INSTITUTE OF PHYSICS PUBLISHING MEASUREMENT SCIENCE AND TECHNOLOGY

Meas. Sci. Technol. 15 (2004) 35–43 PII: S0957-0233(04)65357-4

Defect detection in unpolished Si wafers

by digital shearography
Ganesha Udupa1 , B K A Ngoi1 , H C Freddy Goh2 and M N Yusoff1
1
Precision Engineering and Nanotechnology Centre, School of Mechanical and Production
Engineering, Nanyang Technological University, Nanyang Avenue, 639798, Singapore
2
International Semiconductors Products Pte. Ltd, 629013, Singapore

Received 24 June 2003, in final form 9 September 2003, accepted for

publication 16 September 2003
Published 14 October 2003
Online at stacks.iop.org/MST/15/35 (DOI: 10.1088/0957-0233/15/1/005)

Abstract
Defects in silicon wafers have been of great scientific and technological
interest since before the earliest days of the silicon transistor. Recently
much attention has been focused on crystal originated pits on the polished
surface of the wafer. These defects have been shown to contribute to gate
dielectric breakdown. The present work relates to surface and/or subsurface
defect inspection systems for semiconductor industries and particularly to an
inspection system for defects such as swirl defects and groups of particles in
unpolished silicon wafers before the wafer reclamation and/or the wafer
fabrication process using a digital shearography technique. The method
described here relates specifically to semiconductor wafers, but may be
generalized to any other samples. In the present work, surface or subsurface
defects are detected and evaluated by stressing the silicon wafer while
looking for defect-induced anomalies in a fringe pattern, generated by the
interference of two speckle patterns, in the CCD camera and digital image
processing.

Keywords: COPs, digital shearography, semiconductor defect detection

(Some figures in this article are in colour only in the electronic version)

1. Introduction been utilized for the surface defect characterization [1]. How-
ever, in the semiconductor industry, the main challenge lies in
Silicon wafers are widely used in the semiconductor and the characterization of subsurface as well as surface defects.
microelectronics industries. With this material, there is There has been very little research work in the characterization
an immense need to obtain a defect-free highly polished of subsurface defects in Si wafers.
surface for improved yield and performance of the micro- Optical interferometric techniques have been used for non-
components. The current practice in the semiconductor destructive testing (NDT) of objects. Two of these techniques
industry is to inspect the wafers for any surface defects only are electronic speckle pattern interferometry (ESPI) and
at the end of the final polishing stage. At this stage, the speckle shearing interferometry, also known as shearography.
subsurface defects are visible (as they have been exposed by These techniques have been used to detect hidden defects
polishing) as minute spots forming spiral rings or ‘swirls’. in aircraft parts, turbine blades, space vehicles, automobiles
These subsurface defects, which cannot be detected before and many other products [2]. Shearography is a laser
the reclamation process or wafer fabrication process, cause optical method, which is suited for either NDT or for
a high wafer rejection rate at the end of the finishing stage. strain analysis. In contrast to holography, which measures
Unfortunately, there is no instrument currently available to surface displacements, shearography measures derivatives
inspect the ‘prime’/‘test’ wafers at the subsurface level before of surface displacements. Since strains are functions of
wafer fabrication/reclamation. displacement derivatives, shearography allows strains to be
Several techniques such as x-ray, atomic force mi- determined without numerical differentiating displacement
croscopy, scanning tunnelling microscopy, scanning electron data. Defects in objects normally create strain concentrations;
microscopy, and acoustic scanning electron microscopy have it is easier to correlate defects with strain anomalies

