Home Search Collections Journals About Contact us My IOPscience

A nanometric displacement measurement method using the detection of fringe peak

movement

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2000 Meas. Sci. Technol. 11 1352

(http://iopscience.iop.org/0957-0233/11/9/314)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 130.220.71.19
The article was downloaded on 21/06/2012 at 20:54

Please note that terms and conditions apply.

Meas. Sci. Technol. 11 (2000) 1352–1358. Printed in the UK PII: S0957-0233(00)13044-9

A nanometric displacement
measurement method using the
detection of fringe peak movement
Jong Hoon Yi, Soo Hyun Kim and Yoon Keun Kwak
Department of Mechanical Engineering, Korea Advanced Institute of Science and
Technology, 373-1, Kusong-dong, Yusung-gu, Taejon 305-701, Republic of Korea
E-mail: [email protected] (Yoon Keun Kwak)

Received 4 April 2000, in final form and accepted for publication 8 June 2000

Abstract. This paper proposes a novel method for measuring nanometric displacement by
detecting fringe movement of interferograms which is linearly proportional to the
displacement of an object. Interferometers have been used for precision measurement of
displacement, but they have such error sources as unequal gain of detectors, imbalance of
beams and lack of quadrature. These error sources degrade the accuracy of the
interferometer. However, the fringe movement of interferograms has little relation to these
error sources. In order to investigate the performance of the proposed method, analyses and
simulations of speckle noise, Gaussian noise and wavefront distortion were executed. Results
of the simulations show that the proposed method is robust against these errors. Experiments
were performed to verify this method.

Keywords: displacement, fringe peak, laser interferometer, least-square fitting

1. Introduction

Laser interferometers are most widely used in precise

linear displacement measurements such as semiconductor
metrology and lithographic applications for their capability
of obtaining sub-micrometre order resolution. Many
techniques have recently been proposed in order to
improve the performance of interferometers measuring linear
displacement: multiple-wavelength [1], homodyning [2]
and heterodyning [3] techniques and so on. However,
the nonlinearity and systematic errors of interferometers
in linear displacement measurement have been reported
[4–6]. For homodyning interferometers, these errors result
from unequal gain of detectors, imbalance of beams, the
zero offset and lack of quadrature. Hence, in order to
obtain nanometric resolution, one has to correct for these
error sources. Heydemann [7] proposed a technique of Figure 1. The configuration of the proposed interferometer
system.
least-square fitting of experimental data in order to assess
and correct for these errors. Similar approaches were
performed in order to improve the phase measuring accuracy mirror is slightly tilted to generate equally spaced fringes.
[8, 9]. However, their efforts did not thoroughly solve these The system acquires an image of interference fringes on the
problems and could not correct for the error source occurring linear CCD to find fringe peak positions and traces the fringe
in processing measurements. movement to obtain the displacement of the target mirror.
We propose a novel interferometric method and a Wang et al [10] proposed an image-matching algorithm
signal processing algorithm for precision displacement for detecting the movement of this type of fringe, but this
measurement by minimizing the effects of the phase error. algorithm resulted in a very complex and time consuming
The proposed configuration, shown in figure 1, is similar method. We present a simple method capable of detecting
to the Michelson interferometer except that the reference the fringe movement by identifying fringe peak positions.

0957-0233/00/091352+07$30.00 © 2000 IOP Publishing Ltd

 Nanometric displacement measurement

2. Theoretical background y N 1 y( N +1)1 y( N + 2)1 ... y( N +M −1)1 y( N + M )1

I ( y)
The configuration of the proposed interferometer is shown in
figure 1. An incident beam from the laser source and spatial
filter is divided by the beam splitter into reference and target
beams. The reflected beams make interference fringes on the
linear CCD as in figure 1.
The magnitudes of the electrical fields can be written as y
(a)
Er (x, y) = Ero (x, y) e{i[kWr (x,y)−ωt+φr ]}
...
Et (x, y) = Eto (x, y) e{i[kWt (x,y)−ωt+φt ]}
y N 2 y( N +1) 2 y( N + 2 ) 2 y( N +M −1) 2 y( N + M ) 2
(1) I ( y)

