measurement of thin film thickness by electronic speckle pattern interferometry
TRANSCRIPT
Optics Communications 234 (2004) 269–276
www.elsevier.com/locate/optcom
Measurement of thin film thickness by electronicspeckle pattern interferometry
Canan Karaalio~glu, Yani Skarlatos *
Department of Physics, Bo~gazic�i University, Bebek 80815, Istanbul, Turkey
Received 2 June 2003; received in revised form 20 September 2003; accepted 6 February 2004
Abstract
The surface profile of an Al thin film and its thickness have been observed by electronic speckle pattern interfer-
ometry (ESPI). The Michelson interferometer was used as our basic interferometric system to obtain interference
fringes on a CCD camera. These interference fringes, arising from the path differences due to the surface contours of the
thin film, were analyzed with three different techniques: fast Fourier transform (FFT), phase shifting, and digital image
subtraction; and the results were compared with each other. We also derived a new formula for the image subtraction
method, which is only valid in one-dimensional calculations. An unwrapping procedure was used to obtain a contin-
uous phase in the FFT and phase shifting methods. Results on thickness measurements are presented and are found to
be in good agreement with each other.
� 2004 Elsevier B.V. All rights reserved.
1. Introduction
When an optically rough surface is illuminated
by a beam of coherent laser light and the light
scattered from the surface is observed on a plane
some distance away, then at any point P on this
plane, the electric field is the vector sum of all the
elementary fields which reach the point P after
being scattered by different portions of the entire
rough surface. Since the surface is optically rough,the total electric field and intensity around point P
change randomly with position on the observation
* Corresponding author. Fax: +90-212-287-24-66.
E-mail address: [email protected] (Y. Skarlatos).
0030-4018/$ - see front matter � 2004 Elsevier B.V. All rights reserv
doi:10.1016/j.optcom.2004.02.025
plane. These random fluctuations in intensity are
known as laser speckles. Laser speckles carry in-formation about the rough surface which produces
them and have found many applications, which
are known generally as speckle methods. One of
these methods is electronic speckle pattern inter-
ferometry (ESPI) and it is the most practical and
powerful one among the existing optical tech-
niques for deformation and vibration analysis [1].
Contour measurement by interferometry is usedwidely to determine the shape of a surface. Infor-
mation about the surface of a static object can be
obtained from interference fringes, whose contour
lines characterize the surface on which they are
formed. There are two basic methods for mea-
surement of surface shape or height profile by
ed.
270 C. Karaalio~glu, Y. Skarlatos / Optics Communications 234 (2004) 269–276
optical contouring. These are phase shifting and
Fourier transform. A third, less frequently used
method is called digital image subtraction.
The fast Fourier transform (FFT) method has
been used by many researchers, such as de Angelis
et al. [2] to measure the focal lengths of lenses andthe index of refraction of transparent liquid ma-
terials [3], Spagnolo et al. [4,5] to obtain surface
topography of ancient stone artworks or to ana-
lyze microclimate variation on artworks [6], and
Wu et al. [7] to quantify immunoreactive terminal
area apposed to nerve cells. The phase shifting
method has also been used by many researchers.
The studies of Tay et al. [10], He et al. [11], L€ucket al. [12], Quan et al. [13] and Hobson et al. [14]
are some recent examples for this method; while
Abedin et al. [1,15] have employed the digital
subtraction method.
Thin film thickness is an important parameter
in many applications. In this paper, we obtain a
surface topography for thin films by using three
different shape construction methods: FFT, phaseshifting, and digital subtraction. Previously, we
had used the FFT method to measure the thick-
ness of a thin film [16]. In this paper, we derive a
new formula for the subtraction method in one-
dimensional case. Finally, the results of those three
techniques are compared with each other. The
fringe patterns were captured by a CCD camera.
The deformed surface information is encoded intothese fringes. Contour fringes were displayed on
a monitor and then analyzed with the above
methods.
