Research Article Vol. 56, No.

25 / September 1 2017 / Applied Optics 7089

Line-scanning laser scattering system for fast

defect inspection of a large aperture surface
JINGTAO DONG
School of Instrument Science and Opto-Electronic Engineering, Hefei University of Technology, Hefei, Anhui 230009, China ([email protected])

Received 27 April 2017; revised 5 July 2017; accepted 7 August 2017; posted 7 August 2017 (Doc. ID 294477); published 25 August 2017

Inspection of defects with micrometer level on large aperture surfaces with hundreds of millimeters is one of the
challenges in surface quality evaluation. Various microscopic imaging methods have been applied to inspecting
the surface defects, while they are time-consuming for the small field of view and the sub-aperture stitching. To
tackle this problem, a high-speed line scanning system based on the dark-field laser scattering method is pro-
posed. The laser beam is scanned by the rotating polygon mirror to a laser line for high throughput and then the
telecentric F-theta lens converges each incoming laser beam to a focused spot that creates a high intensity to
enhance the signal-to-noise ratio. The scattered light from surface defect is collected by the designed integrating
sphere for low background noise and the scattering signal is detected for each focused spot at a proper acquisition
rate by a photomultiplier (PMT) detector with extremely short response time. In the meanwhile, the tested surface
is moving perpendicular to the laser line to realize high-speed large area inspection. The defect inspection system
is confirmed experimentally with laser line length of 60 mm, minimum detectable size less than 0.5 μm, and figure
of merit of 9.6 cm2 s−1 μm−1 . The work put forward an effective method for automatic discovery of surface defects
such as scratches, digs, and contaminants on large aperture surfaces. © 2017 Optical Society of America
OCIS codes: (120.5820) Scattering measurements; (120.4630) Optical inspection; (120.4640) Optical instruments; (120.4290)
Nondestructive testing.

https://doi.org/10.1364/AO.56.007089

1. INTRODUCTION collects the scattering light induced by the surface defects. The
The demand of surface defect inspection arises from the necessity imaging resolution is less than 10 μm. However, the camera
for producing high-quality components which have been widely image shows nonuniformity of the illumination that affects
applied to the precision optical systems. For example, hundreds the inspection performance. Liu et al. [4] developed a digital
of large aperture optical elements with size of 810 mm × microscopic imaging system with a motorized zoom lens and
460 mm are employed in the inertial confinement fusion dark-field illumination. To detect the surface defects down to
(ICF) system, while the surface defects with a size of micrometer micrometer-level dimension, the field of view (FOV) is only
level can lead to significant damage of the optical elements when several millimeters so that the sub-aperture stitching [5] and
illuminated by a high-power laser [1]. Therefore, high efficient distortion correction [6] becomes a significant issue. The image
inspection of small defects on large aperture surfaces becomes acquisition time in sub-aperture scanning for 810 mm ×
one of the challenges in surface quality evaluation. 460 mm needs more than half an hour and the implementation
A number of defect inspection methods for large aperture of sub-aperture stitching and image processing costs almost 2 h.
surfaces have been proposed. Rainer et al. [2] adopt a total in- Tao et al. [7] established a dark-field imaging system employing
ternal reflection (TIR) technique to discover the surface defects a line scan camera. Thanks to the high-intensity linear light
of large optics ranged up to 1 m dimension in the national igni- source, maximum line rate, and 8k pixel resolution of the cam-
tion facility system. The tested optic is illuminated by the back- era, the system has an imaging resolution of 10 μm and only
lighting and edge-lighting fiber-optic bundles and imaged by a takes about 10 min to inspect an area of 810 mm × 460 mm.
long working distance microscope coupled with a high-resolution However, the illumination intensity, the exposure time of the
45 M pixel CCD camera. It takes more than 20 min to examine camera, and the line scanning speed should be carefully bal-
a sample surface with a size of 300 mm × 300 mm in about anced in case the micrometer-level defects cannot be discovered
5 μm resolution. Daurios et al. [3] also use the TIR technique for the weak scattering light intensity.
in the laser megajoule system. The component is illuminated Laser scattering is a very sensitive technique for the detec-
by visible LED arrays around the edges and a megapixel camera tion of small imperfections and contaminants on surfaces

