International Journal of Machine Tools & Manufacture 66 (2013) 54–65

Contents lists available at SciVerse ScienceDirect

International Journal of Machine Tools & Manufacture

journal homepage: www.elsevier.com/locate/ijmactool

Origin, modeling and suppression of grinding marks in ultra precision

grinding of silicon wafers
F.W. Huo n, R.K. Kang, Z. Li, D.M. Guo
Key Laboratory for Precision and Non-Traditional Machining Technology of Ministry of Education, Dalian University of Technology, Dalian 116024, China

a r t i c l e i n f o a b s t r a c t

Article history: A phenomenon commonly encountered in grinding of silicon wafers is the grinding marks, which are
Received 17 September 2012 difficult to remove by subsequent polishing process, and have been a great obstacle to the manufacture
Received in revised form of silicon wafers with higher flatness. In this paper, the grinding marks formation mechanism was
22 November 2012
clarified, a grinding marks formation model and an angular wavelength model were developed, and a
Accepted 29 November 2012
Available online 10 December 2012
grinding marks suppression method was proposed. A series of grinding experiments were carried out to
verify the developed models and investigate the effect of the wafer rotational speed, the wheel
Keywords: rotational speed, the infeed rate, the axial run out of the cup wheel and the spark out time. The results
Silicon wafer show that: (1) grinding marks are waviness generated on silicon wafers caused by non-uniform
Grinding marks
material removal circumferentially due to the axial run out of the cup wheel; (2) grinding marks
Cup wheel
present multiple angular wavelengths characteristics; (3) the angular wavelength of grinding marks is a
Axial run out
Flatness one-variable function of the rotational speed ratio of the wheel to the wafer; and (4) grinding marks
Waviness could be suppressed significantly by properly selecting the rotational speed ratio.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction
Silicon wafers used as substrates for the production of semi-
conductor devices have to be extremely flat and smooth to meet
the flatness and nanotopography requirements [1,2]. As the
device feature size reduces continuously, the requirements asso-
ciated with these parameters become increasingly stringent [3].
In past ten years, a trend using the rotational grinding method to
produce silicon wafers with high flatness and minimal damage
cost effectively has been highlighted [4,5]. However, grinding
marks are induced in both single side rotational grinding and
double side rotational grinding [6–8]. The flexible polishing pad is
prone to deform according to the shape of the waviness when the
silicon wafer with grinding marks is subjected to subsequent
chemical mechanical polishing process, and they may remain
even after a considerable amount of material removal. Therefore,
grinding marks have significant impact on site flatness and
nanotopography of the final polished silicon wafers, and have
been a great obstacle to the production of ultra flat wafers [6,7].
Additional processes such as slight wet etching, low damage
lapping and plasma etching have been tried to remove them
and improve wafer flatness [9–11]. In order to reduce grinding
marks, Kato et al. proposed to control the wafer rotation speed to
n
Corresponding author. Tel./fax: þ 86 411 84707430.
E-mail address: huofw@dlut.edu.cn (F.W. Huo).
be low enough in the spark out grinding stage; whereas Vepa
et al. suggested varying the wafer rotational speed relative to the
cup wheel rotational speed during grinding [12,13]. Research
related to wafer grinding marks could be retrieved to the 1990s,
when Tonshoff et al. [14] firstly reported the grinding marks in
grinding of silicon wafers, and argued that they are the char-
acteristics for the wafer rotational grinding. Pei et al. [15] studied
in 2002 the effects of grinding process parameters on grinding
marks through a series of grinding experiments. Chidambaram
et al. then developed a mathematical model to predict the locus of
the grinding lines and the distance between two adjacent grind-
ing lines. The relationships between grinding marks and various
process parameters were then discussed according to the devel-
oped model [6,7]. This model can predict grinding line locus well,
but its prediction of line distance is correct only when the
rotational speed ratio of wheel to the wafer is very low. Sun
et al. [16] studied further the effect of the shape of the wafer
chuck on grinding marks in grinding of silicon wafers, and
developed a mathematical model to predict the depth of grinding
marks for different chuck shapes. Recently Li et al. [8,17] inves-
tigated systematically the grinding marks in simultaneous double
side grinding when the rotational speed ratio of the wheel to the
wafer is an integer and developed a mathematical model for
the distance between adjacent grinding marks. They found that
the distance between the adjacent grinding lines increases
and the grinding lines tend to become less curved with the
increase in the ratio. However, this model was not capable of

