Thank you for the question. From looking at your simulation file it seems to be set up properly. I expect that the experimental loss is higher due to sources of loss not included in the simulation.
In this simulation we would expect two main sources of loss: radiation loss from the taper section, and loss due to mismatch between the modes of the two chips. For a long (>50 um) SOI taper from a width of 500 nm to 2000 nm I would expect the majority of the loss would come from the mode mismatch. We can use the FDE solver to calculate the power coupling between the two modes:
We can see that the power coupling at the interface between the modes is about 67%, so this is likely where most of the loss of the simulation is coming from. Your results, therefore, seem reasonable to me.
However you should keep in mind that these simulations do not include various sources of loss, such as scattering from sidewall roughness. Also, you have used metal boundary conditions so there will be no losses into the substrate in this simulation (to see a discussion of substrate losses in EME simulations, see our Edge Coupler example).
In particular for this device, I would expect misalignment or gaps between the two chips to be a major source of experimental loss that is not included in the simulation. For example, if the chips are misaligned by 400 nm in the z direction, the power coupling drops down to 8%:
I also attempted to include the losses from a gap between the two chips in your file (I have attached the updated file here: simple_taper_lumerical_new.lms
(9.5 MB) ). Here are the results:
Here we can observe a Fabry-Perot effect due to reflections between the air gap and the chips. We can see that a gap of 3 microns reduces the transmission by around half.
I would expect the extra experimental losses you are observing could be due to these effects, though there could be many others. I would recommend including the substrate to obtain a more realistic loss value. However, you should almost always expect the experimental loss to be higher than the loss in the simulation due to the fact that it is difficult to include all sources of loss in the simulation.
I hope this helps. Let me know if you have any questions.