1559-128X/17/257089-10 Journal © 2017 Optical Society of America

 7090 Vol. 56, No. 25 / September 1 2017 / Applied Optics Research Article

[8–12]. Dou and Broderick [8] proposed a laser scattering Finally, the relevant issues for future consideration are discussed
surface scanning system for wafer inspection. The laser beam and the conclusion of the work is made.
is focused on to the surface and off-axis partial mirrors are
used to collect the scattered light from the defects. The surface
defect size down to sub-micrometer level can be detected. 2. THEORY
Marxer et al. [9] makes use of an ellipsoidal mirror that is placed Light scattering can be classified into three cases according to
with its axis along the wafer surface normal to collect light scat- the relationship between the defect size d and the laser wave-
tered by surface defects at collection angles away from the length λ: (1) Rayleigh scattering when d ≪ λ, (2) Mie scatter-
surface normal. The wafer is scanned spirally at a high speed by ing when 0.1λ < d < 30λ, and (3) geometric scattering when
means of a coordinated translational and rotary movement. It d ≫ λ, where d is defined as the diameter of the particle and
takes about 2 min to examine a wafer with a diameter of dig, or the width of the scratch [13]. The defect in micrometer-
150 mm in about 0.1 μm sensitivity of defect inspection. level dimension on a large aperture surface illuminated by a
However, the configuration of single laser point illumination visible laser beam scatters the light in accordance with the
coupled with spiral scanning for wafer inspection is not appro- Mie theory [14]. In order to inspect surface defects as small
priate for large aperture optics because of the weight and vol- as possible, the scattered intensity versus the noise component
ume (more than 40 kg for 810 mm × 460 mm). Kim [10] of the light intensity, signal-to-noise ratio (SNR), is given
applied a dark-field laser scattering system coupled with a approximately by [15]
CCD camera to the defect inspection of flat-panel glasses. S 2λ · I 2sct 1
The surface defect size in several micrometers can be detected. SNR ∝ · ; (1)
S λ
I sct  I B   i D ΔB
However, the incident laser angle and the viewing angle of
the camera, which are the most influential factors for the where I sct is the scattered light intensity by a surface defect, S λ
is the spectral sensitivity of the photodetector (e.g., PMT), I B is
optimal scattering distribution, should be carefully selected for
the background light intensity, i D is the dark current of the
micrometer-level defects. In addition, sub-aperture stitching is
photodetector, ΔB is the frequency bandwidth of the electron-
necessary for large area inspection. Yazaki et al. [11] developed a
ics. In order to maximize the SNR, an increase in both S λ and
dark-field laser scattering system based on a hybrid-dispersion
I sct , together with a decrease in I B , ΔB, and i D should be made.
line scanning configuration that illuminates temporally and
The theoretical model for the scattered light intensity I sct in
spatially dispersed laser pulses on a fast-moving surface and
Eq. (1) can be divided into two different approaches: a numeri-
detects scattered light from defects on the surface with a high-
cal approach that solves Maxwell’s equations [16,17] and an
sensitivity photodetector in a dark-field configuration. The
approximate approach called the Mie theory base on a far-field
length of the dispersed laser line is 30 mm and the surface de- analytical solution of Maxwell’s equations [18,19]. To illustrate
fect size down to 10 μm can be detected at a translation speed of the main parameters of instrumentation design frankly, I sct is
3 m/s of the surface. Whereas, the high-speed scanning makes approximately expressed as a function F of the parameters as
the image of the defect blur due to the long exposure time of follows:
the image sensor. Kim et al. [12] built a line-laser surface light
scattering system by using a laser line generator and a CCD I sct ∝ F
λ2 f
α; n; θs ; ψ s  · ωc · I i ; (2)
camera. The length of the laser line is about 60 mm and where α 
πd ∕λ, d is the defect size, λ is the laser wave-
the particle as small as 0.7 μm can be discovered at the line length, n is the defect refractive index with respect to the sur-
scan speed of 17 mm/s. However, the laser line generator with rounding medium, θs is the in-plane scattering angle (θ), ψ s is
a fan angle of 20° causes the problem that the scattering inten- the off-plane scattering angle (ψ), ωc is the solid angle of the
sity of the defect on positions that are further away from the scattered light collector, and I i is the incident beam intensity.
center of the laser line decreases dramatically, which implies the Referring to Eq. (2), in the aforementioned case (2) of Mie
nonuniform sensitivity of scattering signal detection. scattering, f
α  αn , where 2 < n < 6 [13], the scattered
In this paper, a high-speed line scanning system based on the light intensity I sct is proportional to the range between
dark-field laser scattering method for large aperture surface is
πd 2 and
πd 6 ∕λ4 . It can be seen that as the defect size de-
proposed. The line scan is implemented by a polygon mirror creases to one half, the wavelength has to be decreased to less
and a telecentric F-theta lens. The collector is composed of the than one half in order to keep the scattered light intensity the
designed integrating sphere and a high sensitive photomulti- same. Be noted that for case (1) called Rayleigh scattering,
plier (PMT) detector. The tested surface moves perpendicular I sct ∝
πd 6 ∕λ4 , and for case (3) called geometric scattering,
to the laser line to realize the high-speed large area inspection. I sct ∝
πd 2 , which is independent of the wavelength.
The rest of the paper is organized as follows. The theoretical The response of the scattered light collector is defined by
background of defect inspection by laser scattering is first in- integrating the angular relative-intensity distributions over
troduced. The overall system layout is described in the follow- the collection solid angle ωc [in Eq. (2)] and is expressed by
ing section. The system implementation, including beam Z
shaping, line scanning strategy, and normalization of sensitivity R I s
θsin2 ψ  I p
θcos2 ψdωc ; (3)
ωc
of scattering signal detection, is then discussed explicitly. The
system performances such as background noise, minimum where I s and I p are s- and p-polarized relative scattering inten-
detectable size, and figure of merit are evaluated, and the ex- sities, respectively, and the detailed expressions are found in
perimental result for a large aperture optical surface is given. [10]. The solid angle ωc of the scattered light collector is
 Research Article Vol. 56, No. 25 / September 1 2017 / Applied Optics 7091

required as large as possible to improve the SNR of the system

in the practical design.
Increasing the incident beam intensity I i in Eq. (2) is an
effective means to improve the SNR. Similarly, an increased
SNR can be achieved by reducing the laser spot size with a high
power density as well. Reducing the spot size, on the other
hand, effectively decreases the amount of light that is scattered
from the surface relative to the defect scattering. Fig. 1. Summary of parameters for increasing the SNR of surface
defect inspection.
Referring back to Eq. (1), the background light intensity I B
mainly comes from surface scattering called “haze,” induced by
the surface microroughness, which is commonly quantified by
root-mean-square (RMS) roughness σ rms . A more general de- photocathode, and ΔB is the frequency bandwidth of the elec-
scription as a function of spatial frequencies of the surface tronics. On the one hand, it is clearly important that any cur-
roughness is expressed by 2D power spectral density (PSD) that rent (e.g., from stray light) flowing in the detector will generate
can be found in Eqs. (1)–(4) in [20]. Since the surface inves- the background noise that submerges the real scattered light
tigated in our work is isotropic, the RMS roughness can be signal from the surface defect. In addition, other electronic
obtained by integrating the 2D PSD as noises, such as amplifier voltage noise and amplifier current
Z f noise, are also important and should be properly chosen for
max
σ 2rms  2π PSD
f f df ; (4) low-noise applications. On the other hand, decreasing shot
f min
noise can be realized by means of reducing the frequency band-
where f is the spatial frequency (in unit of μm−1 ) that related width ΔB, which implies that the scattering signal pulses must
to the scatter angle by the one-dimensional grating equation be broader, which, in turn, implies slower scanning of the
f  sin
θs  − sin
θi ∕λ. θs and θi represent the angles of scat- surface.
tered and incident light. (f min , f max ) is the spatial frequency In this section, the theoretical background of defect inspec-
interval and is given by f min  1∕L and f max  1∕2d s , re- tion by laser scattering has been introduced from the improve-
spectively, where L is the scan length and d s is the sampling ment of the SNR as a starting point. The main parameters
distance. The relationship between haze H and RMS roughness related to the system implementation were discussed
σ rms is thus given by [21] specifically and are summarized in Fig. 1.
 
