Machine Learning for OLED Device Design

Tutorial Created with Software Release: 2026-2
Topics: Organic Electronics, Thin Film Processing
Methodology: Machine Learning
Products Used: MS Maestro, OLED Device ML

Tutorial files

1.2 GB
This tutorial is written for use with a 3-button mouse with a scroll wheel.
Words found in the Glossary of Terms are shown like this: Workspacethe 3D display area in the center of the main window, where molecular structures are displayed

 

Tip: You can hover over a glossary term to display its definition. You can click on an image to expand it in the page.
Abstract:

In this tutorial, we will learn how to train a machine learning model to predict properties of OLED devices. We will subsequently apply this trained model and a pre-trained model available in the panel to predict target properties for new OLED devices unseen during training. Lastly, we will train an additional model in a stacking approach, combining structural and electronic features of OLED devices.

Tutorial Content
  1. Introduction
  2. Creating Projects and Importing Structures
  3. Training an ML Model to Predict Properties of OLED Devices
  4. Using Pre-trained Models to Predict Properties of OLED Devices
  5. Training a Stacked ML Model to Predict Properties of OLED Devices
  6. Conclusion and References
  7. Glossary of Terms

1. Introduction

Organic Light-Emitting Diodes (OLED) are a class of optoelectronic devices that emit light in response to an electric current. They are widely used in display and lighting technologies due to their advantages such as high contrast ratios, wide viewing angles, low power consumption (for dark images), and the potential for flexible and transparent displays.

Figure 1. Schematic of a typical OLED device.

OLEDs operate based on electroluminescence. When a voltage is applied across the device, electrons and holes are injected from the cathode and anode into the organic layers. These charge carriers migrate through the transport layers until they recombine in the emission layer (EML) to form excitons. The subsequent radiative decay of these excitons results in light emission. The efficiency of this process depends on factors such as charge balance, exciton formation yield, and quantum yield of the emitter.

To improve performance, OLEDs typically employ a multilayer architecture with each layer serving a specific role:

  • Hole Injection Layer (HIL): This layer is positioned directly above the anode. The HIL facilitates efficient injection of holes into the neighboring hole transport layer (HTL).
  • Hole Transport Layer (HTL): This layer transports holes from the HIL toward the emission layer.
  • Emissive Layer (EML): The core layer where electrons and holes recombine to form excitons. The EML contains the emitter (fluorescent, phosphorescent, or TADF molecules) often dispersed in a host matrix. The choice of host and dopant affects color purity, efficiency, exciton dynamics, and device stability.
  • Electron Transport Layer (ETL): This layer assists in the movement of electrons from the cathode towards the EML.
  • Electron Injection Layer (EIL): Located adjacent to the cathode, the EIL improves electron injection by modifying the interface between the cathode and ETL, lowering the barrier for electron transfer.
  • Blocking Layers:
    • Hole Blocking Layer (HBL): Positioned between the EML and ETL, this layer prevents holes from leaking into the ETL, helping to confine excitons in the emission zone.
    • Electron Blocking Layer (EBL): Positioned between the HTL and EML, the EBL blocks electrons from leaking into the HTL, further promoting recombination within the EML.

These layers can be tuned in terms of thickness, energy levels, mobility, composition, and morphology to achieve optimal charge injection, transport, and recombination, thereby maximizing device efficiency and lifetime. Understanding the functions of these layers is crucial for comprehending how OLED devices operate and how their performance can be optimized (see References).

Despite significant advancements, OLED device performance is strongly influenced by both the device configuration and material composition. Machine Learning (ML) models present a powerful approach to OLED device design optimization. ML can be used to predict and optimize the blend ratios and thickness of different layers to achieve desired device properties like:

  • External quantum efficiency (EQE): Measures how efficiently the device converts electrical energy into usable light output. It's a key metric for assessing the overall performance and energy efficiency of an OLED device.
  • Electroluminescence maximum emission wavelength (λmax): Determines the specific color of light emitted by the OLED device, making it essential for tailoring devices for applications such as displays or lighting.
  • Operation lifetime: Refers to the durability and stability of the OLED device under continuous operation. This property gauges how long the device can maintain its performance, such as brightness and color accuracy, before degradation occurs. Optimizing operational lifetime is vital for ensuring reliable, long-lasting devices.