where Wr and Wt represent the optical paths and φr and φt

are the constant initial phases for each electrical field. The
subscripts r and t stand for the reference and target mirrors,
respectively. The interference intensity produced by two
electrical fields can be expressed as y
I (x, y) = I0 (x, y){1 + V (x, y) cos{k[W (x, y) − δ]}} (2) (b)
Figure 2. Equally spaced interference fringes: (a) z = z1 and
where I0 (x, y) is the average intensity, V (x, y) is the fringe (b) z = z2 .
contrast, W (x, y) is the optical path length difference, k is
the wavenumber (2π/λ) and δ = φr − φt . The interference where n represents the fringe order and yn is the fringe peak
fringe caused by the optical path length difference W (x, y) position for the nth fringe order in the interference pattern.
is The interference fringe patterns described by equations (7)
W (x, y) = Wr (x, y) − Wt (x, y). (3) and (8) are shown in figure 2. When the fringe order is N ,
If the reference and the target mirrors are perfectly flat the fringe peak position for each displacement (zk ) is
and the reference mirror is slightly titled as in figure 1, the Nλ + 2zk
optical paths can be described by .yN k = (10)
2 tan α
Wr (x, y) = 2y tan α (4) We easily obtain the displacement using the difference
between the fringe peak positions for z1 and z2 :
Wt (x, y) = 2z (5) z2 − z1
yN 2 − yN 1 =
. (11)
where α is the tilting angle of the reference mirror and z is tan α
the displacement of the target mirror. Equation (3) can be In practice, wavefront distortion and random noise
rewritten as follows: deform the interferogram, degrading the accuracy of the
measurement. In order to reduce this effect, we take an
W (x, y) = 2y tan α − 2z. (6)
average of the differences of the fringe peak positions. In
In the interferometer, the fringe moves when the target figure 2, the number of fringe peaks is M + 1. From
mirror is displaced. The amount of the fringe movement has a equation (11), taking the sum from the fringe order N + 1
linear relationship with the displacement of the target mirror. to N + M, we have
Each interference fringe is described as

N +M

N +M  
z2 − z 1
yn2 − yn1 = M . (12)
I1 (x, y) = I0 (x, y){1 + V (x, y) cos[k(2y tan α − 2z1 − δ)]} n=N +1 n=N +1
tan α
(7) The tilting angle of the reference mirror is an unknown
I2 (x, y) = I0 (x, y){1 + V (x, y) cos[k(2y tan α − 2z2 − δ)]}. value in equation (12) and can hardly be measured because
(8) it is extremely small. However, the tilting angle can be
Classical methods for detecting the phase difference computed using the interval between the peak positions of
involve identifying the maximum value of the cross fringe order N and N + M, because this one is related to the
correlation between the reference and measured signals. fringe frequency of the interferogram. Subtracting yN 2 from
However, cross correlation is not appropriate for real-time y(N +M)2 and rearranging, we can get this relationship as
applications because of its computational complexity.
Mλ
tan α = . (13)
2.1. The displacement measuring method using fringe 2(y(N +M)2 − yN 2 )
peak position Finally, we obtain the relationship between the relative
We obtain a maximum intensity of the fringe pattern when displacement and the fringe peak positions by substituting
the optical path length difference (W ) is an integer multiple equation (13) in equation (12) and rearranging:
 N +M 
of the wavelength (λ) of the light source, that is N +M
λ n=N +1 yn2 − n=N +1 yn1
z = (14)
2yn tan α − 2z = nλ (9) 2 y(N +M)2 − yN 2

1353
Jong Hoon Yi et al

Figure 3. The procedure for detecting the fringe peak positions.

1.5
Cross correlation
Fringe Peak Detection
Error (σ : nm)

0.5

(a)
0
5 10 15 20
Spatial Frequency of Fringe (cycles/pattern)
(a)
2
Cross correlation
Fringe Peak Detection

1.5
Error (σ : nm)

(b)

Figure 4. Wavefront distortion: (a) M = 6 and (b) M = 30. 0.5

where z = z2 − z1 . From this equation, we know that the

0
displacement can be measured by detecting the fringe peak 5 10 15 20
positions. Spatial Frequency of Fringe (cycles/pattern)
(b)

2.2. The procedure for detecting the fringe peak Figure 5. Simulation results with noise (noise level 2.5%):
(a) speckle noise and (b) Gaussian noise.
positions

The fringe peak positions should be precisely obtained in positions. The experimentally collected fringe images are
order to measure the displacement using equation (14). rarely as good as the simulated ones, so a low pass filtering
Figure 3 shows the procedure for detecting the fringe peak operation is required in order to remove noise. The peak

1354
 Nanometric displacement measurement

2 2
Cross correlation Cross correlation
Fringe Peak Detection Fringe Peak Detection