0890-6955/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijmachtools.2012.11.006
 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65
predicting the grinding marks pattern when the ratio is a non
integer. Li et al. [18] further proposed an extended model for the
grinding marks pattern suitable for both integer and non-integer
ratios. Despite this, the generation mechanism of grinding marks
has not yet been fully clarified, the models developed so far are
not capable of characterizing satisfactorily the angular wave-
lengths of grinding marks and predicting the grinding marks
pattern, little has been written on how grinding parameters affect
grinding marks, and practical measures for suppressing grinding
marks are less than satisfactory. The aim of the present paper is to
rectify these situations. It is divided into seven sections. Section
1 is introduction. The generation mechanism of grinding marks is
clarified in Section 2. A grinding marks formation model and an
angular wavelength model of the grinding marks are developed in
Sections 3 and 4, respectively. Section 5 is the experiments. A
method for suppressing grinding marks is proposed in Section 6.
Finally, conclusions are drawn in Section 7.
2. Formation mechanism of grinding marks
Grinding induced waviness is generally considered to be caused
by grinding vibrations such as forced vibration and self-excited
vibration [19–24]. However, we found that it is the axial run out of
the cup wheel rather than grinding vibrations that causes non-
uniform material removal circumferentially and results in grinding
marks on silicon wafers. In wafer rotational grinding, a segmented
diamond cup wheel with a grinding layer that is only a few
millimeters wide is used to perform an infeed grinding [14,25].
To produce extremely flat surface, it is absolutely necessary that all
the individual segments held on the wheel flange are uniform in
height. However, more or less run out on the end face of the cup
wheel practically always arises. It may arise from a number of sources
such as cup wheel axis offset and tilt, errors in truing and dressing of
the cup wheel, and non-uniform wear of individual wheel segments.
The axial run out of the grinding wheel may not be problematic
for traditional grinding operations, where, the wheel depth of cut
usually is much larger than the run out. However, since the wafer
usually rotates in the range from several tens to several hundreds
of revolutions per minute and the infeed rate is usually in the
range from several to several tens of micrometers, the wheel
depth of cut falls into a range from several tens of nanometers to
several micrometers. Therefore the run out of the cup wheel and
wheel depth of cut have almost the same order of magnitude and
the effect of wheel run out cannot be neglectable [26]. When there
exists axial run out, each of the protruding segments above the
average working surface of the cup wheel will perform an
additional cutting action and generate an extra groove on the
wafer per revolution of the cup wheel, as illustrated in Fig. 1.
A remarkable characteristics of the wafer rotational grinding is
that the kinematics of the cup wheel relative to the wafer is
periodical. This means that any cutting point on the cup wheel, if
neglecting the infeed motion of the cup wheel, will repass the
same cutting path generated just a period before. However, the
previous grooves generated early enough during the infeed
grinding will be completely removed or partially removed and
only a limited number of grooves can survive when the grinding
operation is over. Moreover, these survived grooves are not
evenly distributed circumferentially but aggregate within some
particular regions and result in grooves much wider in most
situations. These large and small grooves form the grinding marks
eventually. For ease of reference, the individual grooves generated
by the most protruding segment are referred to as the elementary
grooves and the wider are referred to as the resultant grooves.
The non-uniformity distribution of the grooves in wafer grinding
is easy to understand by drawing the cutting path of an individual
55
Fig. 1. Illustration of the grinding marks formation in wafer rotational grinding.
cutting point, as shown in Fig. 2, where the curves drawn by black
solid line represent the survived grooves and the dash line curves
represent the motion trajectory of the cutting point. The periodic
motion of the cutting point relative to the wafer and the concen-
tration of the cutting paths are self-evident. When the wheel depth
of cut is larger than the height of the protruding segment only 15
grooves could survive; whereas 90 grooves could survive if the
wheel depth of cut is six times the height of the protruding segment
though other parameters are identical. In wafer rotational grinding,
since only the wafer rotation motion is required for the purpose of
wafer surface generation, all the cutting points of the cup wheel
begin their contact with the wafer at the center region and follow
the same identical forward-curved paths extending radially
outward from the center and they do not cross each other except
at the center region. This non-crossing property enables the shapes
of the grooves to be intact and enables them to overlap each other
to the most.
3. Grinding marks formation model
The practical generated wafer surface is the superposition of
the ideal generated wafer surface without taking grinding marks
into consideration and the survived grooves. Since every rotation
of the cup wheel leaves a groove of some width and depth on the
wafer surface. The geometry of a groove can be represented by
means of a cross-section and a path over the wafer surface. Thus
the grinding marks formation model could be derived by calcu-
lating the Boolean difference of the ideal generated wafer and the
geometry solids representing the grooves generated by the
protruding segment of the cup wheel. Three coordinate systems
are used for the model development, as shown in Fig. 3. Point O
and Point O1 are designated as the wafer and the wheel centers,
respectively. The coordinate system O(X, Y, Z) is fixed on the wafer
center. The coordinate system O1(X2, Y2, Z2) is fixed on the cup
wheel center. We only consider the most protruding segment,
since it produces the deepest grooves and these grooves are the
main composition of the grinding marks. For simplicity, the most
protruding segment was assumed to be a cutting point. Suppose
that the wafer is stationary and the cup wheel rotates about its
own center at a speed of n2 and orbits simultaneously around the
56 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65

Fig. 2. Distributing characteristics of the survived grooves (the rotational speed ratio of the wheel to the wafer is 2383/99. (a) The height of the protruding segment is less
than the wheel depth of cut) and (b) The height is six times the wheel depth of cut).

where j and f are the phase angles of point P and the infeed
rate of the cup wheel, respectively. j is independent of t and
0r j r a.
The equation of the same grinding contact arc in the coordi-
nate system O1(X1, Y1, Z1) can be obtained by rotation transfor-
mation as
2 3 2 32 3 2 3
x1 cosy siny 0 x2 rcosy sinj þ r siny cosj
6y 7 6 76 y 7 6 rsiny sinjrcosy cosj 7
4 1 5 ¼ 4 siny cosy 0 54 2 5 ¼ 4 5
z1 0 0 1 z2 f t
ð3Þ