0957-0233/04/010035+09$30.00 © 2004 IOP Publishing Ltd Printed in the UK 35

G Udupa et al

using shearography than displacement anomalies applying be a defective wafer. This is because, each time the wafer is
holography. Furthermore, rigid body motions do not produce reclaimed, about 15 µm thickness of wafer will be processed
strain, thus shearography is insensitive to such motions and and inspected for particles.
does not need to adopt any particular device for vibration The study of defects in silicon crystals has been an integral
isolation [2]. part of silicon research activity from the earliest days of the
The bare, ground, lapped or etched wafer surfaces are used silicon transistor. In the mid-1970s, Rozgonyi [5] noted the
to inspect the subsurface defects by digital shearography [3]. importance of suitable diagnostics in detecting and identifying
The qualitative inspection of subsurface defects and the the various types of defects in the crystal. Over the last
principles and method of inspection are described in this paper. two decades, many new diagnostic tools have been developed
The quantitative analysis and serial automatic measurement of and effectively employed. This led the industry to a greatly
small areas are being carried out using a macro focus lens at improved understanding of defects in silicon and resulted in
different locations on the wafer to detect distributed subsurface a significant reduction in yield losses while the number of
wafer particles of micron size. processing steps increased more than eightfold [6].
Typically, a semiconductor wafer may have a very large
2. Inspection in semiconductor wafer manufacturing number of defects, of varying patterns, such as swirl, cluster
or random particles, voids, scratches, cracks or damage,
The Semiconductor Industry Association’s (SIA) International which may have resulted from a great number of causes,
Technology Roadmap for Semiconductors [4] identifies lack such as crystal pulling during crystal growth or improper
of progress in the inspection and characterization of defects control of process parameters during the lapping, etching
and particles on wafers to be a potential barrier to device and polishing processes. Among all of these, the most
miniaturization. The roadmap specifies that by 2005, 30 nm interesting were the ‘swirl’ defects, which were attributed to
particles must be detectable on bare silicon and nonmetallic vacancy clustering such as voids or vacancy-type dislocation
films, 39 nm particles on metallic films and 100 nm particles loops, until their discovery by electron microscope. Swirl
on wafer backsides, for which no solutions currently exist. defects are classified into two types: ‘A’ (larger) and ‘B’
Semiconductor wafer manufacturers already use lasers to (smaller). In 1975, ‘A’ defects were identified as interstitial-
detect particles on expensive silicon wafers, which contain type dislocation loops by electron microscopy, although ‘B’
hundreds of chips. But the manufacturing operation must be defects could not be detected [7]. The practitioners in
shut down while workers try to determine what the particles the production lines prefer abbreviations like ‘COP’ (crystal
are made of and where they came from, especially when large originated particles or pits) or ‘LPD’ (light point defects). The
quantities are found. The source of the particles must be COPs have attracted much interest because they may decrease
eliminated before production can resume. As the features in reliability and manufacturing yield of semiconductor devices.
circuits are getting smaller every 18 months or so, the size of a Recently, COPs have been recognized as surface defects
killer defect is getting smaller and smaller. Because circuits in or micro-pits generated during the crystal ingot growing
new computer chips are only slightly wider than the particles, process and detected by particle counters after surface cleaning
the contaminants are large enough to short-circuit the tiny processes [8].
‘wires’ in the chips. With the need to detect smaller defects, In order to increase the yield in the manufacturing process,
the costs of inspecting wafers are skyrocketing. In order for the said defects are to be detected at an early stage of the
new advances to be implemented in production environments, process as well as controlled during the production process.
improvements in sensitivity must be achieved. In the mid 1980s, surface (visual) defects (scratches, voids,
The processing cost of silicon wafers and the control of particles, masking errors, etc) were considered to have the
defects (at sub-micron size and, especially, those present at greatest impact on semiconductor yield. This led to the
subsurface level) on these wafers are most critical to the wafer development of automatic surface defect detection equipment
fabrication/reclamation industries. It has been reported that and fabrication procedures designed to identify, control and
millions of dollars were lost each year owing to the failure eliminate sources of surface defects. Commercial wafer defect
of detecting these defects in silicon wafers prior to the wafer inspection systems, such as KLA’s Tencor instrument, are
fabrication/reclamation processes. The wafers produced by currently available to surface inspect the defects at the end
the wafer fabrication process are called as ‘prime’ wafers. of the manufacturing process in semiconductor industries.
Wafers which fail to meet certain standards, will be rejected Unfortunately the same level of success has not been achieved
at different stages of the fabrication process. These rejected for subsurface (non-visual) defects. Optical or e-beam
wafers are known as ‘test’ wafers. ‘Test’ wafers are useful techniques that have been used successfully to identify and
for monitoring the operation of the device manufacturing steps remove sources of surface defects cannot be applied to
during trial runs before the start of actual device manufacturing subsurface defects. Some of these defects are generally
using the ‘prime’ wafers. ‘Test’ wafers can be used 5–6 associated with open contacts or vias, gate dielectric defects or
times for trail runs and, prior to each trial, they have to be parametric variation, and residues or voids within the device
processed or reclaimed (stripping and lapping through to fine structures. Without tools to identify, measure and analyse these
polishing). Wafer reclamation is a re-processing technology defects, attempts to eliminate them are limited to trial and
used on rejected wafers during the wafer fabrication. error efforts. As a result, subsurface defects comprise 65%
In wafer reclamation industries, a general rule is that a of all the reasons for yield loss [9]. So it is very important to
wafer with a subsurface defect at a depth of 15 µm and a detect and analyse the subsurface defects in the wafer before
defect size of more than 10 µm in diameter is considered to making them into devices. An attempt has been made here for

