is the impedance spectroscopy, in which a sinusoidal voltage signal is applied to the organic device and the phaseshifted periodic current response measured. The impedance resulting from this alternating current (AC) technique is acquired for a range of oscillation frequencies covering several orders of magnitude on the time scale (thus the name impedance spectroscopy). The large time scale can reveal fast and slow charge transport processes such as charge trapping and release. In IM3OLED we are aiming for the implementation of algorithms for impedance spectroscopy in SETFOS. The first version of a robust algorithm for impedance spectroscopy has successfully been implemented in SETFOS. It has become publicly available as part of the first software release during the project.
Task 3.3: Parameter extraction methods
Correlations of fitting parameters were studied for currentvoltage curve analysis. Correlation analysis is available as a feature of Setfos. A modified fitting target was proposed for parameter extraction from currentvoltage curves with a reduced number of parameters allowing for a faster optimization process. Successful fits of IV curves over large dynamic range in current were demonstrated, see below.
Figure 3: fits of IV curves using correlation analysis
For transient current measurements (transient spacecharge limited currents, TSCLC), a suitable fitting method for driftdiffusion modeling had to be developed. Thus a demo tool for fitting TSCLC data was developed that iteratively adjusts the model parameters to achieve a good fit. The fitting feature of Setfos was also applied to the extraction of source spectra of the lightemitting material in the OLED.
Task 3.4: Adopting 3D advanced material, charge transport and exciton dynamics models
On the Russian side of the project, for charge transport modeling in 3D, mainly the Monte Carlo method was used and effective mobility models were studied. At ZHAW an alternative 3D charge hopping modeling method was implemented in a prototype software, namely the Master Equation approach, which combines the advantages of transport modeling in 3D and arbitrary DOS distributions with reasonable computation speed. The overview of methods is shown in the figure below. The prototype software was successfully implemented and used to study TSCLC data in comparison to driftdiffusion results and published (Szymanski et al, IEEE JSTQE, 2013). Moreover, current voltage curve results were obtained and compared to driftdiffusion results for the case of a bilayer with a discontinuity of the internal HOMO/LUMO energy levels. The software is implemented efficiently and runs on a desktop PC.
Figure 4: demonstration of the “Master equation approach” developed at ZHAW
The Master Equation solver of ZHAW was not only developed and applied for DC and transient simulations but also for AC simulations, providing an alternative for the impedance solver of the Driftdiffusion model (Setfos). The successful AC Master Equation simulation was reported in D3.4. The domain size in 3D and its implications on accuracy of the result was studied in the same report.
As the energetic DOS distribution of the semiconducting layer material is an input for these simulations, it can be taken from molecular scale simulations (performed on the Russian side).
OLED optics and light extraction [WP4]
In this work package, we work towards the following objectives:
• Develop a multiscale simulation tool for the prediction of light extraction efficiencies integrating the description of photon propagation on the nano (wave optics) and on the meso and macroscale (geometrical optics)
• Integration of wave and geometrical optics
According to the DOW we planned to reach these goals with performing the following tasks:
Task 4.1: Benchmarking optical simulations
Task 4.2: Extension of existing simulation tools
Task 4.3: Extension of FDTD and UWVFFEM
Task 4.4: Investigation of full wave approach to multiscale optics based on UWVFFEM
Task 4.5: Integration of wave and geometrical approach in one simulation method (Russian Scope)
To overcome the limitations in light extraction from OLEDs, thereby enhancing the efficiency, scattering layers have to be used in various guises (Figure 5). Propagation of light through an OLED multilayer requires the solution of Maxwell's equation which in this simple case can be accomplished by semianalytical means in an accurate and efficient way. The underlying algorithms are well known and documented in the literature. They have been implemented in softwares like SETFOS (Fluxim) or LIGHTEX (Philips) and allow for reliable predictions of extraction efficiencies and intensities as a function of emission angle.
In the presence of scattering layers however Maxwell's equations for a radiating dipole have to be solved by numerical means which complicates the task considerably, the main issue being the required size of the computational domain. Considering the emission by an individual dipole we have to deal with lateral extensions of several tens of microns and in some cases the vertical extension of the structures is also in the range of 510um. The application of standard methods for the numerical solution of Maxwell's equations like FDTD (Finite Difference Time Domain) and FEM (Finite Element Method) is severely limited by this fact. These require an artificial truncation of the computational domain and the light escaping through the vertical walls (which are ideally nonreflecting) of the computational domain is lost for the analysis. It was an important finding of this project that this truncation error cannot be neglected even for large domains. To overcome this limitation JCMwave developed the socalled BlochFloquet method which allows to calculate the radiation of a single dipole in a periodic structure in an effective way and to assess the "truncation" error induced by finite computational domains accurately.
Figure 5: three out of eight example light extraction layers investigated in WP4
Light propagation through "optically" thick layers can be described by raytracing (geometrical optics) using programs like ZEMAX and RAYPLAY(Philips proprietory). However it is not clear how thick is thick and whether raytracing can be safely applied when dealing with layers of intermediate thickness in the range from a few microns to say 20 microns or whether interference effects have to be taken into account. JCMwave was able to elucidate this question by developing so called "domain decomposition" methods which allow to treat the problem in the framework of "full wave" optics.
The results of this analysis also serve as a backdrop for the so called "hybrid" methods which employ raytracing in conjunction with BSDFs characterizing the scattering layers obtained on the basis of full wave optics.
In the end the practitioner needs to have a better understanding on how well the various methods allow to predict light extraction for a given investment in computational resources. From the work performed in the project we can draw the following conclusions:
1: the standard approach where the emission from the (planar) OLED stack is calculated semianalytically with programs like SETFOS or LIGHTEX and subsequent light propagation is modelled by raytracing is only applicable in certain cases like external scattering layers or internal extraction layers comprising many scattering particles.
2: hybrid methods which are based on raytracing and the description of scattering interfaces with wavelength feature sizes by BSDFs computed with full wave methods give a better approximation than pure raytracing
3: by applying the new "full wave" approaches for "large" problems developed by JCMwave in the project to some model problems further insight should be generated in which cases these "expensive" methods yield new insight and to the domain of applicability of the various approaches delineated.
In this sense the project has laid the foundations for important work in the future.