4π cos
θi 
H · R 0 · σ 2rms ; (5)
λ 3. SYSTEM LAYOUT
where R 0 is the specular reflectance as a function of wavelength, The high-speed laser line scanning system for large aperture
incident angle, and polarization. If the inspection of the surface surface inspection, as shown in Fig. 2, has been set up based
defect as small as possible is interesting, the surface must be on the parameters analyzed in Section 2. Figure 2(a) shows
sufficiently smooth that the diffused light remains substantially the overall system layout consisting of motorized XYZ stages
weaker than the amount of light scattered by the defect accord- assembled on a marble vibration isolation platform, an optical
ing to Eq. (5). probe mounted on the Z stage, and a holder used to fix the
It also can be seen from Eq. (5) that the background light sample.
intensity is able to be reduced, provided that the incident angle Figure 2(b) is the schematic diagram of the optical probe.
and the polarization state of the laser are properly chosen. In A linearly polarized beam from the laser (central wavelength of
general, larger incident angles (i.e., dark-field illumination)
provide better results because the penetrating electromagnetic
radiation interacts with the lineated subsurface defects, which
acts like a grating and scatters the light back out through the
surface. Therefore, the incidence of p-polarized light at the
Brewster’s angle would minimize the scattering intensity as well
as the Fresnel reflectance of the background surface.
Both the spectral sensitivity S λ and the dark current i D in
Eq. (1) are the given characteristics of the photodetector that
can be used for the optimum selection of the PMT and the laser
wavelength in the practical design. For a very smooth surface
where the background light intensity signal is extremely low,
Fig. 2. High-speed line scanning system based on the dark-field
the limiting background noise is shot noise, a statistical and laser scattering method. (a) Overall system layout. (b) Schematic dia-
broadband fluctuation of the detector output associated with gram of the optical probe. (c) Detailed schematic diagram of the line
the dark current, which is given by scan. HWP, half-wave plate; HR, high reflector; PM, power monitor;
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi BE, beam expander; BH, beam homogenizer; APPs, anamorphic
i n  2e · G · i · ΔB ; (6)
prism pairs; PMS, polygon mirror scanner (8 faces); TF-theta, telecen-
where i n is the RMS shot noise current, e is the unit electronic tric F-theta; IS, integrating sphere; PMT, photomultiplier; CF, color
charge, G is the PMT gain, i is the mean current leaving the filter; SLB, scattered light baffle; BD, beam dumper; PD, photodiode.
 7092 Vol. 56, No. 25 / September 1 2017 / Applied Optics Research Article

532 nm, TEM00 mode) passes through the half-wave plate and
is then reflected by the high reflector (HR). A small portion of
the intensity of the laser beam through the HR is detected by
the power monitor, which is used to continuously monitor the
power fluctuation of the laser beam as feedback to the designed
automatic power control circuit to maintain the stability of the
laser output power. The rest of the laser beam propagates
through into the beam expander (BE), the beam homogenizer
(BH), and the anamorphic prism pairs (APPs) successively, and
is then redirected by another HR to the polygon mirror scanner Fig. 3. Computer simulation result of beam shaping by anamorphic
(PMS), which scans the incoming laser beam to a laser line with prism pairs. The ratio of the long axis and the short axis of the output
a fan angle, as shown in Fig. 2(c). The photodiode (PD) placed elliptical beam κ depends on the incident angle of the focused laser
beam. For the Brewster’s angle, κ ≈ 1.8, β1  6.5°, β2  23°.
at one side of the laser line generates a line triggering pulse for
each scanning period, which can be used to realize the velocity
stabilization of the PMS and the triggering synchronization of
the line scan. The telecentric F-theta lens converges each in- used to convert the collimated Gaussian beam into a collimated
coming laser beam from the PMS to a focused spot that creates flat-top beam with a uniform intensity distribution. The var-
a focused laser line along the Y direction on the tested surface. iable BE (Edmund Optics, #87-567) placed in front of the BH
The advantage of the telecentric F-theta lens is that all of the is used to adjust the beam diameter to a proper entrance pupil
focused laser beams are parallel to each other and have the same size required by the telecentric F-theta lens (LENSTEK LASER
incident angle θi with respect to the surface normal. The specu- OPTICS, TL-532-70-100).
lar reflected laser beams by tested surface are absorbed by the As discussed in Section 2, the incident angle of the focused
beam dumper. The dark-field scattered light by surface defects laser beam is at the Brewster’s angle (about 56° for glass). The
is collected by the designed integrating sphere (IS) and detected circular flat-top beam will project an elliptical beam on the
by the PMT. The color filter is used in front of the PMT to tested surface, leading to different sampling intervals in X
reduce access of extraneous stray light. The scattered light baffle and Y directions. Therefore, the APPs (Thorlabs, PS873-A)
placed at a distance before the PMT is used to block out the are used after the BE to transform the circular beam of the
light scattered directly from the defects and thus makes the flat-top profile into the elliptical beam by magnifying the cir-
scattering signal more uniform along the laser line. cular beam in one dimension. The tilt angles of the two prisms
The workflow of the system is described as follows. The are optimized according to the incident angle of the focused
optical probe is first focused on the tested surface by the mo- laser line, as shown in Fig. 3, such that the laser spot projected
torized Z stage with the help of a noncontact laser triangular on the tested surface is in circular shape.
displacement sensor. Then the surface is scanned by the optical
B. Line Scanning Strategy
probe along the X direction at a constant velocity and the scat-
tering signals are captured by a high-speed data acquisition Since the laser spot on the surface has been shaped as a circle
(DAQ) card on-the-fly to form a 2D scattering image. The with a flat-top intensity distribution and has a diameter of 9 μm
scanning length and the scanning width of the surface depend focused by the telecentric F-theta lens with the entrance pupil
on the traveling range of the X stage and the length of the size of 10 mm, the line scanning strategy is straightforward,
focused laser line of the optical probe. When the width of as shown in Fig. 4(a). The centers p of adjacent spots cannot
the surface exceeds the length of the focused laser line, the be separated by more than 2ρ  6.3 μm (spot radius
Y stage is used to expand the scan area. The on-the-fly imaging ρ  4.5 μm) in both of the X and Y directions for the full-
technique avoids the complicated sub-aperture stitching algo- size inspection. As the effective length of the focused laser line
rithm and eliminates the stitching time. is 60 mm (this will be discussed in Section 4.C), the effective
number of the acquired points is 10,000 for every single laser
line with the sampling interval of 6 μm for the convenience of
4. SYSTEM IMPLEMENTATION calculation. Note that there is an extremely small angle (about
A. Beam Shaping 0.0057°) between the direction of the laser line and the Y -axis
The intensity distribution of the beam from the laser of the coordinates because of the on-the-fly imaging technique.
(Coherent, OBIS-532LS) operated in the TEM00 mode is However, for the high-speed scanning system in this work, the
Gaussian. When an isolated defect on the tested surface appears period of a laser line is in tens of nanoseconds. As a result, there
in the illuminated laser spot, some light will be scattered and is a dislocation of no more than one spot (i.e., one pixel) for a
collected by the integrating sphere. In principle, the larger the whole laser line that can be neglected without consideration of
defect the greater the scattered light intensity. However, a small the position errors shown in Fig. 4(b).
defect occurring near the central portion of the laser spot may In fact, the laser spots are not able to be located at the exact
produce scattered light intensity equal to or greater than a large positions shown in Fig. 4(a) due to the random errors of the sys-
defect illuminated by the outer portion of the laser spot. This tem components, such as the motorized X stage, the PMS, and
means a small defect located in the different portions of the the laser. Referring to Fig. 4(b), the spot positions in laser line L1
laser spot will have different sensitivities of the scattering signal are accurate, while spots in L2 and L3 have position errors.
detection. Therefore, the BH (Edmund Optics, #34-257) is Table 1 summarizes the position errors and their sources in detail.
 Research Article Vol. 56, No. 25 / September 1 2017 / Applied Optics 7093