36
 Defect detection in unpolished Si wafers by digital shearography

Shearographic image may be mathematically represented

as [10]
Unpolished Silicon Wafers Unpolished Silicon Wafers
I = I0 (1 + µ cos φ) (1)
Polishing Silicon wafers
no yes where I is the intensity distribution of the speckle pattern
(whole batch) Defect?
received at the image plane of the camera, I0 is the intensity of
Polishing
yes no (good wafers only) the laterally sheared images (dc intensity), µ is the amplitude
Defect?
of modulation of the speckle patterns (visibility) and φ is the
READY FOR IC FABRICATION
random phase angle.
RECYCLE RECYCLE
BIN
Significant reduction BIN After the object is deformed, the intensity distribution
of defective wafers &
Cost savings becomes
Patterned Wafer
Conventional process Our process
I
= I0 [1 + µ cos(φ + φ)] (2)
where φ denotes phase change due to surface deformation
Figure 1. The flowchart for in-line metrology of subsurface defect (change in the optical path length of light scattered from two
detection for semiconductor wafer manufacturing/reclamation
industries. neighbouring points). The difference of intensities I and I
is

Id = I
− I = 2I0 [µ sin(φ + φ/2) sin(φ/2)] (3)

the first time to detect subsurface defects in an unpolished Si-
wafer by digital shearography. Figure 1 shows the flowchart where Id manifests itself as a fringe pattern in which dark fringe
for in-line metrology of subsurface defect detection proposed corresponds to
to benefit the semiconductor wafer manufacturing/reclamation
industries. φ = 2nπ with n = 0, 1, 2, 3 . . . .

Bright fringes correspond to φ = (2n + 1)π.

3. Principles of digital shearography
It may be shown that φ is related to the relative
Digital shearography falls in the family of digital speckle displacement (δu, δv, δw) of the two neighbouring points
pattern interferometry (DSPI). Digital shearography is an P(x, y, z) and P(x + δx, y, z), where δx is the amount of
optical interferometric technique that measures surface strain shearing in the x direction, as follows:
concentrations caused by surface and subsurface flaws or
φ = 2π/λ(Aδu + Bδv + Cδw) (4)
defects due to some sort of load, usually either thermal, vacuum
or vibration excitation. In shearography one object point splits where (u, v, w) and (u + δu, v + δv, w + δw) are the
into two in the image plane by a shearing device, thus two displacement vectors of P(x, y, z) and P(x + δx, y, z).
laterally sheared images are observed using CCD camera. The A, B and C are sensitivity factors that are related to the
shearing device may use a Michelson interferometric principle positions of the illumination point Ps (xs , ys , z s ) and the camera
or a double refractive prism. The two laterally sheared images Pc (x0 , y0 , z 0 ), as represented in figure 2, and
interfere with each other producing a random interference
pattern commonly known as a speckle pattern. The pattern A = (x − x0 )/R0 + (x − xs )/Rs
is random, and depends on the characteristics of the surface
of the object. When the object is deformed, by temperature, B = (y − y0 )/R0 + (y − xs )/Rs
pressure or other means, the random interference pattern will
change. The amount of the change depends on the soundness C = (z − y0 )/R0 + (z − xs )/Rs
of the object. A comparison of the random speckle patterns
for the deformed and undeformed states, and their respective R02 = x02 + y02 + z 02 and Rs2 = xs2 + ys2 + z s2 .
fringe patterns, gives information about the structural integrity
of the object. A flaw or defect in the object usually induces a When surface points are considered, φ is related to the
strain concentration which is translated into an anomaly in the displacement derivative through the following:
fringe pattern. The method is called shearography because one
image of the object is laterally displaced, or sheared, relative φ = 2π/λ[Aδu/δx + Bδv/δx + Cδw/δx]δx or
to the other image. (5)
φ = 2π/λ[A∂u/∂x + B∂v/∂x + C∂w/∂x]δx
Digital speckle shearing interferometry or digital
shearography uses a CCD camera and computer image where λ is the wavelength.
processing to produce the fringe anomaly patterns indicative If the beams are confined in x–z plane and θ is the
of the defects in objects. The applicability of shearing angle between the illumination beam and the z-axis (imaging
interferometry to measuring deformations is further enhanced direction), the sensitivity factor B becomes zero and the
using phase shifting (also called phase stepping). Under measurement is insensitive to ∂v/∂x. For measuring the slope
stable conditions, phase shifting interferometers have a higher (first order derivative) of the out-of-plane displacement as
sensitivity than systems without phase shifting. Phase shifting in our experimental configuration, the phase changes due to
interferometers calculate the phase distribution from several surface deformation in this case can be expressed as
interference patterns, which is then displayed on a video
monitor. φ = 2π/λ[(sin θ )∂u/∂x + (1 + cos θ )∂w/∂x]δx. (6)