Task 4.1: Benchmarking optical simulations
In collaboration with partners Kintech and MEPhI a benchmark will be made of the available simulation tools for typical device structures that can be considered relevant to the project. Focus is on identifying their shortcomings and limitations like
• presence of incoherent layers between the organic stack and the substrate
• effective description of random or periodic scattering layers,
• randomly or periodically corrugated OLEDs (including the cathode)
• presence of metal structures
• large structures with an interplay of wave and geometrical optics
• treatment of structures on the mesoscale (measuring many wavelengths)
A detailed account of the benchmarking is given in deliverable D4.1. All items were covered and new insight provided. Here we give a short summary:
In benchmark 1 we considered a prototypical bottom emitting OLED without scattering layers whose emission can be modeled by the “semianalytical” programs LIGHTEX and SETFOS. These analytical solutions are compared to the results obtained with the numerical programs FDTD II (Kintech), FDTD Solutions and JCMwave. The FDTD programs matched the “exact” solutions only to 510% even for large computationa domains in the case of nearly lossless structures. Surprisingly the algorithm implemented by JCMwave matched very well with the analytical solutions.
In benchmark 2 a bidirectional OLED (BiOLED) emitting through the (semitransparent) cathode and the anode without scattering layer was simulated. The bottom and top emission can again easily be described with LIGHTEX/SETFOS. The bottom emission could also be simulated quite accurately with the numerical tools, for the top emission however appreciable differences between the analytical and FDTD solutions were observed. In particular it appeared to be impossible to describe the intensity emitted through the top cathode as a function of emission angle even in an approximate way. This appears to be a major shortcoming for FDTD based OLED simulations. Again JCMwave furnishes results in perfect agreement with the analytical ones due to a special numerical implementation.
The subject of benchmark 3 are structured anodes and cathodes. For the anodes regular arrangements of cylinders as shown in figure 4.3 were simulated (extraction efficiencies and emission patterns into the substrate). Here the agreement between the programs was reasonably good in the presence of absorption. For OLEDs with negligible absorption however the truncation of the computational domain by the FDTD programs introduced a noticeable error compared to the JCMsuite results which take into account the full periodicity of the problem without truncation.
In benchmark 4 the coherence properties of the light emitted by the OLED stack into a thick optical layer (050um) between the ITO anode and the substrate is investigated by “spectral averaging”. It is shown that for layer thicknesses above 10um the light emanating from the OLED stack can be considered as incoherent and that its propagation can therefore be safely described by raytracing.
Benchmark 5 compares the efficacy of internal and external scattering particle layers for a given OLED on the basis of raytracing with the softwares RAYPLAY (Philips) and ZEMAX (commercial). The predictions of the extraction efficiencies differ in some instances very considerably. It was discovered by comparison with the standard raytracing software ASAP that the description of scattering particle layers by ZEMAX is not correct.
Task 4.2: Extension of existing simulation tools
Extension of the functionality of the existing simulation tools at partner companies Fluxim and Philips (SETFOS, LIGHTEX) for planar microcavities:
• Inclusion of incoherent layers
• Inclusion of a scalar scattering approach
• Inclusion of layers with uniaxial birefringence (axes perpendicular and parallel to layers)
• Effective description of emission inside or close to absorbing media
Details are given in deliverable 4.2.
Figure 6: Illustration of numerical models for multiscale simulation of OLED lightoutcoupling.
A multiscale toolset, as envisioned in this project, should be capable of covering a range of length scales from (sub) wavelength scale to the macro scale. In Figure 6 the work beyond the stateoftheart is highlighted with red boxes and blue arrows. One approach in this project is the development of fullwave FEM simulation methods (see red boxes above) that can cope with as large as possible domains with reasonable computational resources.
Fluxim and ZHAW reported progress related to the development of a scalar light scattering model for OLEDs (red box below in Fig 4.2.1 route C) and a combination of a semianalytical DE model (SETFOS) with a raytracing tool (route A in Fig 4. 2.1). If the scattering behaviour of the layer (or of the interface) is well characterized, or if the multilayer stack does not contain any scattering layers and is only a sequence of thin coherent and thick incoherent layers, we employ an innovative approach based on nonparaxial scalar scattering to compute the light outcoupling from these complex multilayer stacks. The mathematical way to solve the problem does not rely on a MonteCarlo algorithm, which is the case in raytracing simulations. A prototype version of a numerically efficient and physically sound model of incoherence and isotropic light scattering was fully integrated into the simulation software SETFOS. This allows the user to run a multiscale simulation of an OLED starting from microscopic electrical modelling of the device to optical simulation of the light outcoupling efficiency, using the same software which simplifies the workflow.
Task 4.3: Extension of FDTD and UWVFFEM
JCMwave took over the part of ZHAW (“UWVFFEM”) in the original DOW and start to work on Discontinuous Galerkin methods in April/May 2012. In the time domain the socalled TDDG (time domain discontinuous Galerkin) methods have a number of well documented advantages as compared to FDTD or frequency domain methods (FEM). Since JCMsuite already implements a rigorous expansion of the computed fields into plane waves (Fourier transform) a coupling of the DDFEM with a rayoptical treatment of the upper and lower half spaces can be implemented in a straightforward manner. This will allow for a fullwave numerical analysis of a large multiscale device.
Integration of wave and geometrical approach in one simulation method with the following options:
• Hybrid approach: Effective description of nanostructured layers by BSDFs (Bidirectional scattering distribution functions) which can be computed offline by FDTD or FEM and used as building blocks in the raytracing part of the simulation
• Full wave approach based on FEM and/or FDTD (depending on the outcome of task 4)
• Integration of FDTD I and/or FEM building blocks in the “planar” microcavity model that is extended by scalar scattering in Task 2.
• Benchmarking of the various approaches integrating wave and geometrical optics
More details are given in deliverable 4.6
Further development and extension of FDTD and FEM approaches with regard to
• Improved implementation with regard to efficiency, accuracy and handling
• FEM (JCMwave): Generalize and develop domain decomposition (DD) approach for large scale problems. DDFEM will allows for a combination FEM for textured layers with plane wave expansion within plane layers.
• FEM (JCMwave): Implementation of Discontinuous Galerkin (DG) method in time domain and frequency domain.
• Development and implementation of the BlochFloquet method for periodic structures.
• Limitations identified in task 1
The BlochFloquet method, which allows to calculate the radiation of a single dipole in in an OLED with a periodic extraction layer to a high degree of accuracy, was implemented and tested by JCMwave. As an example Table 1 gives the extraction efficiencies into the substrate calculated for the configuration shown in Fig. 4.2 for horizontal and vertical dipoles at wavelengths of 450, 570 and 640nm for various thicknesses of the intermediate layer (0.5 2, 5,10, 20 um) with and without absorption (k=0.001). The corresponding raytracing results are also shown for comparison. The strong influence of the absorption on the extraction efficiency is apparent. We should emphasize that in a standard calculation it would have been difficult to disentangle the effect of the truncation error (which increases with the layer thickness) and the absorption.