1.5 1.5

Error (σ : nm)
Error (σ : nm)

1 1

0.5 0.5

0 0
5 10 15 20 5 10 15 20
Spatial Frequency of Fringe (cycles/pattern) Spatial Frequency of Fringe (cycles/pattern)
(a) (b)
2 2
Cross correlation Cross correlation
Fringe Peak Detection Fringe Peak Detection

1.5 1.5
Error (σ : nm)

Error (σ : nm)
1 1

0.5 0.5

0 0
5 10 15 20 5 10 15 20
Spatial Frequency of Fringe (cycles/pattern) Spatial Frequency of Fringe (cycles/pattern)
(c) (d)
Figure 6. Simulation results with wavefront distortion (maximum wavefront distortion λ/5): (a) defocus (Wd (x, y) = A(x 2 + y 2 )),
(b) spherical aberration (Wd (x, y) = A(x 2 + y 2 )2 ), (c) astigmatism (Wd (x, y) = A(x 2 + 3y 2 )) and (d) coma (Wd (x, y) = Ay(x 2 + y 2 )).

position is determined by fitting the intensity signal to the 3. Error analysis

quadratic curve around the peak position, so the approximate
peak position (ỹn ) must be known. The threshold, the mean As mentioned in section 1, the performance of general
value of the signal, allows us to find the approximate peak interferometer systems is affected by error sources such as
positions in the fringe pattern as in figure 3. The middle polarization mixing, unequal gain of detectors, imbalance
positions of the edges are assumed to be the approximate of beams and lack of quadrature. These error sources have
peak positions. little effect on the fringe movement and the performance
The signal from the linear image sensor is sinusoidal as of the proposed method. The significant error source of
shown in figure 3 and can be given by this system is the wavefront distortion which varies the
  interference fringe. Thus, we perform an error analysis
2π
Ii = b + c cos (yi − ypeak ) (15) considering the wavefront distortion and a simulation of
p wavefront distortion and random noises such as speckle and
where Ii , yi , p and ypeak are the ith pixel intensity, the Gaussian noise in order to compare the proposed method with
ith pixel position, the period of the signal and the peak the cross correlation method.
position, respectively. This equation can be approximated
by a Taylor expansion around the peak position. Ignoring 3.1. Wavefront distortion
the higher order terms, the approximated equation is a
quadratic polynomial; thus the intensity signal is well fitted The collimating lens generates a plane wave as in figure 1.
to the quadratic polynomial. The quadratic curve is fitted to However, an irregular optical surface and a mis-alignment
several observation positions around the approximated peak of the optical element cause wavefront distortion which
positions by using the least-squares method to determine the seriously affects the system performance. The incident
peak positions with sub-pixel accuracy. The error of the beam from the collimating lens is not a plane wave but a
least-squares algorithm is reduced as the number of fringe spherical wave when the focuses of the objective lens and
patterns is increased because the random noise error included the collimating lens do not coincide. Generally, wavefront
in the interference pattern is rejected through the averaging distortion is produced when a plane wave passes through
procedure. optical elements such as lenses and beam splitters. The

1355
Jong Hoon Yi et al

(a)

(b)

Figure 7. The experimental set-up for precision position measurement: (a) a schematic diagram, (b) photography.

wavefront of an optical system has a distortion given by the wavefront distortion term (Wdn ) and the number of
the fringe peak position. As M increases the distance
W  (x, y) = W (x, y) + Wd (x, y) (16) between yn1 and yn2 decreases and Wdn gets close to 0.
Hence, as M increases, the summation term in the numerator
where W (x, y) is the ideal wavefront and Wd (x, y) is the
of equation (18) converges to zero and the denominator
wavefront distortion. In the proposed system, the optical
approaches 1. Therefore, the measured displacement (z̃)
path length difference is written as
converges to the actual displacement (z) and this error can
W (x, y) = 2y tan α − 2z + Wd (x, y). (17) also be eliminated by detecting the wavefront distortion in
the calibration procedure, because it is not dependent on the
The relative displacement of the target mirror can be displacement of the object.
obtained by the same procedure as that in the previous section:
  N
+M