37
G Udupa et al

Ps (x s, ys, z s) Mirror M1
He-Ne Laser

P (x, y, z)
Pc (x 0, y0, z 0) Spatial filter SF1
P (x+u, y+v, z+w)
X
Before
Loading Deformation Y Collimator
PC
after loading Z

Figure 2. Position of a point on a specimen in relation to

CCD
illumination and camera point before and after the deformation. Camera
Wafer Mount

For normal illumination θ = 0, equation (6) becomes

Zoom
Lens
φ = 4π/λ[∂w/∂x]δx. Mirror M3
Beam
Splitter
If the shear is along y direction, then

φ = 4π/λ[∂w/∂y]δy.
Wafer Mirror M4

4. Description of the wafer defect detection system

Thermal
In the present work, surface and subsurface defects are detected Loading Device Spatial filter SF 2

and evaluated by stressing the silicon wafer while looking

Mirror M2
for defect-induced anomalies in a fringe pattern, generated by
the interference of two speckle patterns, in the CCD camera Figure 3. A schematic diagram of the wafer defect detection system
and digital image processing. Figure 3 shows a schematic sensitive to out-of-plane displacement gradient.
diagram of the measuring system based on an out-of-plane
displacement gradient sensitive configuration. The major parts shearing element. The shearing element allows a coherent
are the illumination source, shearographic head and the image superposition of two laterally displaced images of the wafer
acquisition. The source of light is a 35 mW He–Ne laser at a in the image plane. The lateral displacement is called the
wavelength of 632.8 nm. The shearographic head consists of a shear of the images. The superposition of the two images
CCD camera and the shearing element. The shearing element is called the shearogram, which is an interferogram of an
is an interferometer in Michelson arrangement. A beam object wave with the sheared object wave as a reference wave.
splitter and two adjustable mirrors (M3 and M4), followed Two such interferograms are recorded for different loading
by a zoom lens, image the wafer onto the CCD camera. conditions of the wafer sample. The loading should induce
The direction and amount of shear are altered by tilting the some deformation or alter the deformation state of the surface
mirror M4 through the required angle. The CCD array of the sample. Typical loading methods are thermal, acoustical
has 752 (H) × 582 (V) pixels. A macro video zoom lens or vacuum and could be applied in a static or dynamic way.
(18–108 mm, f 2.5), with the working distance variable In the present work, an infrared lamp is used as the source of
between a maximum of infinity (without close up lens) and a thermal loading. Figure 4 shows the experimental set-up of
minimum of 140 mm (with close up lens), is fixed to the CCD the wafer defect detection system.
camera (2/3

format). The zoom lens can be manually adjusted The absolute difference of two shearograms recorded
for focus and aperture control. With the working distance at for different loading situations of the wafer results in an
about 600 mm, the zoom lens and camera were capable of interference fringe pattern, which is directly correlated to the
recording a field of view that ranged from 198 mm × 264 mm difference in deformation state. Defects inside the wafer may
at the low magnification to 33 mm × 44 mm at the high alter the local surface deformation induced by the loading and
magnification. Using the close up lens, the size of the field result in a disturbance of the (more or less regular) loading
of view is about 6 mm × 8 mm at a working distance of fringes. This allows the detection and classification of defects
140 mm. To view a 200 mm diameter wafer, the camera-to- using the shearographic fringe images.
object distance was about 600 mm. The camera was connected
to a Pentium 4 computer for image acquisition and analysis.
The wafer under test is illuminated by a laser beam through
5. Results and discussion
a collimator and a spatial filter (SF2). A 10×-microscope
5.1. Study of surface defects in Si wafer
objective combined with 25 µm pinhole spatial filter (SF2)
is used to get the required beam expansion. The scattered Before inspecting unpolished wafers for subsurface defects, a
light from the wafer is imaged on a CCD camera through the study was carried out to investigate the nature and existence

