Acceleration

Bumblebee provides acceleration methods to enhance sampling effectiveness and reduce computation time.

Module Selection

A rate booster is available to promote anisotropic charge hopping along the electrode current. This increases the rate at which the system equilibrates and enhances the sampling of rare events that are typically of interest to OLED performance.

The relative rates of the parallel and perpendicular transfer processes can be modified. Specifying a priority factor of 1.2 for the carrier transport will increase the parallel current sampling frequency by 20%.

Note

The acceleration module adds an anisotropic bias to the distribution of process frequencies. This bias can alter the kMC trajectories, influencing the device statistics.

Similar to the determination of the electrode transfer barrier, it is necessary to verify that the priority factor does not alter the device behavior.

A single disorder instance can be evaluated in order to determine the correct acceleration settings. Once determined, the parameter set can be updated. Subsequent simulations can be performed at significantly reduced cost.

Because the acceleration module does not alter the actual opto-electronic process parameters, the same acceleration settings can be used at various currents. Care should be taken, however, that the priority factor was determine at a current where opto-electronic processes were energetically accessible.

For example: when annihilation processes are included in the simulation, you should confirm that sampling of these processes occurred in the disorder instance that was used to determine the acceleration settings.

The necessary process event counts can be found in the Profiles section of the Sweep Report.

Configuring Output

Writing output to the file system can become a bottleneck for the kMC simulation, as write speeds are typically slow compared to the evaluation of processes in Bumblebee.

Faster simulation times can be obtained by choosing longer output intervals. This does come with a risk that fewer checkpoints are generated to enable simulation restarts in case of job interruptions. As the average device statistics typically contain the simulation properties of interest, the reduced amount of intermediate output itself is often considered inconsequential.

Aside from reducing the reporting interval, the content of the simulation output itself can also be reduced. The Output settings in the parameter sets allow users to disable undesired statistics from the output by de-selecting the appropriate set of files. This reduced the memory volume of the output, particularly for transient data, event logs or 3D distributions.

Cost Scaling

The cost of a Bumblebee simulation scales with the volume of the device. Modeling of devices with many layers or large surface area will require longer simulation times, simply because the carriers have to traverse longer distances and sampling has to be conducted over a larger number of gridpoints.

For devices with a large surface area, one has the choice between conducting a single large-area simulation or splitting the total surface area between multiple parallel instances. This choice typically has little effect on the overall simulation cost. Parallel distribution tends to be preferred when a larger number of cores is available, as utilization of multiple CPUs allows faster access to simulation results. The required number of CPU-hours nevertheless remains mostly unaltered.

When conducting simulations on very large stacks, the time required to equilibrate the device increases, as does the required number of samples to properly characterize the device. Since most large-stack OLED devices tend to represent multi-stack junctions, it is possible to coarse-grain the simulation by splitting up the stack. Simulations can be conducted on individual stack segments. By matching the current-voltage profiles of the segments, the behavior of the multi-junction system can be reproduced. Note however, that this is only possible when interactions between the junctions are negligible. For cases where coarse-graining is not viable, utilization of the acceleration module is recommended to improve sampling efficiency.