In this tutorial, we will utilize the OLED Device Machine Learning panel to predict OLED device properties. We will begin by importing device data, which includes layer composition and thickness specifications. Subsequently, we will train an ML model using the experimental data for λmax and assess its performance. We will use the trained model for prediction on a different dataset. Additionally, we will use a pre-trained model that is available in the panel for predictions. Lastly, we will use an approach called model stacking to train an improved model. This combines the structural features of OLED devices with electronic features, which are predicted by additional ML models. The overall workflow is as follows:

Figure 2. Overall workflow of the OLED Device Machine Learning panel. The input CSV file containing the OLED device is inputted into the panel and used to train a ML model to predict OLED properties. The trained model is then used to predict properties of new OLED devices.

2. Creating Projects and Importing Structures

At the start of the session, change the file path to your chosen Working Directorythe location where files are saved in MS Maestro to make file navigation easier. Each session in MS Maestro begins with a default Scratch Projecta temporary project in which work is not saved, closing a scratch project removes all current work and begins a new scratch project, which is not saved. A MS Maestro project stores all your data and has a .prj extension. A project may contain numerous entries corresponding to imported structures, as well as the output of modeling-related tasks. Once a project is saved, the project is automatically saved each time a change is made.

Structures can be built in MS Maestro or can be imported using File > Import Structures (or drag-and-dropped), and are added to the Entry Lista simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion and Project Tabledisplays the contents of a project and is also an interface for performing operations on selected entries, viewing properties, and organizing structures and data. The Entry Lista simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion is located to the left of the Workspacethe 3D display area in the center of the main window, where molecular structures are displayed. The Project Tabledisplays the contents of a project and is also an interface for performing operations on selected entries, viewing properties, and organizing structures and data can be accessed by Ctrl+T (Cmd+T) or Window > Project Table if you would like to see an expanded view of your project data.

  1. Double-click the Materials Science icon

Figure 2-1. Change Working Directory option.

  1. Go to File > Change Working Directory
  2. Find your directory, and click Choose
  3. Pre-generated files are included for running jobs or examining output. Download the zip file here: schrodinger.com/sites/default/files/s3/release/current/Tutorials/zip/oled_ml.zip
  4. After downloading the zip file, unzip the contents in your Working Directorythe location where files are saved for ease of access throughout the tutorial

Figure 2-2. Save Project panel.

  1. Go to File > Save Project As
  2. Change the File name to oled_ml_tutorial, click Save
    • The project is now named oled_ml_tutorial.prj

3. Training an ML Model to Predict Properties of OLED Devices

In this section, we will use the OLED Device Machine Learning panel to train a model for the prediction of electroluminescence maximum emission wavelength (λmax) of OLED devices and use it for predictions.

Figure 3-1. Loading the training data.

  1. Go to Tasks > Materials > Informatics > OLED Device Machine Learning
  2. To load the training data click Load CSV and navigate to the provided files, presumably in your working directory. Choose Section_03 > oled_device_ml_train_input.csv and click Open
    • The panel is populated with the training data

Before we proceed, let’s understand the input data. The input data, supplied as a CSV file, defines individual OLED devices through their architectures and associated properties.

Each row in the input file is an OLED device with different specifications. Device architectures are defined by layer types, thicknesses, and material composition (SMILES, % composition). The input data is organized into four primary column types:

  1. SMILES Columns:

These columns contain information about the layer type, the layer index and SMILES representation of the materials and the SMILES index. The column headers follow the format: {Layer_type}_{Layer_index}_SMILES_{SMILES_index}.

  • Layer_type specifies the layer's function (e.g., Emitter, HIL, etc).
  • Layer_index indicates the layer's sequential order.
  • SMILES_index denotes the index of each component within that layer.