where y is the rotation angle of the wafer and 2pn1t–2p r

y r2pn1t.
The equation of the ideal wafer surface generated at time t in
the coordinate system O(X, Y, Z) is
2 3 2 3 2 3 2 3
x x1 r cosy r cosy sinj þr siny cosj þ r cosy
6y7 6y 7 6 7 6 7
4 5 ¼ 4 1 5 þ 4 r siny 5 ¼ 4 r siny sinjr cosy cosj þ r siny 5
z z1 0 f t
ð4Þ
Fig. 3. Coordinate systems for model development.
The coordinates of the most protruding segment (point P) in
the coordinate system O1(X2, Y2, Z2) are
2 3 2 3
wafer at the speed of –n1 at a distance of the wheel radius. At the x2 rsinf
start, the wheel center is at point O0, the most protruding 6 y 7 6 rcosf 7
4 25¼4 5 ð5Þ
segment is at point O, and the y2-axis is coincident with the z2 f td
y-axis. It arrives at point P after time t. The origin of the
coordinate system O1(X1, Y1, Z1) is at point O1 and it can be where f is the revolution angle of point P and f ¼2pn2t. d is the
obtained by translating the coordinate system O(X, Y, Z) from height of the most protruding segment above the average end
point O to point O1. The coordinate system O1(X1, Y1, Z1) can be face of the cup wheel.
obtained by rotating the coordinate system O1(X2, Y2, Z2) about The coordinates of point P in the coordinate system O1(X1, Y1, Z1)
the z1-axis. The central angle of the grinding contact arc, a, is are
2 3 2 32 3 2 3
R x1 cosy siny 0 x2 rcosysinf þ rsinycosf
a ¼ arc sin ð1Þ 6y 7 6 76 y 7 6 rsinysinfrcosycosf 7
2r 4 1 5 ¼ 4 siny cosy 0 54 2 5 ¼ 4 5
z1 0 0 1 z2 f td
where R and r are the wafer and the cup wheel radii, respectively.
The ideal generated wafer surface can be described as a helix ð6Þ
surface created by the helical movement of the grinding contact where y is the rotation angle of the wafer at time t and y ¼2pn1t.
arc. The equation of the grinding contact arc in the coordinate The motion trajectory equation of the most protruding
system O1(X2, Y2, Z2) is segment in the coordinate system O(X, Y, Z) is
2 3 2 3 2 3 2 3 2 3 2 3
x2 rsinj x x1 rcosy rcosysinf þ rsinycosf þrcosy
6 y 7 6 rcosj 7 6 7 6y 7 6 7 6 rsinysinfrcosycosf þrsiny 7
4 25¼4 5 ð2Þ 4 y 5 ¼ 4 1 5 þ 4 rsiny 5 ¼ 4 5 ð7Þ
z2 f t z z1 0 f td
 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65
If r, n1, n2, d and f are known, the motion trajectory of the most
protruding segment can be completely determined by Eq. (7).
Since only a portion of the end face of the cup wheel contacts the
wafer surface during grinding, the cutting duration by the most
protruding segment occupies only a small portion in a revolution.
The cutting duration of the most protruding segment during one
rotation of the cup wheel, T0, is determined by
a k¼
T0 ¼
2pn2
The cutting time for the first cutting path is
0 r t rT 0
After the most protruding segment completes the first cutting
path, it will begin the next pass at the commencement of the
second revolution of the cup wheel. It is easy to see that the
cutting time for the ith cutting path is
i1 i1
rtr þ T0
n2 n2
where iZ1
In the ideal wafer grinding process free of grinding marks, all
the individual wheel segments perform an identical depth of cut,
dw, which is the ratio of the infeed rate to the wafer rotational
speed
f
dw ¼
n1
In the case that the axial run out exists, the depth of cut for the
most protruding segment, dp, is
dp ¼ d þ dw
While grinding, both the wafer and the cup wheel rotate at
constant speeds. As long as the cup wheel revolves one revolu-
tion, the most protruding segment will pass the wafer once and
make an elementary groove on the wafer surface. It will repass
the cutting paths generated on the wafer surface made by it
previously after the wheel completes a certain number of revolu-
tions, and hence the motion of the most protruding segment
relative to the wafer is periodic. Here the least period is denoted
as q. Once the cup wheel completes q revolutions, the wafer
will complete p revolutions, and there will be q cutting paths
distributed evenly on the wafer surface circumferentially. p and q
are determined by
p n1
¼ ð13Þ
q n2
where p and q are the smallest integers that hold for Eq. (13).
Suppose that i elementary grooves have been produced on the
wafer surface after time t and denote the geometry of the ith
groove in O(X, Y, Z) coordinate system by p(i). Since d is limited as
compared to the wheel depth of cut, and the grooves generated
previous enough will be erased completely or partially by the
infeed grinding action of the cup wheel, only a limited number of
the grooves could survive, here it is denoted by k. k can be
determined according to the geometry interference principle as
8
< g ðn2 ,n1 Þ
> if d r dw
k ¼ u if d 4 dw and u o q ð14Þ
>
: q if d 4d and u Z q
w
where u ¼ g ðd,dw Þg ðn2 ,n1 Þ and the function g(x1,x2) means the
smallest integer not less than x1/x2.
In wafer rotational grinding, a spark out mode is usually
carried out to improve wafer flatness and reduce surface rough-
ness. In this case, the wheel infeed rate f is equal to 0, and the
grinding marks pattern is different from that in the infeed mode.
In this case, number of the survived grooves can be determined if
57
d rdw as
(
n2 t s if n2 t s o q
k¼ ð15Þ
q if n2 t s Z q
where ts is the spark out duration.
Else
(
v if v oq
ð16Þ
ð8Þ q if v Zq
where v ¼ g ðd,dw Þg ðn2 ,n1 Þg ðn2 ,n1 Þ þ n2 t s
ð9Þ The practical generated wafer surface is the superposition of
the ideal generated wafer surface without taking grinding marks
into consideration and the survived grooves. Define geometrical
solids of the practical generated wafer and the ideal generated
wafer as Op and Oi, respectively. Let function Y(j) represents the
geometry of the jth groove. Op can be obtained by Boolean
difference of Oi and the geometry of all the jth grooves sequen-
ð10Þ tially numbered from j kþ 1 to i as
X
i
Op ¼ Oi  YðjÞ ð17Þ
j ¼ ik þ 1
Since this model takes into account the wafer rotational speed,
the wheel rotational speed, the infeed rate, the axial run out and
the spark out time, it could be applied to the infeed mode with
ð11Þ the wheel depth of cut larger than the height of the most
protruding segment, infeed mode with the wheel depth of cut
smaller than the height of the most protruding segment, incom-
plete spark out mode, and complete spark out mode. Typical
ð12Þ grinding marks patterns on silicon wafers for these four modes
are shown in Fig. 4, where the rotational speeds of the wheel and
the wafer are 2383 rpm and 99 rpm, respectively, and the curves
represent the waviness at a radius of 100 mm in the circumfer-
ential direction. If the wheel depth of cut is larger than the height
of the most protruding segment, the grooves generated before the
previous revolution of the wafer will be removed completely by
the grinding action of the cup wheel and only the grooves
generated in the last revolution are reserved, as shown in
Fig. 4a. In this case, the grinding marks pattern is simple and
intuitive straightforward. However, if the height is larger than the
wheel depth of cut, the grooves generated in the previous
revolutions of the wafer will not be removed completely by the
grinding action of the cup wheel and more grooves will survive.
Those survived grooves may not distribute evenly on the wafer
surface, and their aggregation will result in a series of repeating
grooves much wider, as shown in Fig. 4b. In the spark out mode,
the grinding marks pattern is different from that of the infeed
mode. If the spark out time is not long enough, the final grinding
marks pattern will depend on the infeed rate and the spark out
time, as shown in Fig. 4c. The final wafer surface contains q
identical elementary grooves distributed uniformly along the
peripheral direction, as shown in Fig. 4d.
4. Angular wavelength model of the grinding marks
We first begin by analyzing the circumferential arrangement
characteristics of n cutting paths sequentially generated on a
blank wafer surface until the number of these cutting paths
reaches from 1 to q, and then we resolve the evolution process
of the cutting paths into a certain number of stages according to a
multistage classification method. In each stage, the likelihood of
cutting paths aggregation were calculated as a function of the
rotational speed ratio of the wheel to the wafer. The angular
wavelengths of the grinding marks and the number of the
resultant grooves corresponding to these angular wavelengths
58 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65