2. Principle of the methods
2.1. Digital speckle subtraction method
The incident plane wavefront is divided into
two plane wavefronts of equal intensity by a beam
splitter, BS. The plane waves fall on the two rough
surfaces D1 and D2, which produce two indepen-
dent, random speckle distributions. These speckle
distributions are imaged by the camera lens L in its
image plane. The system is essentially equivalent toa Michelson�s interferometer, where the two mir-
rors are replaced by two rough surfaces. Let
E1 ¼ e1 expði/1Þ and E2 ¼ e2 expði/2Þ be the com-
plex amplitudes of these wavefronts in the image
plane, where e1; e2 and /1;/2 correspond, respec-
tively, to randomly varying amplitudes and phases
of the image plane speckles. The total intensity fcat a given point P ðx; yÞ in the image plane must beproportional to the square of the sum of E1 and E2,
i.e.
fcðx;yÞ¼ f1ðx;yÞþ f2ðx;yÞþ2½f1ðx;yÞf2ðx;yÞ�1=2 cos/;ð1Þ
where f1ð¼ E1E�1Þ and f2ð¼ E2E�
2Þ are the intensity
distributions of the two speckled wavefronts,
/ð¼ /1 � /2Þ is their phase difference, f1ðx; yÞþf2ðx; yÞ ¼ aðx; yÞ, and 2½f1ðx; yÞf2ðx; yÞ�1=2 ¼ bðx; yÞ.Therefore, we can write the intensity equation as
fcðx; yÞ ¼ aðx; yÞ þ bðx; yÞ cos/: ð2ÞSurface D2 is now deformed by some mechani-
cal or thermal disturbance, while surface D1 is left
undisturbed (E2 can be considered as the ‘‘object’’
wavefront, and E1, the ‘‘reference’’ wavefront). In
our experiment, D2 and D1 are the Al surfaces with
and without a thin Al strip on top respectively. Let
zðx0; y0Þ be the displacement of any given point of
Surface D2 in the normal direction (known as out-of-plane motion in the literature), then the corre-
sponding phase change for this point is
D/ðzÞ ¼ 4pzk
: ð3Þ
Due to this phase change, the intensity distri-
bution at the image plane will change to a new one,f 0cðx; yÞ, given by
f 0cðx; yÞ ¼ aðx; yÞ þ bðx; yÞ cos½/þ D/ðzÞ�: ð4ÞWe have two different images with mirror–
mirror and object–mirror reflectivity surfaces in
Fig. 2. The image obtained by mirror–mirror sur-faces, i.e. fcðx; yÞ is subtracted from the image
obtained by object–mirror surfaces, i.e. f 0cðx; yÞ.
The intensity distributions fcðx; yÞ and f 0cðx; yÞ are
captured (photographically) by the digital camera,
which produces electrical signals ultimately stored
in the computer memory in the form of digital
JPEG image files . After the image files are ac-
quired and stored, image fcðx; yÞ is subtracted fromf 0cðx; yÞ in the computer and the subtracted image
C. Karaalio~glu, Y. Skarlatos / Optics Communications 234 (2004) 269–276 271
is displayed on the monitor. This subtraction is
performed pixel by pixel. If we assume that the
displayed brightness is proportional to the differ-
ence between f 0cðx; yÞ and fcðx; yÞ, then
B / ½f 0cðx; yÞ � fcðx; yÞ�; ð5Þ
B / ½f1ðx; yÞf2ðx; yÞ�1=2
�fcos½/þ D/ðzÞ� � cos/g: ð6Þ
If, for some points on D2, z is such that
D/ðzÞ ¼ 2pn, then B ¼ 0, and dark fringes occur
on the computer monitor. On the other hand, for
some other points, if z is such that
D/ðzÞ ¼ ð2nþ 1Þp, then B has maximum values
and bright fringes occur. Using Eq. (3), theseconditions mean
• z ¼ nðk=2Þ for dark fringes.
• z ¼ ðnþ 12Þðk=2Þ for bright fringes.
The dark and bright fringes on the computer
monitor will therefore map the displacements
zðx0; y 0Þ of the different parts of the surface D2.
Successive dark or bright fringes on the computer
monitor will indicate displacement differencesof ðk=2Þ of the corresponding areas of surface D2,
or about 0.32 lm if a He–Ne laser is used for the
illumination.
The presence of the factor ½f1ðx; yÞf2ðx; yÞ�1=2 in
Eq. (6) means that the dark and bright fringes on
the computer monitor are accompanied by some
random speckle ‘‘noise’’ due to the speckle distri-
butions f1ðx; yÞ and f2ðx; yÞ. The presence of this‘‘noise’’ is however unavoidable in ESPI, since it is
this speckle noise which encodes and carries the
essential phase information D/ðzÞ into f 0cðx; yÞ in
Eq. (4) [1].