Multiple materials within a single layer are represented by sequential SMILES_index values (e.g., Emitter_0_SMILES_0, Emitter_0_SMILES_1). The emissive layer has a distinct structure compared to other layers. It is composed of two components: the "Host" material and the "Emitter" dopant material. The same indexing rules apply as with other layers, organizing the "Host" and "Emitter" information into separate columns. An example structure is as follows:

  1. Composition Columns:

The composition of each layer is specified in the composition columns. The column headers follow the format: {Layer_type}_{Layer_index}_comp_{Component_index}

In the above example, HIL has two components with 50% composition of each component. Composition columns should be explicitly defined when a layer comprises multiple materials. In the absence of such specification, a uniform composition of 100% is presumed. For the Emitter, we employ a unique composition specification that follows a different format: Emitter_{Layer_index}_wt%_{Component_index}

The associated Host_{Layer_index} column does NOT require any wt% column

  1. Thickness Columns:

These columns specify the thickness of each individual layer. The column headers follow the format: {Layertzype}_{Layer_index}_thickness. An example structure is as follows:

For the emissive layer, the host material's thickness and the emitter material's composition are combined to define the layer.

  1. Property Columns:

These columns contain the experimentally measured properties of each OLED device, such as EQE and λmax.

The provided input file incorporates supplementary columns that specify the literature origin of the experimental data used.

Figure 3-2. Viewing device information.

The device information can be visualized in the panel. Layer names are assigned according to naming conventions mentioned above. The index "n" signifies the "nth" layer of a specific type within the device architecture.

A stacked layer visualization is generated, with layer colors assigned based on the layer names. The device representation includes numerical values for overall device thickness, total number of layers, and the number of material components within the device. The properties of each device are listed in the table.

Figure 3-3. Viewing the architecture of a device.

  1. Click on the arrow () for one device
    • The device expands to show more information. The composition and SMILES of each layer are shown with color coded bars. Recollapse the dropdown by clicking the arrow again.

Figure 3-4. Editing options in the panel.

  1. To edit the device configuration before training, click the eye () button. In this mode, you can edit the composition.
    • Clicking it again saves any modifications to the loaded data. Please be aware that changes made here do not alter the original input CSV.
    • When the composition dropdown is open at the same time, you can click on any of the SMILES labels to see or edit the structure in the 2D Sketcher.

Figure 3-5. Running the job.

  1. Go to the Build tab
  2. Choose Lambda(EL)[nm]_max as the Target property
  3. Change the Job name to oled_device_ml_train
  4. Adjust the job settings () as needed
    • The job can be run on a single CPU and will be completed in approximately 15 minutes
  5. If you would like to run the job yourself, click Run. Otherwise, proceed with the provided results from the zip file.

Note: When the job is complete, no new entry group is added to the entry lista simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion.

Figure 3-6. Loading the ML model.

Upon completion of the calculation, click OK in the popup window. The ML Model Manager panel opens automatically and the trained model appears in the list.

If you want to load the provided model, switch to the Performance tab and click Load Model to open the ML Model Manager panel. Click Add Model, browse to Section_03 > oled_device_ml_train > oled_device_ml_train.mlform, and click Open.

  1. Select the oled_device_ml_train entry and click OK
    • If you get a warning saying that the uploaded ML model was trained on an older version, feel free to click OK and proceed. If you wish to retrain the model, you can use the ML Model Manager panel.

Figure 3-7. Visualizing the results.

The plot contains predicted versus actual values from the train and test sets, along with the corresponding distributions of the data sets for both predicted and actual values. The table below shows corresponding R2 and RMSE values for both data sets.

In this case, the model generalized well on the test set, even though both the training and test sets were quite small. This results in lower correlation (R2) and higher errors (RMSE). We will make more predictions with the model and compare them with experimental values to better understand its performance. Model stacking could also be used to improve performance in this case (see Section 5).

Figure 3-8. Loading the prediction dataset.

