Accuracy 2D vs 3D in comparison to experiments

[Paul18811]

Hello everyone,

Currently, I am simulating the response of a rib 50/50 MMI splitter of dimensions ~ 35 x 5 x 0.22 microns.
In order to perform a 2D simulations, an effective index method is needed. The effective index method for rib waveguides that is described in most papers does not include an optimization step to match the beat length (coupling length) of the 3D and 2D cases. However, this beat length optimization by tuning the background effective index, is supposed to be necessary for an MMI splitter.

The experimental results of this MMI showed a total loss between the input and output between 3.13 and 3.43 dB. The 2D results showed a total loss of 3.28 and 3.5 dB for the effective index method without and with a beat length optimization. These results were obtained with a resolution of 100.

Even though the beat length optimization should result in more accurate results, it seems like its simulated response matches the experiments less than the one without beat length optimization. Despite the fact that this would be a surprising result, it is not totally sure that the beat length result matches the 'real' system in a lesser amount. The experimentally tested design might have been under/over edged and therefore have a different result from the simulations.

In order to check the accuracy of the effective index methods, also a 3D simulation is done. This result was totally off all previous values as it had a total loss of 4.49 dB. This is far from both the experimental results as well as the 2D calculations. The resolution of this 3D simulation was 70.

Does someone know why this 3D simulation does not seem to have a good accuracy even though it is supposed to be more accurate?

Thanks in advance!