We derived a new formula for this method, but
it is valid only for one dimensional calculations.
By considering an interference pattern like in
Fig. 3, if we plot the intensity values of one row ofthe surface without an Al strip on it, this plot
corresponds to a sinusoidal curve for a flat surface.
In a similar way, if we plot the intensity values of
one row of the surface with an Al strip on it, this
should still correspond to a sinusoidal curve; but
this plot should have a shifting at the edge of the
strip because of the height difference. When we
plot one row intensity distributions from deformed
and flat interference patterns, this shifting amount
gives us the desired information
/d ¼ 2pDxx
�þ n
�: ð7Þ
Dx is the shifting amount in the deformed case
plot and x is the amount of one full period in the
flat case plot. If there is no shifting, i.e. Dx ¼ 0,then /d ¼ 2p and no shifting. n is an important
integer here and we should have an idea about the
order that we are looking for. n is determined from
the order of the height. If we know the height
value approximately, we can determine the value
of n. The height value can be obtained for each
row separately and then two-dimensional surface
profile can be obtained by plotting all these heightvalues next to each other in this technique.
2.2. Phase shifting method
Three or more phase shifted images are required
to produce a wrapped phase map; this map
indicates the contour of the surface under exami-
nation. Using the unwrapping technique, a con-
tinuous phase distribution is obtained from which
the required measurement parameter can be cal-
culated [10].
The light intensity, f ðxÞ of a digital specklefringe pattern can be expressed in the form
f ðxÞ ¼ aðxÞ þ bðxÞ cosð/þ d/Þ; ð8Þwhere a; b are constants, / is a phase angle intro-
duced due to object deformation, and d/ is a
subsequent additional phase change. Four differ-
ent f ðxÞ equations are obtained corresponding tofour different d/ values. If we solve these four f ðxÞequations, the following is obtained:
tan/¼ ½ðf3� f1Þðcosd/4� cosd/2Þ�ðf4� f2Þ�ðcosd/3�1Þ�=½ðf3� f1Þðsind/4� sind/2Þ�ðf4� f2Þsind/3�; ð9Þ
where f1ðxÞ, f2ðxÞ, f3ðxÞ, and f4ðxÞ represent the
light intensities corresponding to phase angles of
/, /þ d/2, /þ d/3, and /þ d/4, respectively.
The phase angle / of a deformed object can
thus be obtained using Eq. (9). For additional
Fig. 1. Schematic of the Al thin film.
272 C. Karaalio~glu, Y. Skarlatos / Optics Communications 234 (2004) 269–276
phases of d/2 ¼ p=2, d/3 ¼ p, d/4 ¼ 3p=2, Eq. (9)would appear in its simplest form
/ ¼ arctanf4ðxÞ � f2ðxÞf1ðxÞ � f3ðxÞ
: ð10Þ
Even though we have used a 4-frame phase shift
algorithm to find phase shifting amount, the
choice is not unique. The reason for chosing this
particular algorithm is that it is very common and
easy to apply on our set-up, as there is no error
coming from phase shifting increments other than
the manufacturing error. In the literature, there
are algorithms from 3-frame to 12-frame as well asFourier transform methods of phase-shift deter-
mination [8,9].
Since arctan function is mathematically defined
from �p to p, the phase distribution is wrapped
into this range and, consequently, has disconti-
nuities with 2p phase jumps for variations larger
than 2p. These discontinuities can be corrected
easily by adding or subtracting 2p according to thephase jump involved.
The height of the surface z follows from the /0,
i.e. unwrapped phase, according to
z ¼ k4p
/0; ð11Þ
where k is the wavelength of the light source
[10–13].
The phase shifting algorithms used are less
complicated than the Fourier transform analysis
and hence greatly simplify the computational me-
dium. Errors might come from estimating thevalues of the experimental parameters [10].
2.3. FFT method
If structured light consisting of parallel stripes,
or fringes, is projected onto a surface, then the
surface acts as a phase modulator, with the
amount of modulation at any point dependingupon the height of the surface at that point.