  1. Go to the Predict tab
  2. Make sure the correct model (oled_device_ml_train) is selected from the dropdown at the top
  3. With the selection set to Prediction Input, click Load
  4. Navigate to the provided files and choose Section_03 > oled_device_ml_predict_input.csv file and click Open
    • The tab is populated with the data we are going to use for predictions

Figure 3-9. Running the job.

  1. Change the Job name to oled_device_ml_predict
  2. Adjust the job settings () as needed
    • The job can be run on a single CPU and will be completed in approximately 2 minutes
  3. If you would like to run the job yourself, click Run. Otherwise, proceed with the provided results from the zip file.

Note: When the job is complete, no new entry group is added to the entry lista simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion. A window opens asking to load the prediction results. Click OK.

Figure 3-10. Loading the prediction output.

  1. If you do not want to run the job yourself and want to load the provided results, click on Prediction Input, and change the option to Prediction Output.
  2. Click Load, navigate to Section_03 > oled_device_ml_predict > oled_device_ml_predict_predict.csv, and click Open

Figure 3-11. Comparing the prediction output.

The table displays the prediction results – prediction uncertainties in the first column, predicted values in the second column – and the experimental values available from the imported data set in column 6.

We can see close alignment between the predicted and experimental values with errors of a few nm only.

4. Using Pre-trained Models to Predict Properties of OLED Devices

In this section, we will use a pre-trained model available within the OLED Device Machine Learning panel to predict the λmax of OLED devices, instead of training a new one.

Figure 4-1. Loading a pre-trained model.

  1. Use the reset button () to reset the panel
  2. Remain in the Predict tab
  3. From the drop-down option, choose Electroluminescence Maximum Peak Position (nm)
    • A pre-trained ML model is loaded into the panel. This model is trained on a larger dataset of approximately 2000 devices. Feel free to explore the results in the Performance tab.
    • A question could appear asking “This ML model was trained on an older version than the current one. Re-training the model is strongly recommended. Proceed anyway?” Click OK.

Figure 4-2. Loading the predication data set and running the job.

  1. Choose Prediction Input from the dropdown options
  2. Click Load
  3. We are going to use the same data we used in the previous section. Navigate to the provided files and choose Section_03 > oled_device_ml_predict_input.csv file and click Open.
  4. Change the Job name to oled_device_ml_predict_pretrained
  5. Adjust the job settings () as needed
    • The job can be run on a single CPU and will be completed in approximately 2 minutes
  6. If you would like to run the job yourself, click Run. Otherwise, proceed with the provided results from the zip file.

Note: When the job is complete, no new entry group is added to the entry lista simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion. A window opens asking to load the prediction results. Click OK.

Figure 4-3. Loading the prediction output.

  1. If you do not want to run the job yourself and want to load the provided results, click on Prediction Input, and change the option to Prediction Output.
  2. Click Load, navigate to Section_04 > oled_device_ml_predict_pretained > oled_device_ml_predict_pretrained_predict.csv, and click Open

Figure 4-4. Comparing the results.

The table displays the prediction results – prediction uncertainties in the first column, predicted values in the second column – and the experimental values available from the imported data set in column 6.

We can see close alignment between the predicted and experimental values.

5. Training a Stacked ML Model to Predict Properties of OLED Devices

In this section, we will use the pre-trained models available in the OLED Device Machine Learning panel to train a new model for the prediction of λmax of OLED devices. While Section 3 focused on utilizing features derived from the chemical structures (SMILES), composition, and thickness of OLED device layers, here we will now enhance our input space. Specifically, we will incorporate predicted electronic properties from DeepAutoQSAR alongside device-level performance metrics from existing OLED device ML models. This approach, known as model stacking, can be especially useful when the training data set is small. By including these “QM-like” electronic features and performance-based features with our original structural features, we expect significant gains in model predictive performance.

Figure 5-1. Loading the training data.

  1. Use the reset button () to reset the panel
  2. Go to the Training Data tab
  3. To load the training data click Load CSV and navigate to the provided files, presumably in your working directory. Choose Section_03 > oled_device_ml_train_input.csv and click Open (same data as in Section 3)
    • The panel is populated with the training data

Figure 5-2. Settings in the build tab.