38
 Defect detection in unpolished Si wafers by digital shearography

Zoom Lens
Mirror 4
CCD
Mirror 3 Camera

Mirror 1 Cluster Swirl

Defects
Collimator
SF1

Mirror 2

TOP VIEW

ϕ 200 mm

Figure 5. Typical ‘swirl’ defects in a polished Si wafer revealed by

a KLA Tencor instrument.

SF2

He- Ne Laser

B-swirl
Wafer mount

Unpolished
Wafer

IR Lamp A-swirl

Figure 6. Types of swirl defects as seen by scanning electron

microscope.
FRONT VIEW

Figure 4. The experimental set-up of the wafer defect detection pits (see inset figure) whereas the A-swirls are designated as
system. hillocks.
To characterize the defects on the wafer during the
of COPs on the processed wafer surface. First the polished reclamation processes, an optical profiler is used to measure
wafers were inspected using a wafer inspection system, in the surface defects generated either during the wafer processing
this case a KLA Tencor instrument. The processed wafers or during crystal growth or both. Figure 7 shows the results
were subsequently measured using a Wyko optical profiler of 2D and 3D surface topography measurements of a single
to study the defects quantitatively. Figure 5 shows a ‘swirl’ defect on the lapped Si wafer by an optical profiler. The
defect revealed after final polishing as seen by the KLA Tencor three-dimensional representation of the surface topography
instrument. These are micro-defects, located in a spiral pattern provides a clear indication of size, depth and shape of the
in wafers cut perpendicular to the crystal growth direction. The defect. The defects are almost circular (or rectangular) in
wafer defect map obtained by this instrument does not show shape at the surface and tapered down like a cup up to a
whether the defects are of class ‘A’ or ‘B’. The instrument depth of about 205 nm as shown in figure 7. The diameter
shows these defects in the form of black dots on the wafer map. of the defects varies from 5 to 10 µm. The depth and shape
To classify these defects requires a high magnification scanning of the defect change as the process changes from lapping
electron microscope, which shows clearly the defects as in to fine polishing. Figure 8 shows the results of 2D and
figure 6 [11]. The A-swirl defect (the black–white contrasts) 3D surface topography measurement of a single defect on
are larger size defects much smaller in number whereas the the fine polished Si wafer. The defects are irregular (or
B-swirl defects (white dots) are a lot of small defects. Close elliptical) in shape at the surface and taper down like cones
evaluation shows that the B-swirls are designated as shallow or pyramids to a depth of about 5 nm. The size (diameter)

39
G Udupa et al

Table 1. Surface topography parameters of processed Si wafers (measurement area: 225.7 µm × 296.7 µm).
Sl. Parameters/ Ra Rq Rz Rt Defect Defect
no processes (nm) (nm) (nm) (nm) diameter (µm) depth (nm)
1 Lapping 1.92 6.31 221.1 296.5 10–15 200
2 Etching 1.44 1.99 44.61 67.09 5–10 50
3 Stock polishing 1.10 1.32 8.10 11.84 2–8 10
4 Fine polishing 0.85 1.08 7.94 8.58 0.05–5 5

um nm um X Profile
14.6 54.41
-0.00
30.00 -0.02
12.0
10.00 -0.04
10.0 -0.06
-10.00
-0.08
8.0 -30.00 -0.10 Rq 56.93 nm
-50.00 -0.12 Ra 36.63 nm
6.0 -0.14 Rt 205.32 nm
-70.00
4.0 -90.00
-0.16 Rp 7.43 nm
-0.18 Rv -197.89 nm
2.0 -0.20 um
-127.86 0 10 20 30 40 50
0.0
um
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.2

um Y Profile
nm
-0.00
54.4
-0.02
-0.04
-0.06
-0.08
-0.10 Rq 63.90 nm
14.6 Ra 45.88 nm
-0.12
-0.14 Rt 204.32 nm
-127.9 -0.16 Rp 8.31 nm
0.0 -0.18 Rv -196.01 nm
0.0 um
18.2 -0.20
um 0 5 10 15 20 25 30 35 40

