Source code for pwspy.utility.reflection.multilayerReflectanceEngine

# Copyright 2018-2020 Nick Anthony, Backman Biophotonics Lab, Northwestern University
#
# This file is part of PWSpy.
#
# PWSpy is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# PWSpy is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with PWSpy.  If not, see <https://www.gnu.org/licenses/>.

"""
===============================================================================================
Multi-layer Reflectance Engine (:mod:`pwspy.utility.reflectance.multilayerReflectanceEngine`)
===============================================================================================

Using the wave transfer matrix formalism from chapter 7 of Saleh and Teich
Fundamentals of Photonics, this script calculates the reflectance of a
multilayer dielectric.
http://www.phys.ubbcluj.ro/~emil.vinteler/nanofotonica/TemeControl_FCMD014_Vinteler.pdf
https://en.wikipedia.org/wiki/Transfer-matrix_method_(optics)

m is the final transfer matrix. It should be made by multiplying the matrices
representing each element of the system. If the transmitted light is considered
to be propagating from left to right then the matrices should be in multiplied
in reverse, from right to left.


Classes
----------
.. autosummary::
   :toctree: generated/

   Polarization
   Layer
   Stack
   NonPolarizedStack

"""
__all__ = ['Polarization', 'Layer', 'Stack', 'NonPolarizedStack']

import typing
from enum import Enum, auto

import matplotlib.pyplot as plt
from cycler import cycler
from numbers import Number
from typing import Union, Optional, List
from pwspy.utility.reflection import Material
import pandas as pd
import numpy as np
import matplotlib as mpl


