FAQ¶

How do I determine the required number of simulation steps?¶

The required number of samples is determined by the desired accuracy of the results.

kMC estimates the properties of the OLED device through a stochastic sampling process. Increasing the number of samples results in greater accuracy.

This accuracy differs for each simulation output. Rare simulation events will be sampled less frequently, resulting in poorer statistics.

One strategy to determine the required number of simulation steps is to perform adaptive refinement. Start by running the simulation for a set number of steps, then check the output accuracy. If the results are unsatisfactory, you can add additional trajectories to the simulation in order to increase the number of samples.

How do I determine the maximum wall time for my workload manager?¶

The total simulation time varies based on the process complexity, device size and the simulation length. In general, kMC simulations can have a duration from minutes to days.

In order to work well with typical schedulers, it is recommended to use parallel trajectories to distribute the workload for more expensive jobs.

Note that even if a simulation exceeds the allotted wall time, the web interface can still interpret the data obtained thus-far from the intermediate simulation output. As such, there are 2 scheduling strategies:

• You can use a generous wall time along with a short simulation time to split up the sampling process between a large number of instances. When evaluating the simulation output, it is possible to commission additional trajectories to increase the accuracy of the statistical estimates, if deemed necessary

• You can specify a long simulation time (along with a reasonable output interval). The duration of the simulation will then be determined by the wall time. Most simulations will not reach their end, but instead be terminated by the scheduler. Termination will cause a small amount of output to be lost at the end of a simulation. This strategy is only recommended on systems with slow convergence behavior

Can I recover the results from a killed job?¶

Bumblebee updates the output files as the simulation progresses. The update frequency can be changed in the parameter set.

The web interface and analysis modules are able to interpret this intermediate data, even when a job was killed by the server.

It is possible to continue a terminated run from the last recorded state. Simply re-run the submission script on the server to resume the simulation.

How do I perform two-dimensional parameter screenings?¶

The parameter screening option in the web interface is designed for single-variable screenings.

Even though multiple screenings can be added to a single simulation, the visualization routines are not able to identify this second parameter, resulting in overlapping data in the graphs.

When performing multi-variate device optimization, it is recommended to use the Python API to set custom sample points. The simulation output can then be collected for processing using e.g. the PLAMS visualization tool set.

How do I include charge-generation layers in the stack?¶

Charge-generation layers (CGL) are available starting from the 2025 release of Bumblebee. Consult the CGL tutorial for more details on this material template.

In older versions of Bumblebee, it is possible to approximate the CGL as an idealized transport layer:

• Use the Transport template for the CGL material

• In the stack editor, polaron generation can be enabled by including photoabsorption processes inside the CGL layer. The polaron density can be controlled by the fluence

• For bilayer CGL, localization of the charge generation at the interface can be approximated by using a thin film of 1-2 nm for the absorption region

• Use the Photovoltaic template for the parameter set to automatically enable the photoluminescence module during the simulation

How do I model a tandem stack?¶

Bumblebee allows any number of layers to be included in the OLED stack. Tandem devices can therefore be treated straightforwardly by including multiple emissive layers.

Assuming proper operation of the polaron injection layers, interactions between light-emission units (LEU) can be minimal. In this regime, it is also possible to run voltage sweep simulations for isolated LEU. Current-voltage profiles can then be aligned to approximate the tandem stack performance. This strategy is primarily recommended for the screening of LEU materials.

When modeling the full tandem stack:

• Explicit inclusion of the interface between LEU can be achieved by using a CGL layer. (See the application notes in the previous segment.)

• The interface can also be modeled implicitly. Instead of adding a layer to the stack, the inter-layer transport parameters can be modified to account for the additional resistance. See the Advanced Bumblebee tutorials for more details.