Extremely large mode loss

[pascualmunoz]

Describe the bug
Calculating the modes of a silicon or silicon nitride core, and silica cladded, waveguides (typical single mode dimensions in the C-band), provides k_eff values of order 1e-5 for fundamental mode (mostly confined in the core). In the case e.g. of Si the k value is 1.34e-13.

For information, the same cross-section but with loss-less materials provides k_eff = 0.0.

Expected behavior
A much lower value of k_eff, similar to material loss.

[pascualmunoz]

# standard python imports
import numpy as np
import matplotlib.pylab as plt

# tidy3D import
import tidy3d as td
from tidy3d.constants import C_0
import tidy3d.web as web
from tidy3d.plugins.mode import ModeSolver

# central frequency
wvl_um = 1.45
freq0 = C_0 / wvl_um

mat = td.material_library['cSi']['Green2008']
eps = mat.eps_model(freq0)

eps

n = np.sqrt( (np.abs(eps)+np.real(eps)) / 2.0 )

k = np.sqrt( (np.abs(eps)-np.real(eps)) / 2.0 )

print(n,k)

print(np.sqrt(eps))

print(eps)

3.478220107067083 0.0008218546723818859
(3.478220107067083+0.000821854672654681j)
(12.098014437760648+0.005717182895029094j)

while in refractiveindex.info we have:

https://refractiveindex.info/?shelf=main&book=Si&page=Green-2008

wvl = 1.45 µm
n = 3.4850
k = 1.3846e-13
ϵ1 = 12.145
ϵ2 = 9.6507e-13

[weiliangjin2021]

@pascualmunoz it'll be very helpful if we can have your geometry and mode solver parameters to reproduce this issue.

[momchil-flex]

@weiliangjin2021 actually I think the issue is simply in our Green2008 material fit. What Pascal is showing is that it has much higher loss at that wavelength than the reference.

[pascualmunoz]

Yes exactly. And the same happens with silica models. Perhaps worth to revise the material library.

[momchil-flex]

@weiliangjin2021 in 2.7, apart from still creating the two separate Green2008 models, let's plan to switch the default materials to "Palik_Lossless" for both cSi and SiO2. This should be done with a warning that the default has changed, and the warning can be suppressed if the user explicitly puts material_library['cSi']['Palik_Lossless'] instead of material_library['cSi'].

FYI @lucas-flexcompute @tomflexcompute @tylerflex

[pascualmunoz]

Thanks, that sounds like a temporary solution but, will the library material models with loss be checked?

[momchil-flex]

Yes, the other part of the solution would be to split Green2008 into a lossless or nearly lossless model for the range 1.2-1.45, and the current 5-pole lossy model which is more valid for energies higher than the band gap of silicon. It's just close to impossible to construct a model that has both high loss at some freqeuncies and very low loss at others.