fluctuation of the PMS obviously, because the acquisition fre-

quency of the DAQ card (ADLINK, PCIe-9814) is sufficiently
stable.
The random position errors of the laser spot ΔX s and ΔY s
in the X and Y directions (illustrated by the fourth spot in L3 )
result from the combination effects of the pointing stability of
the laser and the velocity fluctuations of both the X stage and
the PMS. In addition, the mechanical vibration and the airflow
disturbance are ignored for the compact integration of the op-
tical probe.
In the implementation of the system, the velocity fluc-
tuation of the motorized X stage (Akribis Systems, DGL180-
AUM3) is less than 10 μm∕s (i.e., 2 pixels) at the desired
Fig. 4. Scanning strategy of the high-speed laser line scanning sys- velocity measured by a laser interferometer (Renishaw, XL-80).
tem. (a) Mapping of laser spots on the surface showing the geometry The pointing stability of the laser is given by less than 30 μrad
for the case of minimum overlap so that no part of the surface is over 2 h, resulting that the position error of the laser spot on
missed. Each spot corresponds to a pixel in the 2D scattering image. the tested surface in both of the X and Y directions is less
(b) Laser spots with position errors induced by the random errors of than 4.5 μm (i.e., 1 pixel) according to the simulation
the system components and the corresponding sequence chart. For the of the optical path. The PMS (Lincoln Laser, modified SOS-
simplification, only three lines and four spots of each line are illus-
BB10) has a relative velocity stability of less than 0.02% at the
trated. S trg−L1 , S trg−L2 , and S trg−L3 are the line triggering pulses of laser
lines L1 , L2 , and L3 generated by the PD, respectively. S acq−L1 , S acq−L2 , velocity range from 5k rpm to 10k rpm by employing the line
and S acq−L3 are the acquisition pulses of DAQ card for L1, L2 , and L3 , triggering pulses as feedback to the homemade control circuit
respectively. T acq is the accurate sampling period of laser spots, while embedded with a proportion-integration-differentiation algo-
0 , T 0 0 , and T 0 0 0 are the inaccurate sampling periods. ΔT
T acq rithm. The position error due to the velocity stability of the
acq acq acq is the
period shift of L2 with respect to L1 . ΔX s and ΔY s are the X and Y PMS is thus less than 2 pixels in the 2D scattering image.
displacements of the laser spot related to its accurate position. Note Since the random errors shown in Table 1 are independent
that the coordinates are the same as those in Fig. 2. of each other, the composite position error is calculated to be
no more than 3 pixels in accordance with the error synthesis
principle.
The difference of the sampling intervals between X 1 and X 2
C. Normalization of Sensitivity of Scattering Signal
(ΔX L  X 1 − X 2 ) in the X direction is mainly caused by the
Detection
velocity mismatch between the X stage and the PMS due to The configuration of single laser point illumination coupled
the velocity fluctuation of the X stage and the velocity drift with an ellipsoidal mirror collector of rotational symmetry de-
of the PMS. If the X stage moves faster than the PMS, the in- scribed in [9] has an advantage that the illuminated laser spot
terval between the adjacent laser lines becomes larger. Conversely, and the collector system’s field remain constant during the
if the PMS rotates faster than the X stage, the interval becomes measurement procedure so that the sensitivity of the scattering
smaller. signal detection is homogeneous over the whole surface.
The line triggering pulse S trg−L2 of laser line L2 is induced Therefore, there are two key issues to be solved for the laser
when the laser beam sweeps over the PD [see Fig. 2(c)]. The line illumination in this work. One is that it is important to
effect of the response time of the PD (Thorlabs, SM05PD2A) design the illumination optics such that the area and the inten-
on the triggering delay is sufficiently small that it can be sity of the illuminated spot remain constant as the beam sweeps
neglected. Thus, the period shift ΔT acq of laser line L2 with across the tested surface. The other is that it is better to keep the
respect to L1 is mostly due to the velocity fluctuation of the constancy of the scattered light collecting angle with respect to
PMS, which means the time of the laser beam sweeping over the clear aperture of the scattered light collector as a function of
the PD is not stable. In addition, the inaccurate sampling the spot location on the tested surface. Otherwise, it would re-
0 , T 0 0 , and T 0 0 0 of L arise from the velocity
periods T acq acq acq 3 sult in different responses to the scattering signal detection as a
function of the defect location along the beam scanning path.
The intensity distribution of the laser line incident to the
Table 1. Summary of Position Errors of Laser Spots and
tested surface, as shown in Fig. 5(a), is measured by rotating
Corresponding Random Errors of the System
Components
the PMS at different angles within the scan angle range
[−25°, 25°] of the telecentric F-theta lens, as shown in
Position Errors Error Sources Fig. 5(b). The nonuniformity of the intensity is less than
ΔX L  X 1 − X 2 Velocity fluctuation of X stage 2.5% over the length of 70 mm.
Velocity drift of PMS In principle, it is difficult to keep the scattered light collect-
0 , T 0 0 and T 0 0 0
ΔT acq , T acq Velocity fluctuation of PMS
acq acq ing angle the same for the defect located along the laser line
ΔX s , ΔY s Pointing stability of laser because the defect on a position that is further away from
Velocity fluctuation of X stage the center of the laser line has a smaller solid angle with respect
Velocity fluctuation of PMS
to the clear aperture of the collector. This work relieves the
 7094 Vol. 56, No. 25 / September 1 2017 / Applied Optics Research Article