The domain decomposition method was successfully implemented and applied.
Horizontal dipole Vertical dipole
Wavelength (nm) 450 570 640 450 570 640
D=0.5um 84 (77) 85 (78) 81 (74) 70 (59) 68 (60) 68 (62)
D=2.0um 81 (63) 85 (73) 87 (73) 67 (49) 72 (55) 69 (55)
D=5.0um 81 (55) 81 (60) 79 (60) 72 (40) 69 (44) 69 (48)
D=10.um 76 (39) 83 (51) 80 (48) 66 (25) 65 (31) 68 (38)
D=20.um 79 (25) 81 (36) 75 (34) 65 (13) 67 (19) 69 (25)
Raytrace (no lenses) 92 (59) 94 (60) 94 (50) 89 (17) 92 (35) 93 (52)
D=0.5um (with abs) 84 88 87 80 84 86
D=2.0um (with abs) 75 80 79 68 74 77
D=5.0um (with abs) 61 68 68 53 60 65
D=10.um (with abs) 45 55 54 37 46 52
D=20.um (with abs) 29 38 38 21 29 36
Table 1: extraction efficiencies Extraction into glass(%) calculated with JCMsuite with 8x8 cells, values in brackets assume an absorption of k=0.001 of the high index layer. Results in red are calculated by raytracing (LIGHTEX/Rayplay)
Unfortunately there was not enough time and capacity for a systematic testing and application of the novel full wave approaches for large structures developed and implemented in task 4.3. Furthermore there are still some issues with the performance speed of the DG method. The aim of the investigation in task 4.4 would be the comparison of "full wave" solutions with hybrid solutions based on raytracing and BSDFs obtained by full wave simulations. The present contribution to task 4.4 consists in the calculation of the BSDFs of rough surfaces by raytracing, FDTD, FEM and the "Fourier method". To calculate the BSDF of an interface one has to calculate the scattering of radiation impinging from the top and bottom of the structure. The FDTD simulations predict much broader scattering lobes than are obtained with raytracing. Due to the statistical nature of the surface we were not able to produce complete BSDFS with closely spaced angles of incidence. But already the analysis of the BSDFs obtained so far indicates that "raytracing" of (microfacetted) rough surfaces does not give the right answers, so that there is definitely a need for full wave solutions.
(Feasibility of) Finite Element Modelling
The assessment of OLED efficacy by simulation requires the numerical solution of the full Maxwell equations for a radiating dipole which is a very challenging computational task. In the project the feasibility of this approach using advanced Finite Element Based (FEM) methods was investigated. The accuracy, efficiency and (computational) cost of these methods for simulating light extraction from realistic OLED structures was compared to Finite Difference Time Domain (FDTD) method.
The simulation of light extraction properties is a numerical challenge for the following reasons:
• A wavelength scan over the entire visible spectrum is needed.
• Many incoherently radiating dipoles have to be simulated at different positions with regard to the scattering structure.
• Metals give rise to the presence of plasmons with singular field profiles near metallic edges or corners.
• In practice material data are only given experimentally. A numerical dispersion model can be costly to implement for methods operating in the time domain like FDTD.
• The computational domain must be sufficiently large to avoid numerical truncation errors.
The FEM package JCMsuite was proven to be a very reliable and accurate tool for a fullwave analysis of the light outcoupling from OLEDs. By studying relatively small cathode features we gained the important insight that the simulation requires a very fine mesh width close to the metallic cathode. This is caused by the presence of a plasmon resonance at the metallic bumps. As a rule of thumb the metallic rim of the cylinders requires a meshing below 0.5nm to guarantee an accuracy of 1%. This kind of accuracy can only be achieved with nonstructured grids and is therefore not in the reach of the classical FDTD methods. To deal with OLED models of larger scale it was necessary to use iterative techniques such as the domain decomposition approach. We mention that the FDTD computations done by Kintech and Philips (using Lumerical) required significantly less computation time and memory for the same OLED benchmarks. However, the different FDTD solvers showed a strong discrepancy between each other and with JCMwave’s FEM implementation. It is therefore difficult to draw a final conclusion which method is preferable.
An important outcome is that the artificial truncation of the computational domain introduces a so far underestimated systematic error. As a consequence JCMwave developed a novel method which allows dealing with the infinitely periodic arrangement. It has been shown that the numerically delicate inverse Floquet transform can be computed accurately and efficiently with an adaptive integration technique. The Floquet transform method is considered as a good candidate for an efficient outofthe box simulator of OLED structures.
The domain decomposition method (or other Maxwell solvers) and the Floquet transform approach are complementary. It is possible and implemented in JCMsuite to apply the iterative domain decomposition solver to the unit cell problems arising after Floquet transform. This is particularly interesting for thick OLED structures, when even the unit cell problem becomes challenging with regard to memory requirements. Especially, when using a plane wave expansion in the plane layers, the computation time and the memory requirements can be reduced tremendously.
Evaluation of devices and experimental validation
In this work package, we work towards the following objectives:
• Develop numerical / analytical models for current distribution and heat dissipation in large area OLEDs
• Provide existing or experimentally collected material and device properties
Task 5.1: Manufacture and evaluate standard OLED model devices
Task 5.2: Evaluate multiscale light extraction technologies
Task 5.3: Current distribution and thermal distribution models for large area OLEDs
Task 5.4: Experimental validation of modelling tool
Small molecule semiconducting materials were evaluated where required in order to provide enough experimental data to support the multiscale modelling approach by EU and RU partners. In parallel, modelling of current distribution and thermal dissipation in large area OLEDs, and validation by experiments, were planned in order to address the important contribution of heating and inhomogeneity in large area devices on the operational efficiency of such devices.
Figure 7: Schematic overview of WP5 topics and interexchange with other work packages.
Task 5.1: Manufacture and evaluate standard OLED model devices
Philips Research defined a small molecule reference stack to be used in the project. This stack is based on a phosphorescent red emitter, Ir(MDQ)2acac, and otherwise contains materials of a known molecular structure that are thus accessible to molecular simulations. Measurement data was provided for holeonly devices (20 mm2) of mMTDATA and NPD, as well as for electrononly devices of TPBI and BAlq (using a nin structure with doped layers with negligible voltage drop). These allow to extract the PooleFrenkel mobility data of the majority carriers; the mobility results were also provided to the partners. Strictly speaking the mobility of the carriers depends not only on field strength but also on carrier density; nevertheless these data already give an indication of the electron or hole transport properties of the materials. Moreover, HOMO values for the materials of the reference stack as determined from UPS spectra were provided. The LUMO values were then estimated from the HOMO data and the optical band gap of the materials. The refractive index and extinction coefficients were determined by ellipsometry by TNO. Another quantity useful for molecular modelling is the density of the layers. Such information has been indicated by Russian partners to be essential when modelling layers of molecules. To this end XRR scans were taken with narrow slits to determine the thickness and density of the organic layers on quartz substrates. The data were fitted using X’pert reflectivity software and provided to the Photochemistry Centre (PCRAS). Tapping mode Atomic Force microscopy measurements were conducted to determine the surface roughness of these evaporated layers. All available measurements were summarized in D5.2.