Fig. 4. Typical grinding marks patterns on silicon wafers ((a) d ¼50 nm, dw ¼ 100 nm, ts ¼ 0; (b) d ¼ 500 nm, dw ¼100 nm, ts ¼ 0; (c) d ¼50 nm, dw ¼ 100 nm, ts ¼ 5 s
and (d) d ¼ 50 nm, dw ¼ 100 nm, ts ¼60 s).

are therefore derived according to the cutting paths aggregation

index.
As we know, after the most protruding segment of the cup
wheel has made some cutting paths sequentially that is greater
than q/p but no more than q on the wafer surface at a constant
wafer rotational speed and a constant wheel rotational speed, the
wafer surface can be divided into a number of sectors by any two
adjacent cutting paths. In general case, these sectors can be
classified into three categories according to their central angles.
All the sectors are congruent when the number of the cutting
paths is equal to q. However, we found that these sectors could be
classified into two categories in different central angles and the
Fig. 5. Illustration of the primary characteristic sector and the secondary char-
number of the larger is greater than that of the smaller if and only acteristic sector (The radial curves represent the cutting paths. There are 14
if the most protruding segment has exactly made some special primary characteristic sectors and one secondary characteristic sector on the left
numbers of cutting paths on the wafer surface. We define the wafer surface and 15 and 14, respectively, on the right).
larger one the primary characteristic sector, and the smaller the
secondary characteristic sector, as shown in Fig. 5. The evolution sectors is denoted as g1 and the number of secondary character-
process of the cutting paths can be accordingly resolved into a istic sector is denoted as l1 in the first stage. g1 and l1 could be
certain number of stages according to the concept of the primary determined by
characteristic sector and the secondary characteristic sector. g1 ¼ N1 ð19aÞ
The first stage starts from the 1st to the cutting path numbered
by the smallest integer not less than q/p. The start cutting path of l1 ¼ 1 ð19bÞ
each stage follows the finish cutting path of its preceding stage.
The last stage ends up with the qth cutting paths. The number of the sectors obtained by dividing the wafer
P and q satisfy the following equation at the end of the first surface at the end of the first stage, M1, is
stage M1 ¼ N 1 þ1 ð20Þ
N1 p þ a1 ¼ q ð18Þ a1 and p satisfy the following equation at the end of the second
stage
where N1 is a positive integer, a1 is a non-negative integer and is
less than p. N2 a1 þ a2 ¼ p ð21Þ
If a1 ¼0, the (N1 þ1)th cutting path will overlap with the first where N2 is a positive integer and a2 is a non-negative integer and
completely and there is only one stage in the cutting path is less than a1.
evolution process. The number of the cutting paths at the end of the second stage,
If a1 a0, there will be N1 primary characteristic sectors and M1, is given by
one secondary characteristic sector in the first stage and there
will exist the second stage. The number of primary characteristic M2 ¼ N 1 N 2 þN 1 þ1 ð22Þ
 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65 59