changing the configuration variables including the distance

from the baffle to the photocathode of the PMT, the size of
the baffle, the dimensions of the clear aperture of the IS,
and the distance from the clear aperture of the IS to the tested
surface. As shown in Fig. 6(b), the baffle placed in front of the
photocathode of the PMT has a dramatic effect on the uni-
formity of the scattered light intensity distribution as the same
Fig. 5. (a) Intensity distribution of the laser line incident to the defect moves along the beam scanning path, because the baffle
tested surface and (b) the corresponding measurement configuration. blocks the view of the scattered light which has not undergone
The scan angle range [−25°, 25°] of the telecentric F-theta lens cor-
at least two reflections from the sphere surface. The simulation
responds to the laser line length of 70 mm.
result indicates the nonuniformity of the scattered light inten-
sity is less than 10% over the length of 60 mm, which is
specified as the effective length of the laser line.
problem by means of configuration optimization of the IS and The IS was made (Labsphere, customized) after configura-
calibration of the sensitivity of scattering signal detection. tion optimization and tested with a precision polished surface
The configuration optimization of the IS is implemented by of Nd-doped phosphate glass. The surface topography of a
TracePro software. The optical properties of the model, as scratch on the surface is measured in Figs. 7(a) and 7(b).
shown in Fig. 6(a), are first defined as follows: the scattering Then the scratch is inspected by the IS and the scattered light
property of the inner surface of the IS and the baffle is assigned intensity distribution across the scratch is shown in Fig. 7(c). It
to be Lambertian distribution with reflectance of more than is found that the sensitivity of the scattering signal along the
99% at 532 nm. The material of the tested surface is set to beam scanning path still shows nonuniformity to some extent.
be BK-7 glass. The detected area of the PMT (Hamamatsu, This is because the scattering distribution of the polished
H10720) is defined as the perfect absorber. To evaluate the surface is different from the Lambertian type specified in
uniformity of the scattered light intensity along the laser line, the simulation, so that the calibration of the sensitivity of
the surface defects with the same scattering distribution located scattering signal detection should be addressed.
at the different positions along the laser line are simulated. The In the first step, the same location of the scratch is scanned
diameter of the simulated surface defect is the same as that of for 100 times in succession at the same position of the laser line
the laser spot (i.e., 9 μm). The surface defects are equally spaced and the voltage amplitudes of the pulses are measured, as shown
at an interval of 2 mm along the beam scanning path. The sur- in Fig. 8(a). The pulse amplitudes have a variance of 6.3%,
face properties of the defects are defined as Lambertian scatter- which is mainly caused by the random errors of the system
ing for the convenience of the simulation in TracePro, although components discussed in Section 4.B. Therefore, the scattering
the scattering distribution of the surface defect is complicated signal described in the following steps is averaged by 100 mea-
in the practice. One can also import the raw data of the mea- surements for the same location of the scratch at the same
sured scattering distribution of a real defect into the surface
property in TracePro and apply this surface property to all
the simulated surface defects. Then the scattering intensity
for each position along the laser line is recorded by tracing
the rays scattered by each simulated surface defect and absorbed
by the detected area of the PMT, as shown in Fig. 6(b). The
optimization procedure involves a trial-and-error technique of