In various optical measurements, we find a
fringe pattern of the form
f ðx; yÞ ¼ aðx; yÞ þ bðx; yÞ cos½2pu0xþ /ðx; yÞ�;ð12Þ
where the phase /ðx; yÞ contains the desired in-
formation and aðx; yÞ and bðx; yÞ represent un-
wanted irradiance variations arising from the
nonuniform light reflection or transmission by a
test object [17].The Fourier transform method for fringe pat-
tern analysis requires an added high spatial carrier
frequency. In this case, the distinct and equally
spaced sinusoidal fringes serve as the carriers. If
the spatial carrier frequency is properly selected,
the contour can be reproduced using the inverse
Fourier transform and phase unwrapping tech-
niques. The method has the advantage of usingonly one interference pattern during processing.
The resulting fringes are analyzed with automatic
phase unwrapping, and the background intensity
variations and speckle noise can be reduced [18].
This method is exhaustively explained in [16],
which is based on [18].
3. Experiments
3.1. Description of the experimental setups
We evaporated Al as a thin film onto a glass
substrate. Then, a second Al evaporation was
performed through a stainless steel mask approx-
imately 0.2 mm from the substrate to obtain a thinstrip on the first Al layer. A schematic diagram of
the Al thin strip is shown in Fig. 1. Since every
material induces a phase shift depending on its
index of refraction, we used the same material for
the surface and the thin strip on top of it, in order
not to cause different phase shifts and not to find
incorrect height values at the boundary. We per-
C. Karaalio~glu, Y. Skarlatos / Optics Communications 234 (2004) 269–276 273
formed the experiments to measure the second
strip�s thickness. When we measured its thickness
by a commercially available interferometric
thickness monitor (Varian Ascope* Interferometer
Model No. 980-4006), we found that it changed
approximately from 284.7 to 261.2 nm dependingon the exact location. Naturally this method was
not a non contact as the Fizeau plate of the
thickness monitor presses against the sample dur-
ing measurement. A schematic diagram of the ex-
periment is shown in Fig. 2.
We used a He–Ne laser with a wavelength
k ¼ 632:8 nm as our light source. The distances
between object–beam splitter and mirror–beamsplitter are not important in our experiment since
the main point is just to obtain the interference
pattern caused by the Al strip. The pictures were
taken of the film surface of size 1.260 cm� 0.945
cm. We obtained reference and deformed states by
sliding the glass substrate which was mounted to a
microblock. When we illuminated an area which
included the part of Al strip�s part, the resultantinterference pattern was our ‘‘deformed state’’.
We used a commercial digital still camera
(SONY, CVX-V18NSP) to directly capture digital
images of the speckle fields. A schematic diagram
of the experiment is shown in Fig. 2. Light from
the laser is expanded by the beam expander BE
and is incident on the beam splitter BS. The two
beams from BS then fall on the object and themirror. The speckle fields created by the object and
MirrorBS
Object
BE
ND filter
Serial cable
PC
CCD Camera
Laser
Fig. 2. Schematic diagram of the experimental setup for FFT
method (BS: beam splitter, BE: beam expander, ND filter:
neutral density filter, PC: personal computer).
the mirror travel in opposite directions and inter-
fere with each other in the image plane of the
digital still camera after passing through the
camera lens. The digital still camera has a
576� 768 pixel charged couple device(CCD) array
as its recording element. A neutral density filter(THORLABS Mounted Metallic ND Filter
D ¼ 0:4; 0:5) was placed in front of the camera to
reduce the intensity of the incoming light, and thus
enhance the images. The aperture and shutter
speed of the camera were selected appropriately to
have the maximum light intensity without satu-
rating the camera. The exposure time of the cam-
era was set at 1/150 s. The image acquisition cardconverts the images directly into PNG image files,
which are in turn converted into JPEG images
since the latter require much less memory space in
the hard disc of the computer. Subsequently, a
special software is used to convert the images from
RGB color mode to gray-scale mode. We thus
have 576� 768 pixel images, where each pixel is
represented by 1 byte of data (256 gray levels). Onesuch image is shown in Fig. 3. Three phase shifted
images are produced by the use of quarter wave
plates (THORLABS WPQ05-532). The sub-
sequent additional phase changes are p=2, p, and3p=2.
3.2. Plots and numerical results from the experi-
ments
The fringe patterns were digitized with an 8-bit
gray scale, corresponding to 256 levels.
Fig. 3. Interference pattern obtained from the Al thin film.