Figure 7. The 2D and 3D surface topography map of a single defect on the lapped Si wafer by a Wyko optical profiler indicating average
roughness (Ra ), RMS deviation (Rq ), peak to valley height (Rt ), maximum peak height (Rp ) and maximum valley depth (Rv ) of the surface
profile.

nm X Profile
0.94 nm
0.00
0

-1.00 -1
Rq 1.29 nm
-2.00 -2 Ra 1.03 nm
Rt 4.56 nm
-3 Rp 0.67 nm
-3.00
Rv -3.89 nm
-4 um
-3.86
0 2 4 6 8 10
nm
0.94
Y Profile
nm 0.00 nm
0.50
0.9
0.00
-1.00 -0.50
-1.00
-1.50
-2.00
-3.9 -2.00 Rq 1.24 nm
10.1 0 Ra 0.99 nm
-2.50
-3.00 -3.00 Rt 4.25 nm
0.0 9.0
Rp 0.36 nm
-3.50
um -3.86 Rv -3.89 nm um
-4.00
0 1 2 3 4 5 6 7 8 9

Figure 8. The 2D and 3D surface topography map of a single defect on the polished Si wafer by a Wyko optical profiler indicating average
roughness (Ra ), RMS deviation (Rq ), peak to valley height (Rt ), maximum peak height (Rp ) and maximum valley depth (Rv ) of the surface
profile.
of the defects typically varies from 5 µm to 50 nm at the designated as average roughness (Ra ), RMS deviation (Rq ),
surface. Table 1 shows the surface topography parameters ten-point height (Rz ) and peak to valley height (Rt ) of the
of the processed Si wafers along with defect size during the surface under measurement. As the value of surface roughness
wafer processing stages. The optical profiler gives four surface parameters decreases, there is a considerable decrease in size
roughness parameters, indicative of the surface roughness and and depth of defects in the Si wafer from the lapping to fine

40
 Defect detection in unpolished Si wafers by digital shearography

mm
1.8

ϕ15 1.6
1.4
1.2
1.0
0.8
ϕ20
ϕ2 0.6
ϕ5
0.4
0.2
ϕ10
0.0 mm
0.0 0.5 1.0 1.5 2.0 2.4

(a) (b)

ϕ15

ϕ20

ϕ2
ϕ5

ϕ10

(c) (d)
Figure 9. Demonstration of the measurement range (a) two bonded wafers with simulated subsurface defects, (b) contour map of a 2 mm
defect by a Wyko optical profiler, (c) fringe pattern showing four subsurface defects and (d) fringe pattern showing a 2 mm subsurface
defect.

polishing process. However, these defects affect the final and a program written using the Auto-Pro scripting facilities
performance and decrease reliability and manufacturing yield available in the software to perform the above methods of
of semiconductor devices. Some of these defects or voids subtraction. The real-time subtraction was carried out using
may be originally embedded (during crystal growth) in the the fixed reference frame method or the permanently refreshed
silicon wafer and will be revealed at the surface after the reference frame method [12]. A lateral shear of 10 mm was
polishing process (as seen in figure 5) and result in rejection used throughout the experiment.
of wafers at the end of the final process. The difficult task is The suitability of the measurement range for this
to detect these subsurface defects in unpolished Si wafers and application can be demonstrated using two bonded wafers as
sort them into defective and non-defective wafers. This helps shown in figure 9(a). The two unpolished wafers of 200 mm
not only in reducing scrap but also saving both the processing diameter were bonded at specific spots with various sizes using
and manpower costs associated with producing swirl free non- a steel filled epoxy adhesive. The simulated defects substituted
defective wafers for IC packaging. between the two wafers vary in size from approximately 2
to 20 mm diameter. Figure 9(b) shows the contour map
5.2. Study of subsurface defects in an unpolished Si wafer of a defect with about 2 mm diameter seen using the low
magnification objective lens (2.5×) available in the optical
Defects/flaws in silicon wafers induce strain concentrations on profiler. The diameter of the defects may change a little after
the wafers. Shearography reveals these defects by translating the bonding. The fringe pattern in figure 9(c) successfully
the defect-induced strain concentrations to anomalies in the reveals the location of the four simulated subsurface defects of
fringe patterns. An unpolished wafer of 200 mm diameter size 5, 10, 15 and 20 mm as seen by the bull’s-eye anomaly in
and thickness about 700 µm was clamped along the edges in the fringe pattern. In comparison with figure 9(a), the bull’s-
a wafer mount, leaving 190 mm diameter exposure area for eye corresponds to the positions of the four simulated defects.
laser illumination on one side of the wafer. Thermal loading However the smallest simulated defect, of size 2 mm, was not
was applied using an infrared lamp was placed at its centre on detected when viewing the whole wafer surface. An attempt
other side of the wafer as shown in figure 4. The temperature has been made to detect this defect by reducing the field of
gradient induces thermal stresses in the wafer. Either the view with the zoom lens. Figure 9(d) shows the detection of
double exposure or the real-time method can be used to perform the smallest defect, which shows the position correctly at the
the subtraction. The Image-Pro Plus software along with centre of the wafer. Similar experiments were also conducted
image processing card was installed in a Pentium 4 computer in detecting debonds between two bonded wafers. The present

