International Journal of Machine Tools & Manufacture 42 (2002) 395–404

Fine grinding of silicon wafers: designed experiments

a,* b
Z.J. Pei , Alan Strasbaugh
a
Department of Industrial and Manufacturing Systems Engineering, Kansas State University, Manhattan, KS 66506, USA
b
Strasbaugh, Inc., San Luis Obispo, CA 93401, USA

Received 2 November 2000; received in revised form 31 July 2001; accepted 2 August 2001

Abstract

Silicon wafers are the most widely used substrates for semiconductors. The falling price of silicon wafers has created tremendous
pressure to develop cost-effective processes to manufacture silicon wafers. Fine grinding possesses great potential to reduce the
overall cost for manufacturing silicon wafers. The uniqueness and the special requirements of fine grinding have been discussed
in a paper published earlier in this journal. As a follow-up, this paper presents the results of a designed experimental investigation
into fine grinding of silicon wafers. In this investigation, a three-variable two-level full factorial design is employed to reveal the
main effects as well as the interaction effects of three process parameters (wheel rotational speed, chuck rotational speed and feed-
rate). The process outputs studied include grinding force, spindle motor current, cycle time, surface roughness and grinding marks.
 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Ceramic machining; Grinding; Grinding force; Grinding marks; Material removal; Semiconductor materials; Silicon wafers; Surface
roughness

1. Introduction

Most IC (integrated circuit) chips are built on single

crystal silicon wafers. These IC chips can be found in
every type of microelectronic applications, including
networking and computing (routers, modems, set-top
boxes, Ethernet cards, disk drives), wireless communi-
cations (portable electronic devices, cellular phones,
pagers, satellite receivers), consumer electronics (DVD
players, home security systems, small household
appliances, smart cards), automotive electronics (GPS
and navigational tools, air bag controls, anti-locking
braking systems), industrial automation and control sys-
tems.
However, in recent years the price of silicon wafers
has dropped significantly. The huge price erosion can be
seen from Fig. 1. The worldwide revenue generated by
silicon wafers in 1999 was US$5.8 billion, a 4% increase
from the revenue of 1998 but with 26% more silicon
produced [1]. The falling price of silicon wafers has

* Corresponding author. Tel.: +1-785-532-3436; fax: +1-785-532- Fig. 1. Worldwide revenue and area production of silicon wafers
3738. (after Mozer [1]).
E-mail address: [email protected] (Z.J. Pei).

0890-6955/02/$ - see front matter  2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 8 9 0 - 6 9 5 5 ( 0 1 ) 0 0 1 2 3 - 7
396 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404

applied a great pressure on silicon manufacturers to

reduce their manufacturing cost. It is critically important
to develop new manufacturing processes, or to develop
new applications for some existing processes, that allow
manufacturing silicon wafers more cost-effectively.
The manufacturing processes for silicon based ICs are
illustrated in Fig. 2. As can be seen, surface grinding
has been (or can be) used at three different manufactur-
ing steps in the manufacturing flow: (a) surface grinding
after slicing (wire sawing) as partial replacement of lap-
ping; (b) fine grinding after etching as partial replace-

Fig. 3. Illustration of wafer surface grinding.

ment of rough polishing; and (c) back-grinding the back

side of the wafer after circuits are developed on the front
side. Here, (a) and (b) take place inside silicon wafer
manufacturers while (c) takes place inside IC manufac-
turers or their outside contractors.
Due to its importance, surface grinding has attracted
more and more interest among investigators. Pei and
Strasbaugh [6] have given a brief summary of reported
investigations into surface grinding of silicon wafers.
Fine grinding of etched wafers first appeared in public
domain through the US patent by Vandamme et al. [5].
The advantages of fine grinding of etched wafers are
two-fold. One is to improve the flatness of etched wafers.
Another is to reduce the removal amount for rough pol-
ishing by 25–50% [5]. The end result will be higher
throughput for rough polishing and better flatness for
polished wafers.
Another application of fine grinding is to ‘re-work’
the background device wafers. Back grinding is normally
done by two steps: rough grinding by grinding wheels
with large size of diamond grains; and fine grinding by
grinding wheels with small (fine) diamond size. How-
ever, sometimes there is a need to re-grind the wafers
that have been background previously. And this re-
grinding is typically done by fine grinding only.
Fine grinding of silicon wafers requires using #2000
mesh (3–6 µm grit size) or finer diamond wheels. The
surfaces to be fine ground generally have no damage or
very little damage and the surface roughness is ⬍30 nm
in Ra [6].
The uniqueness and the special requirements of silicon
wafer fine grinding process were discussed in the pre-
Fig. 2. Manufacturing processes for silicon based ICs (after Bawa et vious paper [6]. The major requirements for fine grinding
al. [2], Fukami et al. [3], Tonshoff et al. [4] and Vandamme et al. [5]). of silicon wafers include:
 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404 397