Fig. 6. Computer simulation result of the IS configuration optimi-

zation. (a) Optical model of the IS configuration. The distance from
the baffle to the photocathode is 52 mm. The diameter of the baffle is
20 mm. The size of the clear aperture of the IS is 100 mm × 10 mm. Fig. 7. (a) Surface topography of a scratch on the tested surface.
The distance from the clear aperture of the IS to the tested surface is (b) Height profile of the scratch along the dotted line indicated in
5.5 mm. (b) Optimization result of the scattered light intensity dis- (a). The width and the depth of the scratch are 41.8 μm and
tribution of defect located along the beam scanning path. The non- 56 nm, respectively. The RMS roughness of the measured area is
uniformity of the scattered light intensity with baffle (solid line) is 3 nm. (c) The scattered light intensity distribution of the tested sur-
9.8% over the length of 60 mm, while that without baffle (dotted line) face. The pulse in the plot is due to the light scattering of the scratch.
is 21%. The background noise is caused by the surface microroughness.
 Research Article Vol. 56, No. 25 / September 1 2017 / Applied Optics 7095

implementation. Other key performance parameters of the sys-

tem, such as background noise, minimum detectable size, and
figure of merit, will be analyzed in this section.
A. Background Noise
As discussed in Section 2, the background noise is mainly
caused by the surface microroughness, the incident angle,
the polarization state of the laser, and the electronic noise.
The incident angle and the polarization state of the laser
are first confirmed by the precision polished surface of the
Nd-doped phosphate glass used in Section 4.C. The scattering
signals in different in-plane scattering angles are measured by
p- and s-polarized laser beams at the Brewster’s angle (i.e., 56°)
and another angle (e.g., 45°), respectively, as shown in Fig. 9(a).
The total scattering signal intensity by a p-polarized laser at an
incident angle of 56° is the lowest in a large range of scattering
angles. Then the background noise is evaluated in the following
Fig. 8. Calibration process of the sensitivity of scattering signal de- cases: (1) without incident laser and ambient light (i.e., elec-
tection. (a) Repetitive inspection result of the pulse amplitude for the tronic noise), (2) with incident laser and without the tested sur-
same location of the scratch. (b) Scattered light intensity for the same face, (3) with incident laser and the tested surface, as shown in
location of the scratch as a function of the position along the beam Fig. 9(b). The RMS electronic noise of the system in case (1) is
scanning path. A total of 31 curves are measured with 2 mm as measured to be 3.7 mV, which is comparable with the PMT
the interval and only seven curves are plotted in different colors for amplifier voltage noise given by 2.2 mV. The RMS noise in
clarification. Each curve is averaged by 100 measurements for the same case (2) is measured only to be 8.3 mV when the laser is turned
location of the scratch at the same position of the laser line. (c) SNR
on, which means the stray light and the ambient light in the
variation as a function of the position along the beam scanning path
with 2 mm as the interval. The dots are the SNR of the scattering optical probe are well controlled. The largest noise mainly
signals and the correction baseline indicated by the solid line is the comes from the surface microroughness and the RMS noise
best fitting of the SNR. (d) Normalization result of the scattered light is measured to be 78.7 mV. Therefore, the defect that the
intensity for the same location of the scratch as a function of the posi- amount of scattered light is smaller than the diffused light
tion along the beam scanning path. The variance of the sensitivity (100 induced by the surface microroughness is not able to be dis-
measurements averaged) is calculated to be less than 3%. covered.
B. Minimum Detectable Size
position of the laser line to improve the stability of the pulse In practical, it is not always easy to apply the SNR as the cri-
amplitude. In the second step, the same location of the scratch terion for defect detectability; rather, the minimum detectable
is inspected as a function of the position along the beam scan- size (MDS) should be used [13]. The MDS indicates the ability
ning path by only moving the tested surface along the Y-axis of of the system to provide positive identification of a given defect
the coordinates, as shown in Fig. 8(b). Obviously, the sensitiv- when subjected to repeated inspection scans. Ideally, a given
ity of the scattering signal for the same location of the scratch is surface defect should always be discovered. Therefore, a calibra-
variant for the different positions along the beam scanning tion standard was fabricated by electron beam exposure and
path. In the third step, the sensitivity variation is plotted for reactive ion beam etching on an ultra-precision polished
each position along the beam scanning path by calculating quartz plate with surface RMS roughness of 1 nm, as shown
the SNR  10 × lg (signal amplitude/signal noise) and a cor- in Fig. 10(a). A series of standard scribed lines with different
rection baseline is obtained by least-square fitting, as shown in widths are etched to serve as simulation of actual defects.
Fig. 8(c). In the final step, the measured scattering signal can be
normalized by the correction baseline such that the sensitivity
along the beam scanning path is almost constant, as shown in
Fig. 8(d). Note that the correction baseline may be changed for
the surfaces with very different materials (e.g., glass and metal),
very different coatings (e.g., antireflection and high-reflection,
dielectric film, and metal film), or very different polishing con-
ditions, and should be recalibrated in accordance with the
aforementioned four steps.
Fig. 9. (a) Scattered light intensities under different polarization
5. PERFORMANCE ANALYSIS states and incident angles. The in-plane scattering angles are measured
from 60° to 170°. (b) Measurement results of the scattered light in-
The effective length of the laser line, the position errors of line tensity (i.e., background noise) for three cases described in the legend.
scan, and the sensitivity normalization of scattering signal Note that the intensities of the incident laser are the same for all the
detection have been analyzed in the part of system measurements.
 7096 Vol. 56, No. 25 / September 1 2017 / Applied Optics Research Article