[docs]class Polarization(Enum): """An enumeration of the possible polarization types.""" TE = auto() # Transverse Electric Field TM = auto() # Transverse Magnetic Field
[docs]class Layer: """This represents a layer with a thickness and an index of refraction. Note: This whole system only supports lossless media, we only use the real part of the index of refraction. Args: mat: This can either be a number or series of numbers representing the refractive index at different wavelengths or it can be a `Material` in which case the refractive index will be automatically calculated. d: The thickness of the layer. The units that thicknesses and wavelengths are specified in must match. name: An optional name which will be dislayed if the layer is plotted """ def __init__(self, mat: Union[Number, pd.Series, Material], d: float, name: typing.Optional[str] = None): if not isinstance(mat, (Number, pd.Series, Material)): raise TypeError(f"Type {type(mat)} is not supported") self.mat = mat self.d = d if name is None and isinstance(mat, Material): self.name = mat.name else: self.name = name
[docs] def getRefractiveIndex(self, wavelengths: np.ndarray) -> pd.Series: """Get the refractive index of the layer. Args: wavelengths: The wavelengths to calculate the refractive index at. Returns: The refractive index. """ from .reflectanceHelper import getRefractiveIndex if isinstance(self.mat, Material): n = getRefractiveIndex(self.mat, wavelengths=wavelengths) n = pd.Series(np.real(n), index=n.index) return n elif isinstance(self.mat, Number): return pd.Series(np.array([self.mat]*len(wavelengths)), index=wavelengths) elif isinstance(self.mat, pd.Series): return self.mat
class StackBase: def __init__(self, wavelengths: Union[Number, np.ndarray], elements: Optional[List[Layer]] = None): assert len(wavelengths.shape) == 1 self.wavelengths = wavelengths if elements is None: self.layers = [] else: self.layers = elements def addLayer(self, element: Layer): """Add a new `Layer` to the `Stack` Args: element: A new layer to add. """ self.layers.append(element)
[docs]class NonPolarizedStack(StackBase): """Represents a stack of 1d homogenous films. Reflectance can only be calculated at 0 incidence angle in which case polarization is irrelevant. This class does not do anything that can't be done with `Stack`. Indices of refraction must be real (no absorption). Args: wavelengths: The wavelengths that calculation should operate over. elements: The initial layers to add to the stack. """ def _generateMatrix(self) -> np.ndarray: """First and last items just have propagation matrices.""" matrices = [] lastItem: Layer = None for el in self.layers: if lastItem is not None: matrices.append(self.interfaceMatrix(lastItem.getRefractiveIndex(self.wavelengths), el.getRefractiveIndex(self.wavelengths))) matrices.append(self.propagationMatrix(el.getRefractiveIndex(self.wavelengths), el.d)) lastItem = el matrices.reverse() previousMat = None for matrix in matrices: if previousMat is not None: previousMat = previousMat @ matrix # Matrix multiplication else: previousMat = matrix return previousMat
[docs] def plot(self): """Open a Matplotlib plot of the stack.""" cycle = cycler('color', ['r', 'g', 'b', 'y', 'c', 'm']) fig, ax = plt.subplots() ax.set_prop_cycle(cycle) startCoord = 0 for el, col in zip(self.layers, cycle()): r = plt.Rectangle((startCoord, 0), el.d, 1, color=col['color']) t = plt.Text(r.xy[0]+r.get_width()/2, .5, f"{el.name}: {np.mean(el.getRefractiveIndex(self.wavelengths))}") startCoord = startCoord + el.d ax.add_patch(r) ax.add_artist(t) ax.set_xlim([0, startCoord]) plt.show()
[docs] @staticmethod def interfaceMatrix(n1: pd.Series, n2: pd.Series) -> np.ndarray: """Generate a matrix representing the interface between two dielectrics with indices n1 on the left and n2 on the right. Actually the order of terms does not appear to matter. This does not account for polarization or incidence angles other than 0 degrees. Args: n1: The refractive indices on one side of the reflective interface n2: The refractive indices on the other side of the reflective interface Returns: A transfer matrix for the reflective interface """ assert len(n1) == len(n2) assert np.all(n1.index == n2.index) n1 = np.array(n1) n2 = np.array(n2) matrix = np.array([[n2 + n1, n2 - n1], [n2 - n1, n2 + n1]]) matrix = np.transpose(matrix, axes=(2, 0, 1)) matrix = (1 / (2 * n2))[:, None, None] * matrix assert matrix.shape == (len(n1),) + (2, 2) return matrix