In the later part of the project attention was paid to the electrical transport measurements of the individual materials in the IM3OLED stack. The charge transport in several materials has been analyzed in unipolar devices. In particular, the hole transport in the emissive layer of αNPD:Ir(MDQ)2acac was studied in detail. For this purpose devices were fabricated by both Philips and Holst Centre. The (temperature dependent) transport measurements were shared with WP3 as input for their modeling work. In addition to the αNPD:Ir(MDQ)2acac system, also the data on the hole transport in mMTDATA, and the electron transport of BAlq was shared with the consortium.
The small molecule reference stack that is being used in the project was defined by Philips Research in D5.1. This stack is based on the known holetransporting materials mMTDATA and NPD, the efficient red phosphorescent emitter Ir(MDQ)2acac, as well as the electrontransporting materials BAlq and TPBI. Devices with sizes of 20 mm2, 10 cm2, 48 cm2 and 163 cm2 were fabricated, of which the smallest devices provided the performance of the OLEDs without issues involved with scaling to larger dimensions. Typical analyses were performed, including currentvoltagelight cycles, angle dependent emission, efficiency analysis in an integrating sphere and brightness uniformity measurements. For further details the reader is referred to D5.1.
Task 5.2: Evaluate multiscale light extraction technologies
To improve the efficiency of OLEDs many techniques have been introduced that facilitate extraction of light from the OLED active layers and the substrate. The most effective way to do this is to make use of a high index substrate. The next step is to extract light from the high index glass. In applications, a combination is required that contains a high index layer on standard glass, combined with a manner to get light into air without significant losses in efficiency. Which methods are most successful cannot be predicted up front. It is quite possibly a combination of nano and macrostructures, where macrostructures could include scatter particles. The Im3OLED project aims to provide methods to calculate the impact of multiple multiscale features. Introducing multiscale features to enhance the light extraction from OLEDs is possible by combining externally applicable light extraction foils based on volume scattering or microlens arrays. Both type of foils are commercially available and wellknown to extract light from the glass substrate effectively. Within the OLED, between the glass substrate and ITO layer, it is possible to introduce a high index layer with some form of structuring, such as periodic nanostructures or for instance randomized surface roughness.
We examined the possibility of introducing nanofeatures into the OLED first. Inside the OLED the quality of NIL layers (nanoimprint lithography) is difficult to quantify once the OLED is finished. We therefore started with imprinting structures on foil (to avoid cracking of a glass substrate under the pressure of imprinting). The features are not subwavelength as is the intention, but have a periodicity close or equal to the wavelength of light. It is wellknown that antireflective films can be made with the Motheye structure. The reflectance measurements, however, did not provide us with a sufficient reduction in reflectance. Efforts to find a suitable resist with appropriate refractive index that gave better results on imprinting proved to be unsuccessful and too timeconsuming and an alternative for achieving structures on glass was examined instead.
At the High Tech Campus facilities it was possible to increase the roughness of glass by means of sandblasting. Sandblasting can create a high topology with too high structures which may cause shorts. For this, we used chemicalmechanical polishing of the substrate after sandblasting as a means to keep the topology low. At low intensity of sandblasting, the glass surface was visibly modified, but the OLED emission was not found to be different with and without external outcoupling film. When increasing the intensity of sandblasting from 0.8 to 1.1 and upon processing the IM3OLED stack on top of the substrates, the efficiency was found to be significantly lower due to shorts coming from the rougher surface.
To cover the randomized roughness induced by sandblasting, a high index organic coating was developed. In early attempts, it was possible to increase the refractive index substantially without consequences to the processibility and transparency. Later batches of the nanoparticles, however, turned out to be unexpectedly different, leading to brittle and opaque films. Despite our efforts to resolve the issues, we could not introduce these layers into the OLEDs as planarization layers.
A few changes were made in order to allow device to be made on sandblasted glass. First, a new stack had to be adopted. This stack design is different from the previous OLED stack because of the method of encapsulation. Several materials from the former Im3OLED stack were reused in this stack design, however, some needed to be replaced. The emitter was replaced with another common Irbased material: Ir(ppy)3. Onto the Al cathode, the encapsulation was deposited, starting with a silicon nitride layer. This SiN layer is deposited at elevated temperatures in a PECVD process. The former combination of organic materials did not perform well after this procedure and thus had to be replaced. The leakage currents increased by a factor 104105 and the performance plummeted in a matter of tens of minutes. Secondly, polyimide (commercial) was used instead of an acrylate containing functionalized nanoparticles. The devices on Eagle glass performed well after encapsulation and showed low leakage currents and high current efficacies of 4050 cd/A.
a) b)
Figure 8: a) glass substrate with 4 patches, each with its own sandblasting intensity (top left 0.8 bottom left 1.1 bottom right 1.2 top right 1.4) b) white light interferometry of the area selected for a green large area OLED of 64 cm2 on a 6” glass substrate (sandblast intensity 1.1). Mean height Ra is 8 nm, root mean square Rq 11 nm, and the maximum peaktovalley height Rt is 383 nm.
a) b)
Figure 9: a) microscopy image of sandblasted glass with PI layer (23 microns) and the unstructured Aluminum layer (intended for metallic outline and internal grid for OLED); b) same, but after etching of the Aluminum layer.
The device on sandblasted glass had a current efficacy of ~ 60 cd/A, which was an approximately 4050% improvement over the reference device (40 – 45 cd/A). The additional application of an external scattering foil increased the efficacy even to ~ 90 cd/A, approximately double the reference efficacy. All relevant measurement of this batch were distributed in the consortium in deliverable D5.7. The efficiency of the devices was further measured with and without structuring and external foil using the DMS at a constant current density level. Evaluation of the emitted photons indicates that the EQE (ratio photons and charge carriers) improved from 14 % to 23 %. These values were obtained by measuring the EL spectra in the angle range of 070, and were extrapolated to the range of 090°.
Figure 10: OLED with internal scattering and a high n planarization layer, in (left) offstate and (right) on state. In the offstate the roughness of the sandblasted substrate is clearly visible.