  1. Go to the Build tab
  2. Choose Lambda(EL)[nm]_max as the Target property
  3. Check the option to use Pretrained Models and from the dropdown, choose Scaled HOMO and Scaled LUMO
  4. Check the option to use a Custom OLED Model and click Browse

Figure 5-3. Loading the custom OLED ML model.

  1. Click Add Model, browse to Section_05 > 26-2_oled_device_ml_elmax_pretrained.mlform, and click Open.
  2. Select the 26-2_oled_device_ml_elmax_pretrained.mlform entry and click OK
    • If you get a warning saying that the uploaded ML model was trained on an older version, feel free to click OK and proceed. If you wish to retrain the model, you can use the ML Model Manager panel.

Figure 5-4. Running the job.

  1. Change the Job name to oled_device_ml_train_stacked
  2. Adjust the job settings () as needed
    • The job can be run on a single CPU and will be completed in approximately 15 minutes
  3. If you would like to run the job yourself, click Run. Otherwise, proceed with the provided results from the zip file.

Note: When the job is complete, no new entry group is added to the entry lista simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion.

The pre-trained models available in the panel are from the Machine Learning Property Prediction panel. Additionally, you can use your own trained DeepAutoQSAR or OLED models via the custom model options available in the panel.

Figure 5-5. Loading the trained ML model.

  1. Upon completion of calculations, click OK in the popup window. The ML Model Manager panel opens automatically and the trained model appears in the list.

If you want to load the provided model, switch to the Performance tab and click Load Model to open the ML Model Manager panel. Click Add Model, browse to Section_05 > oled_device_ml_train_stacked > oled_device_ml_train_stacked.mlform, and click Open.

  1. Select the oled_device_ml_train_stacked entry and click OK
    • If you get a warning saying that the uploaded ML model was trained on an older version, feel free to click OK and proceed. If you wish to retrain the model, you can use the ML Model Manager panel.

Figure 5-6. Visualizing the results.

The plot contains predicted versus actual values from the train and test sets, along with the corresponding distributions of the data sets for both predicted and actual values. The table below shows corresponding R2 and RMSE values for both data sets.

The stacked model shown here shows higher correlation and lower errors and thus performs better compared to the model without the stacking in Section 3.

6. Conclusion and References

This tutorial covered the training and applying machine learning models to predict OLED device properties using λmax as an example. First, we trained a model using only structural properties of OLED devices. Then we made predictions on a distinct dataset using both the newly trained model and a pre-trained model. Next, we used a stacking approach to train a model that included electronic properties from DeepAutoQSAR and performance metrics from OLED ML models in addition to structural properties of OLED devices. This led to improved model performance.

For further learning:

For introductory content, focused on navigating the Schrödinger Materials Science interface, an Introduction to Materials Science Maestro tutorial is available. Please visit the materials science training website for access to 100+ tutorials. For scientific inquiries or technical troubleshooting, submit a ticket to our Technical Support Scientists at help@schrodinger.com.

For self-paced, asynchronous, online courses in Materials Science modeling, including access to Schrödinger software, please visit the Schrödinger Online Learning portal on our website.

For some related practice, proceed to explore other relevant tutorials:

For further reading:

6. Glossary of Terms

Entry List - a simplified view of the Project Table that allows you to perform basic operations such as selection and inclusion

Included - the entry is represented in the Workspace, the circle in the In column is blue

Project Table - displays the contents of a project and is also an interface for performing operations on selected entries, viewing properties, and organizing structures and data

Recent actions - This is a list of your recent actions, which you can use to reopen a panel, displayed below the Browse row. (Right-click to delete.)

Scratch Project - a temporary project in which work is not saved, closing a scratch project removes all current work and begins a new scratch project

Selected - (1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entry List (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries

Working Directory - the location where files are saved

Workspace - the 3D display area in the center of the main window, where molecular structures are displayed