[docs] @staticmethod def propagationMatrix(n: pd.Series, d: float) -> np.ndarray: """Returns a matrix representing the propagation of light. n should be a pandas Series where the values are complex refractive index and the index fo the Series is the associated wavelengths. with wavelength. for a distance of `d`. d and the wavelengths must use the same units. Args: n: The refractive indices of the material d: The distance of propagation. Returns: A transfer matrix for propagation through a material. """ wavelengths = n.index phi = np.array(2 * np.pi * d * n / wavelengths) zeroArray = 0 * phi # Without this our matrix will not shape properly matrix = np.array([[np.exp(-1j * phi), zeroArray], [zeroArray, np.exp(1j * phi)]]) matrix = np.transpose(matrix, axes=(2, 0, 1)) assert matrix.shape == (len(n),) + (2, 2) return matrix
[docs] def calculateReflectance(self) -> np.ndarray: """Calculate the reflectance for this `Stack`. Returns: The reflectance. """ m = self._generateMatrix() assert m.shape == (len(self.wavelengths),) + (2, 2) scatterMatrix = np.array([ # A 2x2 scattering matrix. https://en.wikipedia.org/wiki/S-matrix [m[:, 0, 0] * m[:, 1, 1] - m[:, 0, 1] * m[:, 1, 0], m[:, 0, 1]], [-m[:, 1, 0], np.ones((m.shape[0],))]]) scatterMatrix = np.transpose(scatterMatrix, axes=(2, 0, 1)) scatterMatrix = (1 / m[:, 1, 1])[:, None, None] * scatterMatrix R = scatterMatrix[:, 1, 0] * np.conjugate(scatterMatrix[:, 1, 0]) # The reflectance of the stack. This is a real number. Equivalent to np.absolute(scatterMatrix[1, 0]) ** 2 assert np.max(np.abs(np.imag(R))) == 0 return np.real(R)
[docs]class Stack(StackBase): """Represents a stack of 1d homogenous films. Reflectance for the two polarizations can be calculated for a range of numerical apertures (angles). Indices of refraction must be real (no absorption). Args: wavelengths: The wavelengths that calculation should operate over. elements: The initial layers to add to the stack. """
[docs] @staticmethod def interfaceMatrix(n1: pd.Series, n2: pd.Series, polarization: Polarization, NAs: np.ndarray) -> np.ndarray: """Returns a matrix representing a dieletric interface. n1 and n2 should be a pandas Series where the values are complex refractive index and the index of the Series is the associated wavelengths. Args: n1: The refractive indices on one side of the reflective interface n2: The refractive indices on the other side of the reflective interface polarization: The polarization that should be used for the calculation NAs: An array of the numerical aperture values. :todo: More details would be good Returns: A transfer matrix for the reflective interface """ assert len(n1) == len(n2) assert np.all(n1.index == n2.index) n1 = n1.values[:, None] n2 = n2.values[:, None] NAs = NAs[None, :] theta1 = np.arcsin(NAs / n1) theta2 = np.arcsin(NAs / n2) if polarization is Polarization.TE: a21 = 1 N1 = n1 * np.cos(theta1) N2 = n2 * np.cos(theta2) else: # Polarization.TM a21 = np.cos(theta2) / np.cos(theta1) N1 = n1 / np.cos(theta1) N2 = n2 / np.cos(theta2) matrix = np.array([[N2 + N1, N2 - N1], [N2 - N1, N2 + N1]]) matrix = np.transpose(matrix, axes=(2, 3, 0, 1)) matrix = (1 / (2 * a21 * N2))[:, :, None, None] * matrix assert matrix.shape == (n1.size, NAs.size) + (2, 2) return matrix
[docs] @staticmethod def propagationMatrix(n: pd.Series, d: float, NAs: np.ndarray) -> np.ndarray: """Returns a matrix representing the propagation of light for a distance of `d`. d and the wavelengths must use the same units. Args: n: The refractive indices of the material d: The distance of propagation. Returns: A transfer matrix for propagation through a material. """ wavelengths = np.array(n.index)[:, None] n = n.values[:, None] NAs = NAs[None, :] theta = np.arcsin(NAs / n) phi = n * d * np.cos(theta) * 2 * np.pi / wavelengths # phi = np.array(2 * np.pi * d * n / wavelengths) zeroArray = np.zeros(phi.shape) # Without this our matrix will not shape properly matrix = np.array([[np.exp(-1j * phi), zeroArray], [zeroArray, np.exp(1j * phi)]]) matrix = np.transpose(matrix, axes=(2, 3, 0, 1)) # Move the angle and wavelength dimensions up front. leave the matrix dimensions as the last two. assert matrix.shape == (n.size, NAs.size) + (2, 2) return matrix
def _generateMatrix(self, polarization: Polarization, NAs: np.ndarray) -> np.ndarray: """First and last items just have propagation matrices. """ matrices = [] lastItem: Layer = None for el in self.layers: if lastItem is not None: matrices.append(self.interfaceMatrix(lastItem.getRefractiveIndex(self.wavelengths), el.getRefractiveIndex(self.wavelengths), polarization, NAs)) matrices.append(self.propagationMatrix(el.getRefractiveIndex(self.wavelengths), el.d, NAs)) lastItem = el matrices.reverse() previousMat = None for matrix in matrices: if previousMat is not None: previousMat = previousMat @ matrix # Matrix multiplication else: previousMat = matrix return previousMat