N +M 
z̃ = z − Wd (yn2 ) − Wd (yn1 ) (2M) 3.2. Simulations
n=N +1 n=N +1
Speckle noise, Gaussian noise and wavefront distortion were
×[1 − (Wd (y(N +M)2 ) − Wd (yN 2 ))/(Mλ)]−1 superposed on the ideal interference image for evaluation
 +M
z − N n=N +1 Wdn /(2M) of the measurement error in this simulation. It is often
= (18)
1 − (Wd (y(N +M)2 ) − Wd (yN 2 ))/(Mλ) convenient to express wavefront data in a polynomial form
where z̃ is the relative displacement calculated by the as a means of dissecting the data into physically meaningful
proposed procedure, z is the actual displacement and terms. The individual terms of the polynomials correspond
Wdn = Wd (yn2 ) − Wd (yn1 ). Equation (18) shows that the to many of the aberrations commonly found in optical tests
wavefront distortion causes the displacement error which can such as defocus, spherical, astigmatism and coma aberrations
be divided into the numerator term and the denominator term. [11].
The denominator term results from the difference between The simulation was carried out for each error source
the wavefront distortions at the first and last fringe peak (speckle noise, Gaussian noise, defocus error, spherical
positions. This term is inversely proportional to the number aberration, astigmatism and coma). In the simulation, the
of the fringe peak position (M). Thus, this can be reduced cross correlation method and fringe peak detection method
by increasing M. Figure 4 shows the relationship between were used to calculate the fringe movement. The interference

1356
 Nanometric displacement measurement

3
Intensity

Displacement (nm)
1

Pixel 0

Figure 8. The interference fringe pattern obtained in an

−1
experiment.

−2
pattern used in the simulation was generated by
 
2π −3
Ii = round 100 + 100 cos (2yi tan α − 2z) 0 200 400 600 800 1000 1200
λ Time (sec)
(a)
+Wd (yi ) . (19)
1.5
The displacement of the target mirror z was varied from −λ/4 Line Fitting :
to +λ/4. In the cross correlation method, the displacement y = 0.9984x+0.001

Interferometer output (µm)

was measured by detecting the maximum peak position of σ = 2.703 nm
the cross correlation function D = z/tan α. 1
Figure 5(a) shows the effect of the speckle noise on the
performance of the measurement system. The x axis is the
spatial frequency of the interference fringe which is linearly
related to the number of fringe peak positions as described 0.5
in the previous section. The y axis is the standard deviation
of the displacement error. In the cross correlation method,
as shown in figure 5(a), the error increases as the spatial Line Fitting
frequency of the interference fringe increases. On the other Experimental Result
0
hand, in the fringe peak detection, the error decreases as the 0 0.5 1 1.5
Strain gage output (µm)
spatial frequency of the interference fringe increases. The
effect of the Gaussian noise is similar to the case of the speckle (b)
noise shown in figure 5(b). In these simulations, the noise Interferometer Output
level is 2.5% of the peak-to-peak fringe intensity and the 2 Strain Gage Ouput
Displacement (µm)

standard deviation of the error is less than 1.0 nm both for 1

speckle and for Gaussian noise. The proposed fringe peak 0
detection method is less sensitive to noise than is the cross-
−1
correlation method.
Figure 6 reveals that the measurement error originating −2
0 200 400 600 800 1000 1200
from wavefront distortions decreases proportionally to the Time (sec)
spatial frequency of the fringe in the fringe peak detection. 20
In the low spatial frequency region, the error of the cross
Error (nm)

10
correlation method is less than that of the fringe peak
detection method. In equation (18), the wavefront distortion 0
is averaged over the maximum fringe position of each step.
The difference between the averaged wavefront distortions of −10
0 200 400 600 800 1000 1200
the two steps decreases with increasing spatial frequency of Time (sec)
the fringe. Therefore, the error of the displacement decreases (c)
as the spatial frequency of the fringe increases. In these
Figure 9. Experimental results: (a) under free conditions, (b) for
simulations, the maximum wavefront distortion is λ/5 and linear displacement and (c) for sinusoidal displacement.
the standard deviation of the error is less than 0.5 nm in the
high spatial frequency region.
tilted so that the small back reflection towards the laser cavity
could be removed. Two quarter-wave λ/4 plates were used
4. Experimental results
to prevent optical feedback of the reflecting beam from the
Figure 7 shows the experimental set-up. A He–Ne laser mirrors.
was used as the light source of the interferometer and For the purpose of demonstration, the target mirror was
its wavelength was 0.6329 µm. The components in the mounted on a micro-stage (Newport 461-x-m). A PZT (Tokin
experiment were all anti-reflection coated for the He–Ne laser NLA-5 × 5 × 9) was attached to the micro-stage to generate
wavelength. The beam splitter and the polarizer were slightly the displacement of the target mirror. The PZT was driven by