41
G Udupa et al

Defects
Cluster Swirl
Defects

Area
1000
(a) 100
100

Count
mm2
50
10

1
0.2 0.5 1 2 5 10 20 50 100 500 2000
0.160 9900
Diameter (µm)

Figure 11. The surface defects map after polishing the wafer of
figure 10(a), revealed by a KLA Tencor instrument with a graph
showing defect particle count and the corresponding defect size.

Figure 10(a) shows a typical fringe anomaly pattern for the

subsurface wafer defects. The pattern may vary depending on
the distribution of COPs inside the wafer. Typical flaw/defect
indications depicted in fringe patterns are: (a) bull’s-eyes,
(b) abrupt curvature changes, (c) abrupt fringe density changes
and (d) fringe discontinuity. However in figure 10(b), for non-
defective (good) wafers, the fringe patterns show lines of equal
out-of-plane derivative of displacement due to loading and
(b) there is no abrupt change of curvature and/or discontinuity
of fringes. The results are repetitive, and it shows that the
Figure 10. (a) The subsurface defects in an unpolished Si wafer and technique is capable of differentiating the ‘good’ wafers from
(b) the good unpolished Si wafer.
‘defective’ wafers before the wafer reclamation or fabrication
processes. The defective unpolished wafer of figure 10(a) has
application needs to detect defects within the range of the been sent for polishing to verify the presence of subsurface
thickness of the wafer. This proves that the technique is able defects. After the polishing process, the wafer was inspected
to detect the subsurface defects present 700 µm below the using a KLA Tencor instrument for surface defects. The defect
surface, which is of interest in this application. The minimum
map obtained on the final polished silicon wafer shows defect
size of the subsurface defect that can be detected in a wafer is
sizes varying from sub-micrometre to hundreds and thousands
difficult to say precisely as it depends on several factors such as
of micrometres (up to about 10 mm), as shown in figure 11. The
the shearing amount, the type of the defect, the type of loading
instrument counts the number of particles on the wafer surface
and its condition, and the size of the field of viewing. However,
based on its size and plots it on a graph showing particle count
the study of swirl defects or pits discussed above reveals that
and the corresponding size, as in figure 11.
the size of the subsurface defect in an unpolished Si wafer is
normally about 10 µm. Defects larger than 10 µm may be The technique detects the bigger defect particles or voids
present either individually or in the form of cluster defects as (cluster defects) present inside the wafer in the form of the
seen in figure 5. The cluster defect is a series of swirl defects bull’s-eye as seen in figure 10(a). However the detection of
grouped together to form a bigger defect. The defects in the smaller (tens of micron) individual defect particles could not
outer ring in figure 5 may be considered as a group of cluster be revealed; this may need a higher macro focus zoom lens to
swirl defects as it forms a thick ring about 20 mm wide. In this perform serial measurement on a smaller area or the use of a
case it is easy to detect the defects by digital shearography or lower wavelength light source. Since the location of defects
holography techniques. It is observed that not all the measured in the wafer thickness direction is not known, it is difficult to
wafers are of this type. The defect distribution inside the wafer compare the defects in figure 10(a) with the mapped defects of
varies from circular pattern to a random distribution of swirl figure 11. Further work is needed to quantify the results using
defects from few hundred to few thousand particles or pits. the phase shifting method.