Table 1 all the stages are listed in Table 1. After the wheel had rotated
Example of the cutting paths evolution process. 2399 revolutions, 2399 paths were generated and the 2399 paths
distributed evenly on the wafer surface, which divided the wafer
Stage no. 1 2 3 4 5 6 7
surface into 2399 identical sectors and the whole cutting paths
Mi 15 29 44 189 349 1294 2399 evolution process completed. The evolution process of the cutting
Ni 14 1 1 5 1 5 2 paths in this example owns seven stages.
gi 14 15 29 160 189 1105 2399 Except the first and the last stages, the number of secondary
li 1 14 15 29 160 189 0
characteristic sectors at the end of each stage is equal to the
number of primary characteristic sectors of its preceding stage, as
shown in Fig. 6. Ni represents the number of the cutter paths
The number of the primary characteristic sectors g2 and the needed to fill a single primary characteristic sector of the
number of the secondary characteristic sector l2 are as follows proceeding stage. As can be found that Ni is equal to the quotient
g2 ¼ N 1 N 2 þ1 ð23aÞ by dividing the number of primary characteristic sectors of the
(i  1)th stage by the secondary characteristic sector of the
l2 ¼ N 1 ð23bÞ (i  1)th stage and li represents the residue. A larger Ni means
that more cutting paths are needed to partition the primary
If a2 a0, there will exist the third stage. Similarly, if ai  1 a0, characteristic sectors of its preceding. If Ni is larger than 1, the
the corresponding formulas for the ith stage could be defined by cutting paths needed to fill the space of the primary characteristic
Ni ai1 þ ai ¼ ai2 ð24Þ sectors of the proceeding stage will aggregate. The aggregation
may be significant if Ni is large enough and these aggregated
ðMi ki Þai1 þ ki ai ¼ q ð25Þ elementary grooves made by the most protruding segment may
iP
1
overlap and form a single larger groove. Therefore, the likelihood
where ki ¼ ð1Þi þ j1 M j , M0 ¼1, i42. of resulting in significant non-uniform material removal is higher,
j¼0 as is the case in stage 4 and stage 6 in the aforementioned
gi and li in the ith stage are, respectively example. On the contrary, the aggregation does not exist if Ni is
gi ¼ Mi ki ð26aÞ equal to 1, as are the cases in the second, third and fifth stages.
Therefore, Ni is an indicator of the degree of the cutting paths
li ¼ ki ð26bÞ aggregation, and hence it is defined as the cutting path aggrega-
tion index. Ni could be obtained according to Eq. (24). If there is
According to Eq. (24), if ai ¼0, the ith stage is the last stage, and
only one stage in the cutting path evolution process, or if the
q cutting paths will be generated evenly on the wafer surface
cutting path aggregation index in the second stage is larger than
circumferentially
1, the longest angular wavelength of the grinding marks on the
Mi ¼ q ð27Þ wafer surface, P2, is
The number of the primary characteristic sectors gi and the 2p
P2 ¼ ð29Þ
number of the secondary characteristic sector li in the last stage q
are
The angular wavelength of grinding marks due to a larger
gi ¼ q ð28aÞ aggregation index in the third stage, P3, is
2pp
li ¼ 0 ð28bÞ P3 ¼ ð30Þ
q
The evolution process of the cutting paths owns one or more
Similarly, the angular wavelength of grinding marks due to a
evolution stages and this depends on the rotational speed ratio of
larger aggregation index in the ith stage, Pi, is determined by
the wheel and the wafer. According to the equations defined
above, it is easy to determine the number of the stages of the 2pai3
Pi ¼ ð31Þ
cutting paths evolution process and calculate ai, Ni, Mi, gi and li at q
any stage for any given rotational speed ratio. To better under-
where i Z4and the last stage is exclusive.
stand this model, a calculated example of the cutting path
The angular wavelength of grinding marks due to a larger
evolution process according to the above equations is listed in
aggregation index in the last stage, P, is determined by
Table 1 and its cutting path evolution process is plotted in Fig. 6.
The wafer and the wheel rotational speeds in this example are 2p
P¼ ð32Þ
165 rpm and 2399 rpm, respectively. p and q are equal to 165 and q
2399, respectively, according to Eq. (7). It is not until the most The number of the resultant grooves due to a larger aggrega-
protruding segment completes 2399 passes that those generated tion index in the ith stage, Qi, is
paths can be distributed evenly on the wafer surface, as shown in
Q i ¼ M i1 ki1 ð33Þ
Fig. 6j. Fig. 6a shows the first cutting path generated on the wafer
surface. At the end of the first stage, the wheel rotated 15 where k0 ¼ 0
revolutions, and 15 grooves were generated, which divided the It can be seen from Fig. 6 that the cutting paths first start to
whole wafer surface into 14 primary characteristic sectors and a aggregate in the fourth stage and the number of the resultant
secondary characteristic sector, as shown in Fig. 6b. At the end of grooves is equal to 29, and then the cutting paths aggregate in the
the second stage, the wheel rotated 29 revolutions, and 29 paths sixth stage with a number of 189 resultant grooves. The number
were generated on the wafer surface, which divided the wafer of the resultant grooves is in full agreement with the calculated
into 15 primary characteristic sectors and 14 secondary charac- results according to Eq. (24). The proposed angular wavelength
teristic sectors, as shown in Fig. 6c. After the cup wheel rotated 44 model shows that the angular wavelength of the grinding marks
revolutions, 44 cutting paths were generated, which divided the is an one-variable function of the rotational speed ratio of the
wafer surface into 29 primary characteristic sectors and 15 wheel and the wafer, and is independent of the wheel infeed rate,
secondary characteristic sectors. The values of Mi, Ni, gi and li of the height of the protruding portion relative to the end face of the
60 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65