Task 5.3: Current distribution and thermal distribution models for large area OLEDs
Organic lightemitting diode (OLED) based products for lighting applications have become commercially available in the recent years. Example products include for instance Lumiblade (Philips), Orbeos (Osram) and Fenalene (Lumiotec). The efficiency of OLEDs is the resultant of the interplay between electrical and optical properties and processes at various length scales. An OLED device will typically be characterized by its macroscopic parameters, such as emission colour, intensity and efficiency, and lifetime. The macroscopic performance indicators and architecture of the OLED are expected to be correlated. Especially large area devices intended for lighting applications require sophisticated electrode designs and knowledge on the lateral current distribution. Heat generation, as a result of electrical losses on micro and macroscopic scales, induces a further deterioration of performance that has considerable consequences. The lateral distribution of heat dissipation affects the OLED performance at multiple length scales, even down to the molecular level. Driving a lighting tile of 15 x 15 cm2 at very high brightnesses requires an optimized architecture to avoid significant losses to the OLED performance. Due to limited conductivity of typical transparent anodes, OLED tiles require a metal support structure in order to obtain sufficient light homogeneity (Neyts et al., 2008). An important question is then which geometrical properties (size, grid dimensions, pitch) the metal support structure should have for the realization of homogeneous largearea and highefficiency OLEDs. We will describe here our evaluation of the role of transparent anodes employing various metal support structures in combination with a transparent conductor, and improved emitter materials on the homogeneity and design of largearea OLEDs. This work was published in Journal of Applied Physics.
The aim within this part of WP5 is to be able to simulate large area coupled electrical and thermal behaviour of an OLED. The method mentioned above will always work to solve this problem, however it is fairly inefficient with respect to degrees of freedom and memory consumption and therefore poses its own practical limitations. Therefore, the aim was to perform this kind of modelling more efficiently.
Feasibility of Finite Element Modelling
An analytical solution for the current distribution of a largearea organic light emitting diodes (OLEDs) with parallel equidistant gridlines was derived in 201222. In contrast to numerical methods, this analytical solution allows for a very quick scan of the OLED design space, even for very large OLEDs, providing insight how different model parameters affect each other. The analytical solution for the electrical domain of the anode was programmed within MATLAB (Mathworks, Natick, MA). This analytical solution can handle infinitely large model dimensions while maintaining a high accuracy. However, it includes some assumptions. The assumptions within the analytical derivation were subsequently verified with finite element simulations of the same OLED.
The analytically calculated light distribution was experimentally verified by measuring the light distribution on a largearea OLED (12 by 12 cm2) on heatstabilized Teonex polyethylene naphtalate (PEN, DuPont Teijin Films) with a moisture barrier. The measured homogeneity of the experimental OLED was roughly 84% at a brightness of 500–600 cd/m2. This corresponded fairly well with the calculated homogeneity of the analytical model which was 81%, where the maximum and minimum light output was calculated to be 600 and 485 cd/m2.
Figure 11: Analytical model applied to a 12x12 cm2 flexible OLED on PEN22.
2D vs 3D numerical FEM modelling
Although the analytical model is very suitable for large area calculations, it can only be applied to line grids and cannot be expanded to the thermal domain. Therefore, a second approach is required which will allow us to model any kind of grid including electrothermal interdependency.
The current distribution of OLEDs can also be modelled very well with Finite Element Analysis. With finite element modelling, the object (OLED) is subdivided into small blocks, called elements. In 3D, these small elements should have reasonable aspect ratios. As the different layers of an OLED are very thin and the gridline pattern may result in very fine structures, one can easily imagine that a 3D FE simulation will require an enormous amount of these elements and more elements require more computer memory. These models become big very quickly and therefore 3D FE analysis of OLEDs can only be applied to fairly small OLED sizes, but not to large area OLEDs. Therefore a 2D FE approach was developed which should allow us to perform large area current distribution simulations and which can be expanded to electrothermal distribution simulations.
For a standard 3D FE current distribution simulation, it is required to have the dimensions (length, width and thickness) of the OLED layers, the electrical conductivities of the electrodes, the currentvoltage characteristics of the light emitting material and the potential which is applied to the anode. The electrical conductivity of the light emitting material is many orders of magnitudes lower than the electrical conductivity of the electrodes (approx. 106 to 104 or 107). Therefore, the current will tend to spread very well on the electrodes and will take the shortest route (straight down) through the light emitting material. Hence the current transport within the electrodes is mainly planar (2D) and the current transport through the light emitting material is mainly 1D. This makes it possible to apply an enhanced 2D approach: both electrodes are just modelled in 2D and the connection between both layers is made by a ‘continuity of currents’, which means that when a certain amount of current leaves the anode at a certain lateral position for the light emitting material, the same amount of current should enter the cathode at the same lateral position, but at a different height in the device.
To determine whether the enhanced 2D approach is valid for modelling the current distribution, a comparison was made between a full 3D model and an enhanced 2D model. Both models were made in Comsol Multiphysics; the 3D model in version 4.2a and the enhanced 2D mode in version 3.5a. The potential difference (between anode and cathode) of both models were found to be nearly identical which shows that the enhanced 2D approach provides a good replacement for a full 3D approach.
a) b)
Figure 12: Potential distribution for the 3D model (a) and for the 2D model (b) for a 4 cm2 device
Electrothermal modelling
The 2D current distribution model was expanded with a 2D electrothermal model for OLED simulations. The electrical and thermal model results were compared with the electrical and thermal behaviour of devices of various sizes, which were produced and measured by Philips. The simulations were performed in Comsol 3.5a in a fully coupled (electrothermal) mode, using the direct UMFPACK solver.
The results of the simulations shows that the inclusion of the temperature within the electrothermal models significantly improves the light output results. Thermal behaviour and dependencies need to be incorporated in the OLED simulations: the homogeneity of the electrothermal is a clear improvement over the homogeneity of only the electrical model. The simulated thermal distribution of the models showed a good agreement with the experimental results.
Figure 13: current distribution model with heat dissipation: measurement and model for light emitted and temperature of a 163 cm2 OLED
Task 5.4: Experimental validation of modelling tool
The experimental validation in this project was reflected in two main activities. First, the results of abinitio studies from the Russian partners was verified with the experimental values obtained in the project. Secondly, the modelling tools from the European partners was validated with experimental data, such as currentvoltage measurements.