42
 Defect detection in unpolished Si wafers by digital shearography

6. Conclusions size of the defects using a phase shifting technique to make

this system suitable for in situ inspection on the factory floor.
A wafer defect detection system for detecting subsurface This will enhance the system’s capability greatly by providing
defects in an unpolished silicon wafer has been investigated critical information and further assisting in determining ‘good’
based on digital shearography. In the present work, swirl wafers from ‘defective’ wafers before wafer reclamation or
defects (cluster defects) and groups of particles can be detected fabrication processes.
qualitatively by whole field measurement of the wafer surface
in few seconds. The cluster defects and COPs are detected by Acknowledgments
thermally stressing the wafer while looking for defect-induced
anomalies in the fringe pattern. Preliminary tests show that This research work was supported by the Economic De-
about 95% of the results (in a batch of 100 unpolished wafers) velopment Board, Singapore (Grant No COY-15-IDS/I122-
obtained by the system are in agreement with the results 1S99/50890) in collaboration with International Semiconduc-
obtained by the Tencor instrument in terms of detecting cluster tors Products (ISP) Pte. Ltd, Singapore. Authors would like
swirl defects or particles. Since the depths of the subsurface to thank Dr Usha of ISP for her help in the experimental work.
defects are unknown, its difficult to compare the results with
the surface defect results obtained after processing the wafer. References
However, the results obtained are repetitive and hence useful
to sort defective and non-defective unpolished wafers. The [1] Boving K G 1989 NDE Handbook (London: Butterworths) pp
study of surface defects on processed wafers show that most 1–5
[2] Hung Y Y 1989 Shearography: a novel and practical approach
of the defects in nature are pits or voids and the minimum size for non-destructive inspection J. Nondestruct. Eval. 8
of an individual defect in an unpolished wafer is about 10 µm. 55–66
The system can detect cluster swirl defects of size at least [3] Ngoi B K A, Freddy Goh H C, Udupa G and Yusoff M N 2003
5 mm when viewing the whole wafer surface. By reducing System and method for inspection of silicon wafers Patent
the size of field of view, the working distance and controlling pending no 200300562-6, Singapore
[4] The International Technology Roadmap for Semiconductors
the other parameters such as amount of shear, stable loading 2001: Metrology pp 1–26
conditions etc, the system could be able to detect defect sizes (http://public.itrs.net/Files/2001ITRS/Home.htm)
of a hundred to a few hundred micrometres. To evaluate the [5] Rozgonyi G A 1977 Semiconductor Silicon/1977, PV-77-2 ed
size of the subsurface defect needs correct identification of its H R Huff and E Sirtl (Pennington, NJ: The Electrochemical
location and depth in the thickness direction. To investigate the Society) pp 504–8
[6] Fabry L and Matsushita Y 1988 Semiconductor Silicon/1988,
detection of micro-size defects in the wafer needs improvement PV-98-1 ed H R Huff, U Gosele and H Tsuya (Pennington,
in the performance of the system. Future improvements to NJ: The Electrochemical Society) pp 1459–63
the system include replacing the available zoom lens with a [7] Chikawa J-I and Yoshikawa S 1980 Swirl defects in silicon
long working distance microscope to inspect a field of view single crystals Solid State Technol. January 65–70
of 1 mm or less. The disadvantage in this case is that it takes [8] Murray Bullis W 2000 Current trends in silicon defect
technology Mater. Sci. Eng. B 72 93–8
more time to inspect the whole wafer surface and may require a [9] Semiconductor International
local stressing at the point of measurement in order to register http://www.semiconductor.net/semic
any underlying defects. Also a scanning stage is required to [10] Hung Y Y 2000 Digital shearography and applications Trends
scan the wafer when viewing a small area on the surface. If in Optical Non-destructive Testing and Inspection ed
the subsurface defects are greater in number, the detection P K Rastogi and D Inaudi
(Amsterdam: Elsevier) pp 287–308
and identification of individual defects becomes more difficult [11] Monika C H 2003 Characterisation of defects in Si-wafers
and complex. It is observed that some of the wafers are FYP Project Thesis School of Mechanical and Production
transparent to IR radiation causing difficulty in measurement. Engineering, Nanayang Technological University,
Other loading methods such as vacuum or pressure are being Singapore
developed to avoid such problems and to get a more uniform [12] Steinchen W, Kupfer G, Mäckel P and Vössing F 2000 On the
way to experimental modal analysis by means of digital
loading condition. A higher pixel resolution camera may be shearograpy Trends in Optical Non-destructive Testing and
incorporated for increased image definition. Further work is Inspection vol 21, ed P K Rastogi and D Inaudi
being carried out to determine quantitatively the depth and (Amsterdam: Elsevier) pp 309–22

43