Fig. 6. Example of the cutting paths evolution process (The number in the brackets represents the number of the cutting paths. (a) Stage 1 (1); (b) Stage 1 (2); (c) Stage 2
(15); (d) Stage 3 (29); (e) Stage 4 (44); (f) Stage 4 (87); (g) Stage 4 (189); (h) Stage 5 (349); (i) Stage 6 (538); (j) Stage 6 (1294); (k) Stage 7 (1512) and (l) Stage 7 (2399)).
 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65
cup wheel and the spark out time. However, it should be pointed
out that it is also possible that the infeed motion disenables the
late stages to develop fully, even though there may exist many
stages, since only limited number of elementary grooves can
survive due to the wheel infeed motion.
5. Experimental
A number of grinding experiments were performed to verify
the grinding marks formation model, the characteristic wave-
length model, and investigate the effect of the wafer rotational
speed, the wheel rotational speed, the infeed rate, the axial run
out and the spark out time. The experiments were conducted on
an Okamoto VG401MK wafer grinder using a mesh #600 resin
bond diamond cup wheel of 350 mm in diameter and 3 mm in
rim width. 200 mm (100) p-type silicon wafers were used for this
investigation. Ultra-pure water was used as the coolant. To reduce
the possible deformation induced by vacuum chucking, silicon
wafers was adhered on a stainless steel solid plate by wax and
then the plate was fixed on the rotary table, as shown in Fig. 7.
The stock removal amount was 50 mm for all grinding experi-
ments to assure that the previous grinding marks were ground
away completely. Photos of the ground wafers were taken by
using an annular light to present the tiny grooves more clearly.
Some of the ground wafers were measured with a Tropel Flat-
master 200 surface form measurement to compare the various
grinding marks patterns. It should perhaps be pointed out that
high precision and resolution spindle speed control is not
Fig. 7. Grinding experiment setup.
Fig. 8. Influence of the axial run out of the cup wheel on the characteristic angular wavelengths of grinding marks ((a) before truing and (b) after truing).
61
required for common wafer grinding machines and the wafer
and the wheel rotation speeds decrease as the grinding load
increases. The actual rotation speed was measured with a Sampo
DT-6234B digital tachometer.
The first series of experiments were performed to verify that
grinding marks are induced by the axial run out of the cup wheel.
All the grinding conditions except the axial run out were just the
same to avoid the influence of other factors. The wheel and the
wafer rotational speeds were set at 2000 rpm and 50 rpm,
respectively. The infeed rate was 10 mm/min and no spark out
grinding was used. The cup wheel was trued using a cast iron
lapping plate with #360 silicon carbide abrasives to obtain a
lower run out. Fig. 8a and b show the measurement results of two
silicon wafers ground with the same cup wheel before and after
truing. It can be seen that the grinding marks generated on these
two wafers are just different in the amplitude of the waviness and
increasing the axial run out leads to an increase of the depth and
the number of the grooves. The angular wavelengths of the
grinding marks on the both wafer surfaces are identical. This
indicates that grinding marks are induced by the axial run out of
the cup wheel, rather than grinding vibrations.
The second series of experiments were performed to verify the
grinding marks pattern model and the angular wavelength model.
Four silicon wafers were ground sequentially to assure that the
axial run out value of the cup wheel is constant. The wafer and
the wheel rotational speeds were set at 143 rpm and 2295 rpm,
respectively. The infeed rate and the spark out time, as well as the
wafer rotational speed and the wheel rotational speeds measured
at the end of each grinding operation, are listed in Table 2.
The cutting paths evolution processes for the four wafers are
listed in Table 3 and the photos of the four ground wafers are
shown in Fig. 9.
Though, the waviness with too short angular wavelength are
invisible, at least three groups of waviness with different angular
wavelengths on the first three wafers and at least two groups of
waviness with different angular wavelengths on the last wafer
Table 2
Grinding process parameters.
Wafer no. n1 (rpm) n2 (rpm) f (mm/min) ts (min)
1 2279 141.2 50 0
2 2286 141.8 3 0
3 2288 142.1 3 2
4 2290 142.7 3 20
62 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65

can be distinguished by naked eyes. Hence their numbers of each According to the angular wavelength model of the grinding
group can be counted and their central angles can be measured marks, the cutting path aggregation indexes of the second stage in
accurately. This indicates that grinding marks on silicon wafers the cutting path evolution processes for all the four wafers are 7,
present multiple angular wavelengths characteristics. Another 8, 9 and 19, respectively. The model predicts that a total of 7, 8,
characteristics of grinding marks is that the angular wavelength 9 and 19 identical grooves can form within a primary character-
of the waviness component is positively correlated with its istic sector of the first stage on the four wafer surfaces, respec-
amplitude. Hence, the longest angular wavelength component is tively. However, only the latter 3 resultant grooves survived on
the core of grinding marks. the first wafer surface since the other 4 was removed due to the
The longest angular wavelength of the grinding marks for the infeed motion of the cup wheel. The number of the survived
four wafers was measured to be 14p/113, 16p/129, 18p/145 and grooves increased with decreasing the infeed rate. When the
38p/305, respectively, and the second longest angular wavelength infeed rate decreased from 30 to 3 mm/min, the number of the
are 2p/113, 2p/129, 2p/145 and 2p/305, respectively. These survived grooves increased from 3 to 5, as shown in Fig. 10a and b.
observations are in complete accord with the characteristic Furthermore, the number of the survived grooves increased if a
wavelength model of the grinding marks. Hence this model is spark out grinding was implemented. All the predicted grooves
correct. It should be pointed out that the measured angular could survive after a spark out grinding with 60 s, as shown in
wavelengths of the grooves with the third longest angular Fig. 10c. This proves that the grinding marks pattern is a function
wavelength are not in complete accord with the predicted by of the rotational speed ratio, the run out, the infeed rate and the
the proposed model. Since the rotational speed of the wheel and spark out time. Since the run out of the cup wheel is on the micron
wafer varied during grinding the accuracy of measurements of the level to the submicron level and high-precision measurement of
speed was not high enough. the run out is extremely difficult, we cannot verify the grinding
marks pattern model quantitatively. Despite this, however, these
observations are found in good agreement with this theoretical
model. A longer spark out time was beneficial to weaken the
Table 3 grinding marks. However, it can be found that the waviness with
Cutting paths evolution process for the four wafers.
the longest angular wavelength still existed even after a rather
Wafer no. q p N1 N2 N3 N4 N5 N6 N7 N8 N9 long spark out grinding up to 20 min, though the amplitude of the
grinding marks decreased, as shown in Fig. 10d. This shows that
1 11395 706 16 7 7 1 1 1 1 2 – the effect of suppressing grinding marks by means of prolonging
2 11430 709 16 8 4 10 2 – – – – spark out time may be unsatisfactory.
3 22880 1421 16 9 1 6 1 1 2 1 2
4 22900 1427 16 19 1 67 – – – – –
The third series of experiments were performed to investigate
the effect of the wafer and the cup wheel rotational speeds on