1357
Jong Hoon Yi et al

the signal from the PZT amplifier. A strain gauge was used with sub-pixel accuracy. This algorithm was more
in order to measure the displacement of the PZT. The gain accurate and less time consuming than was the cross-
of the strain gauge was 0.344 µm V−1 . The sensing signal correlation method. The error analysis and simulation
of the strain gauge was collected with a 16-bit A/D converter results showed that this method was effective for precision
which was interfaced to a host PC. The interference fringe position measurement. The wavefront distortion caused by
patterns were detected by the linear image sensor which optical elements was the error source of the displacement
was connected to an image grabber. From these data, the measurement, but its effect became less significant as
fringe peak position was obtained and the displacement of the number of the interference fringe increased. The
the target mirror was calculated using equation (14). In experimental result shows the possibility of measuring
the measurement, the interference fringe was obtained as displacement using the proposed method. The precision
1024 pixels and a 12-bit grey level image. displacement measurement proposed in this paper is more
Figure 8 shows the experimentally collected interference accurate than are the conventional techniques and provides
fringe with noise around the fringe peak positions. The robuster information about displacement. The calibration
spatial frequency of the fringe was about 18 cycles per and further experiments will be carried out in order to verify
pattern. An unstable intensity of the light source caused the the performance of this system.
interference fringe to have an envelope of the fringe peaks.
Figure 9 shows that the displacement of the target Acknowledgment
mirror was measured using the proposed method. In
order to show the stability of the proposed system, the This work was supported in part by the Brain Korea 21
displacement was measured when the target mirror was Project.
fixed as shown in figure 9(a). Under this condition, the
standard deviation and the mean of the measurement error
References
are 0.897 nm and 0.196 nm, respectively. The linear
displacement was generated to verify the performance of [1] Dandliker R, Hug K, Politch J and Zimmermann E 1995
this system. The displacement of the target mirror was High-accuracy distance measurements with
measured by the proposed method and monitored by the multiple-wavelength interferometry Opt. Eng. 34 2407–12
strain gauge attached to the PZT, simultaneously. Figure 9(b) [2] Rugar D, Mamin H J, Erlandsson J E R and Terris B D 1988
Force microscope using a fiber optic displacement sensor
shows the experimental results, in which the x axis is the Rev. Sci. Instrum. 59 2337–40
strain gauge output and the y axis is the output of the [3] Zhao Y, Zhou T and Li D 1999 Heterodyne absolute distance
proposed system. In figure 9(b), the experimental results interferometer with a dual-mode hene laser Opt. Eng. 38
are plotted as the dotted line and the linear fitting as the 246–9
full line. In this experiment, the standard deviation of the [4] Hou W and Wilkening G 1992 Investigation and
compensation of the nonlinearity of heterodyne
measurement error is 2.703 nm. Figure 9(c) shows the interferometer Precision Eng. 14 246–9
experimental result when the sinusoidal displacement was [5] Wu C-M and Su C-S 1996 Nonlinearity in measurements of
generated. In figure 9(c), the measured displacement is length by optical interferometry Meas. Sci. Technol. 7
well matched with the strain gauge output and the standard 62–8
deviation of the error was 2.77 nm. These results include [6] Bobroff N 1993 Recent advances in displacement measuring
interferometry Meas. Sci. Technol. 4 907–26
not only the error source mentioned in the previous section [7] Heydemann P L M 1981 Determination and correction of
but also the misalignment of the mechanical element and the quadrature fringe measurement errors in interferometer
displacement error of the strain gauge. These experiments Appl. Opt. 20 3382–4
show that nanometric displacement measurement could be [8] Birch K P 1990 Optical fringe subdivision with nanometric
achieved using the proposed system with precise calibration. accuracy Precision Eng. 12 195–8
[9] Wu C-M, Su C-S and Peng G-S 1996 Correction of
nonlinearity in one-frequency optical interferometry
5. Conclusion Meas. Sci. Technol. 7 520–4
[10] Wang Z, Graca M S, Bryanston-Cross P J and
We proposed a novel method and signal processing Whitehouse D J 1996 Phase-shifted image matching
algorithm for displacement measurement Opt. Eng. 35
algorithm for precision displacement measurement using an 2327–32
interferometer. For measuring the displacement, the fringe [11] Malacara D 1992 Optical Shop Testing 2nd edn (New York:
peak position was detected by the least-squares fitting method Wiley)

1358