During the project the results from the Russian side were continuously compared with the experimental values. An example of such a comparison was already shown in section 3.1.1 on WP2. The main conclusion was these comparisons was that the abinitio studies often yield the correct trends and magnitudes of values such as HOMO and LUMO energies and disorder parameters, but that there is some deviation. Electrical driftdiffusion simulations are extremely sensitive to the strength of the disorder and to the energy levels, in the case of injection barriers, or internal barriers. The values obtained from abinitio studies are therefore not suitable to be used directly as input parameters in the electrical modelling. However, the tools developed may be used to screen a large selection of materials for trends in their electrical properties, or gain a better understanding of the structureproperty relation.
For electrical and optical modelling, Setfos was used throughout the project for the interpretation of currentvoltage measurements. For instance: Setfos was used to extract mobility parameters from the (temperaturedependent) transport measurements. As a final test, Setfos was used to model the currentvoltage characteristics of the complete IM3OLED device. In this simulation the electrical transport in the layers was expressed by a fielddependent mobility, of which the parameters were measured by Philips. The result is shown in the figure below.
Figure 14: Experimental and simulated current for a 20 mm² OLED with the IM3OLED stack. The simulation was performed with Setfos.
The simulation is in qualitative agreement with the measurement. However, there is some difference between the experiment and the simulation. This difference may be explained in the uncertainty of many of the parameters that are involved in the modelling of a full OLED stack. For instance, the energy levels, and more specifically the offset between the energy levels of the layers have a critical influence on the current density. Furthermore, in the presence of traps was not taken into account since no characterization of traps was done in this project.
Due to the late realization of the combination of internal and external structures, it was not possible to compare the experimental results to the models. In D4.6 a planarized rough substrate was considered. The refractive index of polyimide reaches 1.7. The modelling takes into account a value of 1.8. The layer thicknesses do coincide. All relevant elements have been implemented in JCMsuite, including a stochastic roughness.
Material d N
Air semiinfinite 1
Glass 1 mm 1.5
Planarization 3 um 1.8+i*2*104
ITO 100 nm 1.94
HTL 100 nm 1.75
ETL 80 nm 1.75
Cathode 100 nm 0.77+i*5.6
Air semiinfinite 1
Comparing the measured enhancement of the light extraction samples with polyimide and sandblasted glass to examples in literature shows that similar gains are obtained. Our green OLED improved by 30 % in brightness and close to 40 % in EQE. One benchmark method in D4.1 to improve light extraction was a 2D photonic crystal, achieved by etching glass and depositing SiN. The OLED improved by 52% in brightness and 41 % in EQE when a subwavelength periodicity was used. A higher periodicity (500 nm) provided a gain of 35 % to brightness and 55 % in EQE [Do et al., J Appl Phys 96 7636]. In another document [Optics Express 2014 22 A705], internal scattering with a high index hybrid layer (with TiO2 particles) was published to provide a gain of 2.36 in total when combining it with low index scatter particles (SiO2) within the TiO2 layer. In our case, a close to 2.0 gain to brightness (1.7 to EQE) was obtained with scatter foil. Unfortunately both examples are too different to allow good comparison. These promising results encourage us to continue to investigate further along the lines of IM3OLED.
Multiscale modelling tool assembly [WP6]
Concerning the EU scope of this project, the objective of this workpackage is to expand the commercially available software SETFOS with models developed in other workpackages. The development of the RU platform is summarized in D5.8. The final objective of WP6 was to get commercially available software covering different scales: from microscopic charge transport and dipole emission to large area panels and light propagation in ‘optically’ thick devices featuring scattering properties for light outcoupling enhancement.
In this project we thus have worked on the following task (according to the DoW):
Task 6.8: Expand Setfos with models developed in WP3, WP4, WP5 (Fluxim)
Description of SETFOS for multiscale/multiphysic modelling
The driftdiffusion module in Setfos makes use of specific numerical methods to calculate the charge carrier distribution, the recombination and the current in the device. This module also solves the exciton diffusion equation. This last quantity can then be used by the optical emission module to compute, for example, the emitted spectrum of a flat OLED, the radiance but also the outcoupling efficiency of the device. The emission module of SETFOS solves the Maxwell equations for an oscillating dipole embedded in a flat OLED stack using a semianalytical formalism. The light scattering module (combined with the emission module) allows calculating the light emission of OLED stacks containing scattering structures like rough interfaces or scattering layers. Therefore, combining these three modules, it is possible to perform an OLED simulation on different length scales from electrical to optical simulations. The figure below summarizes this workflow:
Figure 15: Simulation workflow of Setfos for optoelectronic device simulation of OLEDs
Since the beginning of the project the simulation software SETFOS has continuously improved including some new features coming from the different workpackages. Deliverable 6.2 reported an exhaustive list of the new features which are now available in SETFOS. This list of new features is summarized in the following.
In February 2013 SETFOS 3.3 was released and included a first version of the impedance solver using a smallsignal analysis formalism. This tool allows simulating, for example, the capacitance of the device versus the voltage and/or the frequency.
SETFOS 3.4 was released in December 2013. This new version came with several improvements:
 A new physical description of the organic/organic interface model was implemented, improving the simulation results of multilayers OLEDs. The impedance solver was extended such that the impact of traps and the impact of their dynamic can now be simulated in this simulation mode.
 A first version of the multiscale optical model was released, allowing simulating OLEDs made of any sequences of thin coherent and thick incoherent layers. This feature is of practical interest to simulate OLEDs containing a thick color filter or encapsulation layers.
In July 2014, SETFOS 4.0 was successfully released, see the product flyer at
http://www.fluxim.com/fileadmin/user_upload/downloads/setfos_4.0_light_scattering.pdf
This new release includes a new light scattering module which can simulate OLEDs with scattering interfaces or layers.
The corresponding optical multiscale modeling framework of Setfos 4.0 is illustrated in Figure 15. This release forms the basis for further feature extensions, such as implementations of the Mie particle scattering model or Fourier optics algorithm, within SETFOS. The inclusion of a Mie scattering algorithm inside the develop version SETFOS came at the end of the project, nevertheless this algorithm was successfully benchmarked with other simulations as presented in D4.6.
Thanks to the efficient model implementation in Setfos, the computation time is short and the calculation of fullspectra for whiteemitting OLEDs with scatter layers is feasible (completion within a couple of minutes). The outcoupling enhancement predicted for this particular case of Figure 5 is 2.1 which is an excellent value rarely achieved in experiment. In more realistic scattering films, light may suffer absorption and particles may not be monodisperse. Setfos takes care of both of these generalizations that lead to somewhat reduced outcoupling enhancement figures.
Electrothermal multiscale model.