Fig. 9. Photos of the ground silicon wafers with grinding marks ((a) f¼ 50 mm/min, ts ¼ 0 min; (b) f ¼3 mm/min, ts ¼0 min; (c) f ¼ 3 mm/min, ts ¼2 min and (d) f ¼3 mm/min,
ts ¼ 20 min).
 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65 63

grinding marks. Eight silicon wafers were ground sequentially to change in the angular wavelengths of the grinding marks. Since
assure that the axial run out was constant. The wheel rotational the rotational speed resolution of the motor driving the wheel
speeds were set at 2295 rpm and the wafer rotational speed was spindle is about 10 revolutions per minute, we can not change the
set at 147, 148, 149, 150, 151 and 152 rpm, respectively, to wheel rotational speed with an increment down to one revolution
investigate the effect of the wafer rotational speeds on grinding per minute. However, it can be perceived that increasing the
marks. Then the wafer rotational speeds was set at 148 rpm and wafer rotational speed by one revolution per minute will lead to
the wheel rotational speed was set at 2304 and 2314 rpm, an insignificant change in the angular wavelengths.
respectively, to investigate the effect of the cup wheel rotational
speeds on grinding marks. The infeed rate was 30 mm/min and no
spark out grinding was used. It can be seen that the wafer
rotational speed has a dramatic influence on the angular wave- Table 4
lengths of grinding marks. Increasing the wafer rotational speed Example of the cutting paths evolution process.
by one revolution per minute led to a significant change in the
Stage no. 1 2 3 4 5 6 7 8 9 10 11
resultant angular wavelengths. It can also be seen from Fig. 10b, g
and h that the wheel rotational speed has an influence on the Ni 15 1 1 1 1 1 1 1 1 1 2
angular wavelengths of the grinding marks. Increasing the wafer Mi 16 33 50 83 133 216 349 565 914 1479 2393
rotational speed by ten revolutions per minute led to an obvious

Fig. 10. Effect of the wafer and the wheel rotational speeds on grinding marks ((a) n1 ¼147 rpm, n2 ¼2295 rpm; (b) n1 ¼ 148 rpm, n2 ¼ 2295 rpm; (c) n1 ¼ 149 rpm,
n2 ¼ 2295 rpm; (d) n1 ¼ 150 rpm, n2 ¼2295 rpm; (e) n1 ¼ 151 rpm, n2 ¼ 2295 rpm; (f) n1 ¼152 rpm, n2 ¼ 2295 rpm; (g) n1 ¼ 148 rpm, n2 ¼2314 rpm and (h) n1 ¼ 148 rpm,
n2 ¼ 2304 rpm).
64 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65

6. Suppressing grinding marks by reducing the cutting paths
aggregation
The angular wavelength model of the grinding marks developed
in Section 4 clearly indicates it is the aggregation the elementary
grooves produced by the protruding portion of the cup wheel that
leads to a series of wider grooves and forms grinding marks on the
wafer surfaces, and the longest wavelength component of the
grinding marks is formed in the stage in which the aggregation of
the elementary grooves occurs firstly. Therefore grinding marks