274 C. Karaalio~glu, Y. Skarlatos / Optics Communications 234 (2004) 269–276
For the digital image subtraction method,
Fig. 3 is our interference pattern which is obtained
and used in this part of our experiments. When we
illuminated an area which included the part of Al
strip�s part, the resultant interference pattern was
our ‘‘deformed state’’, i.e. f 0cðx; yÞ. When the area
which did not include the Al strip was illuminated,
the resultant interference pattern was our ‘‘refer-
ence state’’, i.e. fcðx; yÞ. The subtraction was made
by using a software program, MATLAB. By
considering the explanations in Section 2.1 and
from Fig. 4, we can find the results as
Dx ¼ 27 pixels; x ¼ 32 pixels:
Fig. 4. One-dimensional plots from the interference pattern of
Fig. 3.
Assuming n ¼ 0, we find the height value to be
approximately z ¼ 267 nm by using Eq. (7) with a
wavelength k ¼ 632:8 nm. As we know from the
interferometric measurement, the film thickness is
to be between 285 and 261 nm. This value is also
checked by measuring with a thickness monitor(Varian Ascope). We cannot accept taking n ¼ 1
since in this case the thickness is found to have an
out of range value of nearly z ¼ 583 nm. The error
on z comes from reading the plots of Fig. 4. This
error in reading the plot is 1 pixel and this corre-
sponds to �10 nm error on z.For the phase shifting method; by considering
the explanations in Section 2.2, if we take thephase value as /0 ¼ 6:15 rad from the right hand
plot of Fig. 5, the thickness value is found to be
z ¼ 261 nm by using the formula /0 ¼ 4pz=k,where k ¼ 532:5 nm is the wavelength of the laser.
Fig. 5. Wrapped phase mesh and one-dimensional unwrapped
phase plot of row100 for the phase shifting method.
C. Karaalio~glu, Y. Skarlatos / Optics Communications 234 (2004) 269–276 275
By plotting one-dimensional unwrapped phase
plots for each row, this method can give infor-
mation on the surface profile on both coordinates
ðx; yÞ. The error on z value comes from the phase
value which is obtained from the right hand plot of
Fig. 5. The error reading the phase value is 0.01rad and this corresponds to �1 nm error on z.
For the FFT method, we found the thickness
261� 2 nm between pixel values from 280 to 420.
We used a He–Ne laser with a wavelength
k ¼ 632:8 nm as our light source. Since phase
values can be measured with a depth (z) resolutionis greater than k=200 with the FFT method [16],
our error is �3 nm. By using the Varian thicknessmonitor mentioned above on the same location,
we found this value to be 261� 3 nm. The plots
and explanations of them can be found in [16].
4. Discussion and conclusion
Our work represents a new formula for imagesubtraction technique for one-dimensional case,
and the comparison of a new application of a well
known FFT technique [16], image subtraction
method, and phase shifting method. We have
measured the thickness of an Al thin film with a 2
nm accuracy by using three different image con-
struction methods. In all methods, the fringe pat-
terns obtained contain information on whether thesurface of an object is flat or deformed.
In the phase shifting method, the phase steps
were chosen such that the phase of the speckle
pattern at each pixel could be extracted by arith-
metic operations on the data. In this method, we
saw that the path difference values could be found
with good accuracy when we compared it to a
known result. In addition to this high accuracy ofthe thickness value, all surface contours can be
obtained in detail. This method is quite powerful,
at the cost of taking four records rather than just
one. Unlike the FFT method, as noise cannot be
separated from the signal, the plot in Fig. 5 is not
as smooth as the unwrapped phase plot of FFT
method [16].
The subtraction image is proportional to thedifference between the undeformed and deformed
images. The difference images contain intensity
fringes located at points where the difference is a
multiple of the quantity d, i.e. displacement. These
intensity fringes are, of course, speckled. The dig-
ital image subtraction method is best when the
deformed area is to be seen. Subtraction enhances
the visibility of the deformed spot; however, it isnot as effective as the FFT or phase shifting
methods if the amount of deformation is to be
measured.
We can summarize all these explanations as:
1. To minimize environmental noise effects, FFT
is preferred to the phase shifting technique since
the time required for fringe acquisition is much
shorter [19].2. The subtraction method results in better accu-
racy than the FFT method. The FFT method
utilizes shifting of the Fourier spectra, and
hence its accuracy is dependent on the digital
frequency resolution [20].