From an electrical point of view, an OLED device for lighting application can be regarded as a succession of thin (few tens of nanometers) semiconducting layers sandwiched between an anode and cathode. One of these electrodes requires being transparent so that the emitting light can escape the device, however transparent electrodes are fairly bad conductors leading to an inhomogeneous potential distribution in the large area electrodes and thus to electrical losses. As already demonstrated in D5.5 due to the high aspect ratio (large area electrode versus the through plane thickness) of an OLED, the electrical model can be split up into a 2D FEM model, to solve the current distribution in the large area electrodes, and a 1D microscopic model, to compute the current flowing vertically in the device solving the semiconductor equations. This approach was compared with experiments in D5.6.
From a thermal point of view, the thin film layer stack can be seen as a heat source that will dissipate into the electrodes and substrates, obviously the heat flux entering the electrodes will depend on the power consumption of the thin film layer stack.
Figure 16: Combination of microscopic and macroscopic electrical models
Within the IM3OLED project a multiscale electrothermal model was developed. Unfortunately this feature is not yet commercialized and further developments will still be undergone at Fluxim after the end of the IM3OLED project. However some first simulations could already been done using our in house solver.
Figure 17: Potential distribution in the top OLED electrode including highly conductive stripes and shunts.
Potential Impact:
The potential impact
4.1.4.1 Impact
Photonics will play a key role in addressing the challenges of energy efficiency and moving to a lowcarbon economy. In the future, solidstate light sources will outperform almost all other light sources in terms of efficiency, offering potential energy savings of 50% or even more. At present, the high costs of research and development has grown beyond what a single company can sustain. Large subsidized consortia are formed to pool resources, share facilities, reduce costs and accelerate development. The availability of sophisticated modelling tools that provide insight into complex optoelectronic devices at various levels of operation provides another level of cost reduction since the necessity to fabricate complex devices is reduced. In particular on topics such as the electrical and optical performance of optoelectronic devices, the modelling tools have matured and are capable of providing the necessary insights. IM3OLED contributed to the array of modelling tools by investigating many aspects of the organic lightemitting diode and evaluating the methods by which the properties of such devices are calculated. This knowledge is immediately available to the industry via the software packages that have been expanded on very relevant topics.
One of world’s leading OLED scientists, Dr. Tyan (former Eastman Kodak) described light extraction to be the last hurdle in order to reach the performance required to commercialize OLEDs in lighting applications. In the laboratory OLEDs already demonstrated efficiencies higher than those of luminaires based on other lighting technologies. Large area OLEDs, however, still lag behind. IM3OLED had these two items incorporated in its work program.
Adding flexibility to OLEDs will create huge number of innovative products, given their already high versatility in terms of colour tuning, daylight simulation, comfort of use, form factor, and the possibility of integrating other electronics into the light source. Flexible OLED technology has great potential to strengthen Europe’s leading position in general lighting and establish new development and manufacturing facilities within Europe. It is important that available modelling tools are equipped to accurately describe such devices.
 IM3OLED expanded existing software to include the presence of multiple incoherent layers, thereby making it appropriate to future OLEDs on flexible substrates with thin film barriers
 IM3OLED expanded existing software to include scaling effects as resulting from Ohmic losses when going from mm2 to cm2 and beyond. The software is also capable of integrating temperature effects. Such models have been examined in the project and have been verified with devices from Philips.
 IM3OLED investigated the impact of computational resources and made improvements to Setfos and JCMwave. Setfos has enabled multicore calculations and has developed a more rapid method to calculate scattering in devices. JCMwave developed novel methods to handle multiple length scales in a single model, as well as alternative fullwave approaches with higher accuracy than FDTD.
 IM3OLED examined various means to improve the efficiency of OLEDs and assured that these can be predicted accurately using the tools examined in the project. Application of the available fullwave methods, namely FDTD, 3DFEM and TDDG (time domain discontinuous Galerkin method) to these benchmarks has provided valuable insights into the applicability of these methods and their reliability. In some cases raytracing suffices, but in others (when it concerns random surface textures) it does not. This information is crucial to avoid predicting incorrectly what combination of light extraction features will result in optimal performance.
As stated before, IM3OLED is unique in the direct implementation of the results obtained within the project into commercial tools. The relevance of the project’s objectives on today’s topics in optimizing OLED therefore directly impacts the research and development of OLEDs for lighting application.
4.1.4.2 Impact to beneficiaries of the project
For the beneficiaries of the project in particular, the project provided valuable insights into the requirements of the software tools for the OLED industry. Established and new computational methods were examined. With an expanded arsenal of computational physics, and examination of their reliability in the benchmarks in this project, it is possible for Philips as an OLED Lighting developer to apply these methods without concerns about their reliability. Moreover, their participation in the project guarantees immediate distribution within the company and the work floor.
For software companies Fluxim and JCMwave, their software packages have been upgraded while doing the research in the IM3OLED project, thereby improving the fit of the programs to the requirements specified by the OLED companies. Three Fullwave methodologies have been investigated (FDTD, 3DFEM, TDDG) with each their own advantages and disadvantages. For computation physics institute ZHAW there are opportunities to examine novel fullwave methods with existing and proven benchmarks. This exceptional benefit will enable a faster development of new software tools that may ultimately become available via Fluxim.
For research institute Holst Centre, IM3OLED provided insight into the performance of large area devices with and without multiple light extraction features. While developing the experimental methods to apply high index layers on roughened surfaces, several experimental issues have to be overcome. At the end of the project, Holst Centre succeeded in making a 64 cm2 demonstrator on glass featuring doubling of the efficiency. The project also strengthened the ties to software companies that provide the necessary tools to gain insight into the behaviour of such devices. Via the Holst Centre a community of companies is exposed to the developments in IM3OLED. Holst Centre has monthly meetings with its industrial partners in program meetings and additional partner days where all industrial partners meet and discuss the advances in the programs. Advances as made in the IM3OLED project are important to the industry, and thus the events with industrial partners will be taken to disseminate the projects objectives and achievements.
The RU partner’s exceptional knowledge on mathematics, computational physics, and software development led to many interesting discussions during the project. The collaboration with the Russian partners resulted in a discussion at the Holst Centre on the participation of these Russian partners in some of the programs. Although at present, such a collaboration is not in place, the project has revealed the many qualities of the Russian scientists. Kintech offers the OLED database on trial license and may, with sufficient interest, commercialise this.
4.1.4.3 Socioeconomic impact
The targets and achievements of the IM3OLED project contribute to the combined efforts in EU to strengthen its industry. The European lighting industry has always been at the forefront of innovation. Still today Europe is the leading region when it comes to technological development and scientific research related to light in its different applications.