Fig. 11. Example of the cutting paths evolution process on the wafer surface (The number within brackets represents the number of the cutting paths. (a) Stage 1 (1);
(b) Stage 1 (16); (c) Stage 2 (33); (d) Stage 3 (50); (e) Stage 4 (83); (f) Stage 5 (133); (g) Stage 6 (216); (h) Stage 7 (349); (i) Stage 8 (565); (j) Stage 9 (914); (k) Stage 10
(1479) and (l) Stage 11 (2393)).
 F.W. Huo et al. / International Journal of Machine Tools & Manufacture 66 (2013) 54–65
could be suppressed if this stage could be postponed by control-
ling the cutting path aggregation index to be equal to 1. It can be
achieved by properly selecting the rotational speed ratio of the
wheel to the wafer. Since the number of the cutting passes needed
to complete a period is equal to q, if q is too small, the number of
the cutting paths evolution process may not be large enough and
the aggregation of the cutting paths may be remarkable. Therefore,
q should not be too small. The ideal conditions for suppressing
grinding marks by reducing the cutting paths aggregation can be
expressed as
q- þ 1
Ni ¼ 1
where i¼2, 3y.
A calculated example is shown in Table 4, where the wafer
rotational speed is 144 rpm and the wheel rotational speed is
2393 rpm. We could find that q and p are 2393 and 144,
respectively, according to Eq. (13) and there are 11 stages in the
cutting paths evolution process. Except the first and the last
stages, the cutting paths concentration index is equal to 1.
The cutting paths evolution process is shown in Fig. 11. It can
be seen that the elementary grooves produced by the protruding
portion of the cup wheel are distributed evenly on the wafer
surface all the time. Hence the aggregation the elementary
grooves produced by the protruding portion of the cup wheel
can be reduced to a minimum, the grinding marks could be
decreased dramatically and a remarkable improvement on the
wafer flatness could be achieved, although the axial run out of the
cup wheel remains the same. Accurate spindle speed control is
not available for today’s wafer grinding machines, and hence the
rotational speed ratio of the wheel to the wafer cannot be
positively selected and controlled yet. However, it is worthwhile
taking this into account at the grinding machine design stage.
7. Conclusions
(1) Grinding marks are waviness generated on silicon wafers
caused by non-uniform material removal circumferentially
due to the axial run out of the cup wheel.
(2) Grinding marks on silicon wafers present multiple angular
wavelengths characteristics.
(3) The grinding marks pattern is a function of the rotational
speed ratio, the axial run out of the cup wheel, the infeed rate
and the spark out time.
(4) The angular wavelength of the grinding marks is a one-
variable function of the rotational speed ratio of the wheel
to the wafer.
(5) A method for suppressing grinding marks by reducing the
cutting path aggregation was proposed. It can be achieved by
properly selecting the rotational speed ratio of the wheel to
the wafer.
Acknowledgments
The financial support for this research by the National Natural
Science Foundation of China (No.50975040) is greatly appreciated.
65
References
[1] Hideki Tsuya, Present status and prospect of Si wafers for ultra large scale
integration, Japan Journal of Applied Physics 43 (7A) (2004) 4–18.
[2] Mike Seacrist, Silicon starting materials for sub-65 nm technology nodes, in:
9th International Symposium on Wafer Cleaning and Surface Preparation,
June 8–9, 2005, Boise, Idaho, USA.
[3] International Technology Roadmap for Semiconductors 2009 Edition, Front
End Processes, /http://www.itrs.netS.
[4] P.S. Sreejith, G. Udupa, Y.B.M. Noor, et al., Recent advances in machining of
silicon wafers for semiconductor applications, International Journal of
Advanced Manufacturing Technology 17 (3) (2001) 157–162.
[5] Z.j. Pei, A study on surface grinding of 300 mm silicon wafers, International
ð34aÞ
Journal of Machine Tools and Manufacture 42 (3) (2002) 385–393.
[6] Z.J. Pei, Alan Strasbaugh, Fine grinding of silicon wafers: grinding marks, in:
ð34bÞ Proceedings of IMECE2002, ASME International Mechanical Engineering
Congress & Exposition, November 17–22, 2002, New Orleans, Louisiana, USA.
[7] S. Chidambaram, Z.J. Pei, S. Kassir, Fine grinding of silicon wafers: a
mathematical model for grinding marks, International Journal of Machine
Tools and Manufacture 43 (15) (2003) 1595–1602.
[8] Z.C. Li, Z.J. Pei, Graham R. Fisher, Simultaneous double side grinding of silicon
wafers: a literature review, International Journal of Machine Tools and
Manufacture 46 (12-13) (2006) 1449–1458.
[9] Hashii, Tomohiro, Semiconductor wafer grinding method, US Patent 6969302,
2005.
[10] Nozawa Kimio, Maeda Takeshi, Murakami, Katsuyoshi, et al., Method for
manufacturing semiconductor wafer, Japanese Patent JP2003229385, 2003.
[11] Michihiko Yanagisawa, Tadayoshi Okuya. Nanotopography removing method.
US Patent 6875701, 2005.
[12] Tadahiro Kato, Fukushima,Oshima Hisashi,et al., Processing method for a
wafer, US Patent 6358117, 2002.
[13] Vepa Krishna, Dobson Duncan. Systems and methods to significantly reduce
the grinding marks in surface grinding of semiconductor wafers, US Patent
Application Publication, US 2002/052169, May 5, 2002.
[14] H.K. Tonshoff, W.V. Schmieden, I. Inasaki, W. König, G. Spur, Abrasive
machining of silicon, Annals of the CIRP 39 (2) (1990) 621–635.
[15] Z.J. Pei, Alan Strasbaugh, Fine grinding of silicon wafers: designed experi-
ments, International Journal of Machine Tools and Manufacture 42 (3) (2002)
395–404.
[16] W.P. Sun, Z.J. Pei, G.R. Fisher, Fine grinding of silicon wafers: effects of chuck
shape on grinding marks, International Journal of Machine Tools and
Manufacture 45 (6) (2005) 673–686.
[17] Z.C. Li, Z.J. Pei, Simultaneous double side grinding of silicon wafers: a
mathematical study on grinding marks, International Journal of Abrasive
Technology 1 (3–4) (2008) 287–300.
[18] Zhichao Li, Bin Lin, Wangping Sun, Xiaohong Zhang, Simultaneous double
side grinding of silicon wafers: a further investigation into grinding marks
pattern, ECS Transactions 27 (1) (2010) 395–400.
[19] S. Malkin, Grinding Technology Theory and Applications of Machining with
Abrasives, Ellis Horwood, Chichester, UK, 1989.
[20] Ioan D. Marinescu, Mike Hitchiner, Eckart Uhlmann, W. Brian Rowe,
Ichiro Inasaki, Handbook of Machining with Grinding Wheels, CRC Press,
2007.
[21] Mohammed Alfares, Abdallah Elsharkawy, Effect of grinding forces on the
vibration of grinding machine spindle system, International Journal of
Machine Tools and Manufacture 40 (14) (2000) 2003–2030.
[22] S. Matsui, An experimental study on the grinding of silicon wafers, Bulletin
of the Japan Society of Precision Engineers 22 (4) (1988) 295–300.
[23] Hongqi Li, Yung C. Shin, Wheel regenerative chatter of surface grinding,
Journal
of Manufacturing Science and Engineering 128 (2) (2006) 393–403.
[24] I. Inasaki, B. Karpuschewsk, H.S. Lee, et al., Grinding chatter – origin and
suppression, Annals of the CIRP 50 (2) (2001) 515–534.
[25] G.J. Trmal, F. Holesovsky, Wave-shift and its effect on surface quality in
super-abrasive grinding, International Journal of Machine Tools and Manu-
facture 41 (7) (2002) 797–989.
[26] Thomas R.A. Pearce, David C. Fricker, Alan Speight, Indirect measurement of
grinding wheel run-out using acoustic emission, International Journal of
Manufacturing Technology and Management 12 (1–3) (2007) 139–154.