[docs] def calculateReflectance(self, NAs: np.ndarray): """Given an array of numerical apertures this function returns the reflectance as a dictionary of 2d arrays. There is one 2d array for each of the two polarizations. the dimensions of the array is (wavelengths x NAs). The total reflectance can be calculated as the average reflectance of the two polarizations. Other ellipsometric parameters can also be calculated. Args: NAs: The numerical apertures to calculate reflectance at. Returns: A dictionary containing a reflectance array for each of the two polarizations. The polarization is the key to the dictionary. Each reflectance is a MxN array where M is the number of wavelengths and N is the number of NAs passed to this function. """ out = {} for polarization in Polarization: m = self._generateMatrix(polarization, NAs) scatterMatrix = np.array([ # A 2x2 scattering matrix. https://en.wikipedia.org/wiki/S-matrix [m[:, :, 0, 0] * m[:, :, 1, 1] - m[:, :, 0, 1] * m[:, :, 1, 0], m[:, :, 0, 1]], [-m[:, :, 1, 0], np.ones((m.shape[0], m.shape[1]))]]) scatterMatrix = np.transpose(scatterMatrix, axes=(2, 3, 0, 1)) scatterMatrix = (1 / m[:, :, 1, 1])[:, :, None, None] * scatterMatrix R = scatterMatrix[:, :, 1, 0] * np.conjugate(scatterMatrix[:, :, 1, 0]) # The reflectance of the stack. This is a real number. Equivalent to np.absolute(scatterMatrix[1, 0]) ** 2 out[polarization] = R.real return out
[docs] def circularIntegration(self, NAs: np.ndarray) -> pd.Series: """Given an array of NumericalApertures (usually from 0 to NAMax.) This function integrates the reflectance over a disc of Numerical Apertures (Just like in a microscope the Aperture plane is a disc shape, with higher NA being further from the center.) Ultimately the result of this integration should match the reflectance measured with the same NA. Args: NAs: The numerical apertures to calculate reflectance at. Returns: A pandas `Series` with wavelengths as the index and reflectance as the value. """ from scipy.integrate import trapz d = self.calculateReflectance(NAs) r = (d[Polarization.TE] + d[Polarization.TM]) / 2 # Reflectance averaged over polarization. r = r * 2 * np.pi * NAs #The aperture plane is a disk, higher NAs have larger circumference disks contributing to them. hence the 2pi*na factor. inte = trapz(r, NAs, axis=1) int2 = trapz(2 * np.pi * NAs, NAs) #This is used for normalization. basically accounting for the fact that na is proportional to radius in aperture plane, not equal. return pd.Series(inte / int2, index=self.wavelengths)
[docs] def plot(self, NAs: np.ndarray, polarization: Polarization = None): """Plot various graphs of reflectance vs NA. NAs should be an array of Numerical apertures to have the reflectance calculated for. `polarization` can be specified to view the reflectance of only one polarization.""" d = self.calculateReflectance(nas) rTM = d[Polarization.TM] rTE = d[Polarization.TE] if polarization is None: eccentricity = np.sqrt(1 - rTM / rTE) # rTM == a**2, rTE == b**2 r = (rTM + rTE) / 2 elif polarization == Polarization.TE: r = rTE eccentricity = None elif polarization == Polarization.TM: r = rTM eccentricity = None else: raise TypeError(f"`polarization` must be Polarization, not {type(polarization)}.") fig2, ax2 = plt.subplots() fig2.suptitle("R") ax2.set_ylabel("wavelength") ax2.set_xlabel("NA") if eccentricity is not None: fig, ax = plt.subplots() fig.suptitle("Eccentricity") ax.set_ylabel("wavelength") ax.set_xlabel("Angle") im = ax.imshow(eccentricity, extent=[NAs[0], NAs[-1], self.wavelengths[-1], self.wavelengths[0]], interpolation=None, aspect='auto') plt.colorbar(im, ax=ax) fig.show() im = ax2.imshow(r, extent=[NAs[0], NAs[-1], self.wavelengths[-1], self.wavelengths[0]], interpolation=None, aspect='auto') plt.colorbar(im, ax=ax2) fig2.show() fig3, ax3 = plt.subplots() colormap = plt.cm.gist_rainbow colors = [colormap(i) for i in np.linspace(0, 0.99, r.shape[1])] norm = mpl.colors.Normalize(vmin=NAs[0], vmax=NAs[-1]) sm = plt.cm.ScalarMappable(cmap=colormap, norm=norm) for i in range(r.shape[1]): ax3.plot(self.wavelengths, r[:, i], color=colors[i], label=NAs[i]) plt.colorbar(sm, ax=ax3) fig3.show() fig4, ax4 = plt.subplots() ax4.set_xlabel("NA") ax4.plot(NAs, rTM.mean(axis=0), label='TM') ax4.plot(NAs, rTE.mean(axis=0), label='TE') ax4.plot(NAs, r.mean(axis=0), label='R') ax4.legend() fig4.show()
if __name__ == '__main__': num = 40 wv = np.linspace(500, 700, num=100) s = Stack(wv) s.addLayer(Layer(Material.Glass, 100000)) for i in range(20): s.addLayer(Layer(1.2, 650/1.2/4)) s.addLayer(Layer(1.35, 650/1.35/4)) # s.addLayer(Layer(Material.ITO, 2000)) s.addLayer(Layer(Material.Air, 10000)) nas = np.linspace(0, .2, num=1000) s.plot(nas) out = s.circularIntegration(nas) fig, ax = plt.subplots() ax.plot(s.wavelengths, out) fig.show() pass