3. The surface acts as a phase modulator of the
fringe pattern, and the first stage in the evalua-
tion of surface shape is the demodulation of theobserved fringe pattern. Phase stepping is not a
true demodulator since it yields total signal
phase rather than the required phase deviation,
while Fourier transform techniques are complex
and are best used by researchers experienced in
their applications [14].
4. There are problems with both fringe phase step-
ping and Fourier transform analysis. Fringephase stepping requires the capture of at least
three images, with the restriction that the object
must remain stationary while the fringes are
moved. On the other hand, Fourier transform
profilometry requires complex processing, and
the effects of signal filtering in the frequency do-
main are difficult to predict. Indeed, frequency
domain filtering has been the subject of exten-sive research [14].
The image subtraction and phase shifting
methods as well as the new application of FFT are
important in the area of thin film thickness mea-
surements for the following reasons: They are easy
to perform and understand; all apparatus used in
the experiments is standard and relatively inex-
pensive; finally, these measurements do not causeany extra deformation on a thin film�s surface sincethese methods are absolutely non-contact. In the
276 C. Karaalio~glu, Y. Skarlatos / Optics Communications 234 (2004) 269–276
future, better results can be obtained with im-
proved CCD cameras and new software. Larger
powers of 2 for the FFT method can be included in
the calculations, and larger surface topographies
can be obtained with higher resolution.
Acknowledgements
We would like to thank Pietro Ferraro for his
helpful suggestions.
References
[1] K.M. Abedin, S.A. Jesmin, A.F.M.Y. Haider, Opt. Laser
Technol. 32 (2000) 323.
[2] M. de Angelis, S. de Nicola, P. Ferraro, A. Finizio, G.
Pierattini, Opt. Commun. 136 (1997) 370.
[3] M. de Angelis, S. de Nicola, P. Ferraro, A. Finizio, G.
Pierattini, Opt. Commun. 175 (2000) 315.
[4] G.S. Spagnolo, D. Ambrosini, D. Paoletti, G. Accardo, J.
Cult. Heritage 1 (2000) 337.
[5] G.S. Spagnolo, G. Guattari, C. Sapia, D. Ambrosini, D.
Paoletti, G. Accardo, Opt. Lasers Eng. 33 (2000) 141.
[6] G.S. Spagnolo, D. Ambrosini, G. Guattari, J. Opt. 28
(1997) 99.
[7] L.C. Wu, F. D�Amelio, R.A. Fox, I. Polyakov, N.G.
Daunton, J. Neurosci. Methods 74 (1997) 89.
[8] K. Creath, SPIE Proc. 4101A-06, 30 May 2000.
[9] K.A. Goldberg, J. Bokor, Appl. Opt. 40 (2001) 2886.
[10] C.J. Tay, C. Quan, H.M. Shang, Opt. Laser Technol. 30
(1998) 27.
[11] Y.M. He, C.J. Tay, H.M. Shang, Opt. Laser Technol. 30
(1998) 545.
[12] H. L€uck, K.-O. M€uller, P. Aufmuth, K. Danzmann, Opt.
Commun. 175 (2000) 275.
[13] C. Quan, X.Y. He, C.F. Wang, C.J. Tay, H.M. Shang,
Opt. Commun. 189 (2001) 21.
[14] C.A. Hobson, J.T. Atkinson, F. Lilley, Opt. Lasers Eng. 27
(1997) 355.
[15] K.M. Abedin, M. Kawazoe, K. Tenjimbayashi, T. Eiju,
Opt. Eng. 37 (1998) 1599.
[16] C. Karaalio~glu, Y. Skarlatos, Opt. Eng. 42 (6) (2003).
[17] M. Takeda, H. Ina, S. Kobayashi, J. Opt. Soc. Am. A 72
(1982) 156.
[18] C. Quan, C.J. Tay, H.M. Shang, P.J. Bryston-Cross, Opt.
Commun. 118 (1995) 479.
[19] L.-S. Wang, S. Krishnaswamy, Laser Interferometry VIII:
Techniques and Analysis, vol. 2860. Proc. of the SPIE, The
Int. Soc. for Opt. Eng., Denver, USA, 1996, pp. 162–174.
[20] Y.M. He, C.J. Tay, H.M. Shang, Opt. Lasers Eng. 30
(1998) 367.