The sector is driven by high innovation potential, accounts for an estimated €20 billion turnover, and represents over 100,000 jobs in Europe. The lighting industry in Europe is and always has been a sector that is SME driven when it comes to the production of luminaires and of high value products. This SME culture within the industry is a crucial precondition to provide the market with highly decorative, innovative, and sustainable products.
The dual challenge facing Europe is both to lead in photonics technology innovation, and to exploit these results though successful commercialisation. The IM3OLED project has done this in a very direct fashion, namely by direct integration of its research into commercial tools. Furthermore, due to its large network of industrial companies and academia, Holst Centre/TNO is especially well placed to integrate new knowledge via their shared innovation framework.
4.1.4.4 Environmental impact – huge reduction in electricity consumption by going to SSL
The energy savings in the USA is typically forecast to a high degree. The predictions can be taken as an indication for the world. The energy savings were forecast to be be close to 50% by 2030. The annual savings would correspond to 300 terawatthours, which, at today’s energy prices, would be equal to $30 billion. Furthermore, the energy savings would reduce greenhouse gas emissions by 210 million metric tons of carbon and over a time span of 20 years as of now accumulate to 1.8 million metric tons of carbon emission. The energy savings projection assumes significant progress in efficient SSL sources, as well as widespread market adoption. Specifically, by 2025, SSL sources would need to realize a luminaire efficacy of 200 lumens per watt (lm/W) and market penetration, in terms of lumenhours, of about 60 percent. A conservative estimate is that by 2030 high efficacy flexible OLEDs could represent at least 10 20 % of solid state light sources, and therefore be responsible global energy savings of at least 510 % compared to today’s levels (30 terawatthours, or $3 billion saving in energy bills), assuming continued improvements in OLED efficacies.
IM3OLED contributes to achieving the goals set by the industry by providing the necessary modelling tools that can accurately predict the impact of changes within the OLEDs. As stated earlier in chapter 4, the modelling tools are capable of modelling all aspects of flexible OLEDs, including combinations of various light extraction methods.
4.1.4.5 Main dissemination activities
Publications and proceedings
WP2:
 By Russian Partner PCRAS:
“Atomistic multiscale simulation of the structure and properties of an amorphous OXD7 layer”
Chemical Physics Letters, Volume 590, p. 101105
Emelyanova, Svetlana; Chashchikhin, Vladimir; Bagaturyants, Alexander
WP3:
 “The Role of Shallow Traps in Dynamic Characterization of Organic Semiconductor Devices”
Publication, to be submitted in April 2012 by ZHAW
Evelyne Knapp and Beat Ruhstaller
Institute of Computational Physics, Zurich University of Applied Sciences
WP4:
 “Numerical analysis of nanostructures for enhanced light extraction from OLEDs”
Proc. SPIE 8641, LightEmitting Diodes: Materials, Devices, and Applications for Solid State Lighting XVII, 86410B (4 March 2013); doi: 10.1117/12.2001132
Lin Zschiedrich; Horst J. Greiner; Sven Burger; Frank Schmidt
Philips Technology and JCMwave
 “Simulation of advanced OLED light extraction structures with novel FEM methods”
Proc. SPIE 9137, Organic Photonics VI, 91370O (1 May 2014); doi: 10.1117/12.2054146 Lin Zschiedrich; Therese Blome; Horst J. Greiner
Philips Technology and JCMwave

WP5:
 “Analytical model for current distribution in largearea OLEDs with parallel metal grid lines.”
Publication, published in Journal of Applied Physics in April 2012 by TNO
Marco Barink, Stephan Harkema
 “Analytical optimization of metal grids in large area flexible and ITOfree OLEDs”
Publication, rejected by Organic Electronics in April 2012 by TNO. Not resubmitted.
Stephan Harkema, Marco Barink, Gert van’t Heck, Harmen Rooms, Jeroen van de Brand, and Ton van Mol
 “Electrical Modelling of Largearea Organic Lightemitting Devices”
Abstract, submitted to SCEE 2012 (Zürich, Switzerland) in September 2012 by ZHAW
Evelyne Knapp and Beat Ruhstaller
Institute of Computational Physics, Zurich University of Applied Sciences
 “Light management in Flexible OLEDs”
Proceedings of SPIE
Stephan Harkema, Raghu K. Pendyala, Paul L.J. Helgers, Christian G.C. Geurts, Jack W. Levell, Joanne S. Wilson, Duncan MacKerron
WP6:
 By Russian Partners
“Simulation Platform for Multiscale and Multiphysics Modeling of OLEDs”
Procedia Computer Science 2014 Volume 29, 2014, Pages 740–753
M. Bogdanova, S. Belousov, I. Valuev, A. Zakirov, M. Okun, D. Shirabaykin, V. Chorkov, P. Tokar, A. Knizhnik, B. Potapkin, A. Bagaturyants, K. Komarova, M.N. Strikhanov, A.A. Tishchenko, V.R. Nikitenko, V.M. Sukharev, N.A. Sannikova, I.V. Morozov
WP7:
 Presentation of the IM3OLED project at the ISFOE 2012 Conference by TNO.
The project will be presented during the Special Session “Strategy & R&D Projects in Europe, USA & Asia in Flexible Organic Electronics”, Thessaloniki, July 2012.
Alessia Senes (TNO)
Presentations
WP4:
 “Modelling light extraction from (O)Leds with the vector wave Helmholtz equation”
Presentation held at Nano Photonics Workshop at the ZUSE institute (Berlin, Germany) on 23 February 2012
Horst Greiner (Philips)
 “Advanced simulation of OLEDs and organic solar cells”
Keynote lecture held at OE Winter School at the University Heidelberg (Heidelberg, Germany)
Beat Ruhstaller (ZHAW)
 “Light management in Flexible OLEDs”
Invited presentation at SPIE Optics & Photonics 2014
Stephan Harkema (TNO)
Public website
This website has been developed by SEG (Visual Design), Retack Design (Web Design and Implementation) and Talaria Communicatie (Text advice & Translations) in close collaboration with TNO and was launched in Jan 2012. The costs involved were 4,100 Eur.
The website features a content management system based on JOOMLA to allow the project members to adjust the descriptions on the website. The website features all necessary subsites including those for events, news, publications, presentations, partners, etc. and has a convenient method to share the existence and contents of the website on social media.
Another website that has been realized is the sharepoint site for direct exchange of information between project members of EU and RU collaborative projects. Hosting and arranging the site has been taken care of by TNO. It was decided to not expand the external site to also enable file sharing since this would require a much more complicated website with more securities installed. Sharepoint has been hosted by TNO for some time and is and has been used for dozens of EU projects involving TNO. Access to the site is limited by username and personal password that has been provided to all project members by TNO.
List of Websites:
http://www.im3oled.eu/