Table 2. Design and Performance Parameters of the

Proposed Laser Scattering Inspection System
Parameters Value
Laser type 532 nm, TEM00, p-polarized,
polarization ratio of 100∶1
Laser spot 9 μm, circular, flat top
Laser power 10 mW
Incident angle Brewster’s angle
Scanning interval of laser spot 6 μm
Position error of laser spot 3 pixels
Effective length of laser line 60 mm
Scanning speed in Y direction 8 mm/s
Fig. 10. (a) SEM image of the scribed line with a width of 0.5 μm Scanning efficiency 480 mm2 ∕s
on the calibration standard. The RMS roughness of the measured area Data acquisition rate 34.93 MHz
is 1 nm. (b) Scattering image of the scribed lines on the calibration Response time of PMT 0.57 ns
standard. (c) The scattered light intensity distribution of the scribed Velocity of PMS (8 faces) 10k rpm
line with a width of 0.5 μm. RMS electronic noise 3.7 mV
RMS noise with laser on 8.3 mV
MDS <0.5 μm for σ rms  1 nm
Sensitivity nonuniformity of laser line <3%
The linewidths are 0.5 μm, 1 μm, 2 μm, 3 μm, …, and 40 μm, (100 measurements averaged)
respectively. The line depths are the same and measured to be Figure of merit 9.6 cm2 s−1 μm−1
about 100 nm, which is designed to be close to the depth of
defects on optical surfaces in an ICF system. The scattering
image is obtained by the proposed laser scattering system, as contaminants, and particles on the Nd-doped phosphate glass
shown in Fig. 10(b). The linewidth of 0.5 μm can be identified and the cladding glass are clearly discovered.
clearly for 30 measurements and the SNR of the scattering sig-
nal is calculated to be 9.93 dB, as shown in Fig. 10(c).
6. DISCUSSION
Therefore, it can be estimated that the MDS of the system
is less than 0.5 μm for an optical surface with an RMS rough- In the experiment, it is found that the thickness of the trans-
ness of 1 nm. It should be noted that the linewidth shown in parent material has an effect on the scattering image due to the
Fig. 10(c) is much larger than 0.5 μm due to the broadening scattered light collecting configuration. As shown in Fig. 12(a),
effect caused by light diffraction and laser spot size.
C. Figure of Merit
System performance is a trade-off between the MDS and
inspection speed. Decreasing laser spot size to increase light
intensity diminishes area coverage, requiring a concomitant in-
crease in inspection time. Similarly, decreasing shot noise to
enhance SNR increases frequency bandwidth, requiring a
slower scanning speed. Thus, a figure of merit for a surface in-
spection system capable of discovering defects is given by [13]

Inspected area∕time
η
cm2 s−1 μm−1 : (7)
MDS
In this work, the figure of merit is calculated to be
9.6 cm2 s−1 μm−1 in accordance with the design and
performance parameters listed in Table 2.
The defect inspection result for an optical surface of edge
cladding Nd-doped phosphate glass with dimensions of
300 mm × 150 mm × 20 mm is given in Fig. 11(a). The effec-
tive laser line of 60 mm is first scanned forward along the X
direction for a distance of 300 mm at a speed of 8 mm/s
and then scanned backward after the sample is moved by a dis-
tance of 60 mm along the Y direction. Therefore, the inspec-
tion time is only
300∕8 × ceil
150∕60 ≈ 120 s after
consideration of the moving time of the Y stage. The function Fig. 11. Experimental result of the optical surface of edge cladding
ceil
x means to return the nearest integer greater than x. Nd-doped phosphate glass. (a) Scattering image of the optical surface.
Figure 11(b) shows the details of the scattering image indicated (b) Details of the scattering image indicated by the dotted box shown
by the dotted box shown in Fig. 11(a). The scratches, in (a).
 Research Article Vol. 56, No. 25 / September 1 2017 / Applied Optics 7097

To correct the sensitivity nonuniformity of the scattering signal

detection, both computer simulations and experiments are car-
ried out. The incident Gaussian beam is first reshaped into the
flat-top beam with a circular shape on the tested surface. Then
the configuration optimization of the scattered light collector
and the calibration of the sensitivity of scattering signal detec-
tion along the beam scanning path are implemented success-
fully. In addition, the line scanning strategy is described and
the position error of the laser spot is analyzed. The system per-
Fig. 12. (a) Effect of thickness of the transparent material on the formance is experimentally verified and summarized in detail in
scattering image. The local blur regions indicated by arrows are caused
Table 2. It is shown that the system is capable of providing
by the rear surface defects. The thickness of the sample is 4 mm.
(b) Simulation result of SNR of the rear surface defect for different
automatic evaluation for optical surfaces as large as 810 mm ×
sample thickness. 450 mm within 14 min 
810∕8 × ceil
450∕60∕60 with a
MDS less than 0.5 μm. Although the system has a problem
of large FOV, the improvements can be realized by line
confocal detection that will be studied in the future work.
a small portion of the scattered light caused by the defect on the The proposed laser scattering system in this work has potential
rear surface of the transparent material can be detected if the to be employed in fast automatic discovery of surface defects,
thickness is not sufficiently large and thus the scattering image such as scratches, digs, and contaminants, on large aperture
shows a blurred pattern. To investigate the effect of the rear surfaces.
surface defect on the SNR of the top surface defects inspection,
identical defects on both rear and top surfaces of a transparent Funding. Shanghai Institute of Optics and Fine
sample with different thicknesses are simulated. The SNR of Mechanics, Chinese Academy of Sciences (SIOM); National
the scattering signal of the rear surface defect is calculated, Natural Science Foundation of China (NSFC) (50875074).
as shown in Fig. 12(b). It is indicated that the SNR of the rear
surface defect for a sample thickness smaller than 1 mm is com- Acknowledgment. Fruitful discussions with G. Xiao,
parable with that (∼10 dB) of the top surface defect, so that it is Z. Y. Qian, X. Wang, J. Chen, and Z. L. Wu in ZC
difficult to discriminate the rear surface defect from the top Optoelectronic Technologies, Ltd. are gratefully acknowl-
surface defect. With the increase of the sample thickness, edged. K. Z. Ni and S. J. Liu in SIOM are also acknowledged
the SNR of the rear surface defect decreases and the scattering for providing the glass sample and the calibration standard.
signal submerges in the background noise for a sample thick-
ness larger than 10 mm such that the rear surface defect is not
able to be shown in the scattering image [in other words, the REFERENCES
system has a large depth of view (FOV) of 10 mm]. 1. P. A. Baisden, L. J. Atherton, R. A. Hawley, T. A. Land, J. A.
There are probably several ways to eliminate the effect of the Menapace, P. E. Miller, M. J. Runkel, M. L. Spaeth, C. J. Stolz,
rear surface defect. One is the digital image processing algo- T. I. Suratwala, P. J. Wegner, and L. L. Wong, “Large optics for
rithms [22,23] to identify and remove the local blur. This is the national ignition facility,” Fusion Sci. Technol. 69, 295–351 (2016).
2. F. Rainer, R. K. Dickson, R. T. Jennings, J. F. Kimmons, S. M. Maricle,
useful for defects such as scratches, digs, and particles. R. P. Mouser, S. Schwartz, and C. L. Weinzapfel, “Development of
However, the contaminants that show blurred patterns are practical damage mapping and inspection systems,” Proc. SPIE
not able to be discriminated. The other is to select a laser wave- 3492, 556–563 (1999).
length that is absorbed by the material. But this is inconvenient 3. J. Daurios, S. Bouillet, G. Gaborit, and J. C. Poncetta, “Optical met-
in the practical inspection and it is very difficult for materials rology devices for high-power laser large optics,” Proc. SPIE 6616,
661645 (2007).
like quartz glass, which has a wide spectrum transmittance. 4. D. Liu, Y. Y. Yang, L. Wang, Y. M. Zhuo, C. H. Lu, L. M. Yang, and R. J.
Another is by means of line confocal detection that the scat- Li, “Microscopic scattering imaging measurement and digital evalu-
tered light out of the focused laser line is blocked out by ation system of defects for fine optical surface,” Opt. Commun.
the confocal configuration. This approach imposes stringent 278, 240–246 (2007).
5. D. Liu, S. Wang, P. Cao, L. Li, Z. Cheng, X. Gao, and Y. Yang, “Dark-
demands on fabrication of the scattered light collector and over-
field microscopic image stitching method for surface defects evalu-
all mechanical stability of the inspection system, and will be ation of large fine optics,” Opt. Express 21, 5974–5987 (2013).
implemented in our future work. 6. S. Wang, D. Liu, Y. Yang, X. Chen, P. Cao, L. Li, L. Yan, Z. Cheng,
and Y. Shen, “Distortion correction in surface defects evaluating
system of large fine optics,” Opt. Commun. 312, 110–116 (2013).
7. CONCLUSION 7. X. Tao, Z. Zhang, F. Zhang, and D. Xu, “A novel and effective surface
flaw inspection instrument for large-aperture optical elements,” IEEE
A high-speed line scanning system based on the dark-field laser Trans. Instrum. Meas. 64, 2530–2540 (2015).
scattering method has been developed. The theoretical back- 8. L. Dou and M. P. Broderick, “A new technique for automated wafer
ground of defect inspection by laser scattering is introduced inspection and classification of particles and crystalline defects,” in
and the main design parameters related to the improvement IEEE/SEMI Advanced Semiconductor Manufacturing Conference
and Workshop (1997), pp. 180–184.
of the SNR of the scattering signal are discussed and listed 9. N. Marxer, K. P. Gross, H. Altendorfer, and G. Kren, “Process and
in Table 1. The laser scattering system established in accor- assembly for non-destructive surface inspection,” U.S. patent
dance with the design parameters is then described explicitly. 6,271,916 (7 August 2001).
 7098 Vol. 56, No. 25 / September 1 2017 / Applied Optics Research Article

10. G. B. Kim, “A structured mechanism development and experimental scanner for ideal and real-world particles,” J. Electrochem. Soc.
parameter selection of laser scattering for the surface inspection of 140, 1403–1409 (1993).
flatpanel glasses,” Int. J. Prod. Res. 48, 3911–3923 (2010). 18. A. Ungut, G. Grehan, and G. Gouesbet, “Comparisons between geo-
11. A. Yazaki, C. J. Kim, J. Chan, A. Mahjoubfar, K. Goda, M. Watanabe, metrical optics and Lorenz-Mie theory,” Appl. Opt. 20, 2911–2918
and B. Jalali, “Ultrafast dark-field surface inspection with hybrid- (1981).
dispersion laser scanning,” Appl. Phys. Lett. 104, 251106 (2014). 19. B. M. Nebeker and E. D. Hirleman, “Light scattering by particles and
12. M. S. Kim, H. S. Choi, S. H. Lee, and C. J. Kim, “A high-speed particle- defects on surfaces: semiconductor wafer inspection,” in Light
detection in a large area using line-laser light scattering,” Curr. Appl. Scattering from Microstructures (2000), Vol. 534, pp. 237–252.
Phys. 15, 930–937 (2015). 20. D. Angela, F. B. Josep, G. Stefan, N. Gunther, S. Jörg, and M. B.
13. P. Lilienfeld, “Optical detection of particle contamination on surfaces: Jean, “Surface characterization techniques for determining the root-
a review,” Aerosol Sci. Technol. 5, 145–165 (1986). mean-square roughness and power spectral densities of optical
14. H. S. Lee, S. K. Chae, Y. Ye, D. Y. H. Pui, and G. L. Wojcik, components,” Appl. Opt. 41, 154–171 (2002).
“Theoretical and experimental particle size response of wafer surface 21. B. W. Scheer, “Development of a physical haze and microroughness
scanners,” Aerosol Sci. Technol. 14, 177–192 (1991). standard,” Proc. SPIE 2862, 78–95 (1996).
15. K. Takami, “Defect inspection of wafers by laser scattering,” Mater. 22. X. Zhu, S. Cohen, S. Schiller, and P. Milanfar, “Estimating spatially
Sci. Eng. B 44, 181–187 (1997). varying defocus blur from a single image,” IEEE Trans. Image
16. G. L. Wojcik, D. K. Vaughan, and L. K. Galbraith, “Calculation of light Process. 22, 4879–4891 (2013).
scatter from structures of silicon surfaces,” Proc. SPIE 774, 21–31 23. X. X. Zhang, R. G. Wang, X. B. Jiang, W. M. Wang, and W. Gao,
(1987). “Spatially variant defocus blur map estimation and deblurring from
17. B. Y. H. Liu, S. K. Chae, and G. N. Bae, “Sizing accuracy, counting a single image,” J. Visual Commun. Image Represent. 35, 257–
efficiency, lower detection limit and repeatability of a wafer surface 264 (2016).