Table 1 Table 2
Test matrix Variable levels

Test Wheel speed Chuck speed Feed-rate Variables Unit Low level (⫺) High level (+)

1 ⫺ ⫺ ⫺ Wheel speed Rev s⫺1 36.25 72.50

2 + ⫺ ⫺ (rpm) (2175) (4350)
3 ⫺ + ⫺ Chuck speed Rev s⫺1 0.67 9.83
4 + + ⫺ (rpm) (40) (590)
5 ⫺ ⫺ + Feed-rate µm s⫺1 0.1 0.3
6 + ⫺ +
7 ⫺ + +
8 + + +

Table 3
Grinding force data

Wheel speed Chuck speed Feed-rate Maximum grinding force (N)

Wafer 1 Wafer 2 Wafer 3

⫺ ⫺ ⫺ 62.3 62.3 62.3

+ ⫺ ⫺ 71.2 62.3 71.2
⫺ + ⫺ 71.2 66.7 71.2
+ + ⫺ 62.3 66.7 66.7
⫺ ⫺ + 75.6 80.1 80.1
+ ⫺ + 84.5 84.5 84.5
⫺ + + 80.1 66.7 66.7
+ + + 84.5 84.5 89.0

Fig. 4. Grinding marks under Magic Mirror pictures.

1. The grinding wheel should have self dressing ability; reported and discussed preliminary experimental work
2. The grinding wheel should have a reasonable life; on the effects of grinding wheels, process parameters
3. The grinding force should be low and constant; and grinding coolant. As a follow-up, this paper reports
4. Surface and sub-surface damage should be minim- a designed experimental study on fine grinding of silicon
ized; and wafers. Three-factor two-level full factorial design is
5. The ground wafers should have very good flatness. used in this study. The objective is to reveal the main
This usually means sub-micron TTV (total thick- effects as well as the interaction effects of three process
ness variation). parameters (wheel rotational speed, chuck rotational
speed and feed-rate) on such process outputs as grinding
The previous paper published in this journal [6] force, spindle motor current, cycle time, surface rough-
398 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404

Fig. 5. Effects on grinding force.

Table 4
Spindle motor current data

Maximum
Wheel speed Chuck speed Feed-rate motor current
(amp)

⫺ ⫺ ⫺ 5.5
+ ⫺ ⫺ 4.2
⫺ + ⫺ 5.5
+ + ⫺ 4.3
⫺ ⫺ + 5.5
+ ⫺ + 4.4
⫺ + + 5.5
+ + + 4.5

ness and grinding marks.

There are four sections in this paper. Following this
introduction section, Section 2 describes the design of
experiments and the experimental conditions. In Section
3, the experimental results will be presented and dis-
cussed. Conclusions are drawn up in Section 4. Fig. 6. Effect on spindle motor current.
 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404 399

2. Design of the experiments and experimental

conditions

The experiments are conducted on a Strasbaugh

Model 7AF surface grinder in the process development
laboratory of Strasbaugh, Inc. (San Luis Obispo, CA).
The grinding wheel used is a diamond cup wheel. The
grit size is mesh #2000 and the diameter of the wheel
is 280 mm. As illustrated in Fig. 3, the workpiece
(wafer) is held on the porous ceramic chuck by mean of
a vacuum. The axis of rotation for the grinding wheel is
offset by a distance of the wheel radius relative to the
axis of rotation for the wafer. During grinding, the grind-
ing wheel and the wafer rotate about their own axes of
rotation simultaneously, and the wheel is fed towards the
wafer along its axis.
Single crystal silicon wafers of 200 mm in diameter
with the (100) plane as the major surface are used for
this investigation.
During grinding, deionized (purified) water is used to
cool the grinding wheel and the wafer surface. For this
study, the coolant is supplied to the inner side of the cup
wheel. The coolant flow-rate is 3.2 gallons per minute.
Three process parameters are chosen to study their
effects and interactions:

1. Wheel speed: the rotational speed of the grinding

wheel.
2. Chuck speed: the rotational speed of the chuck. It is
the same as the rotational speed of the workpiece
(wafer).
3. Feed-rate: the feed-rate of grinding wheel (spindle)
towards the wafer.

A 23 (three variables, two levels, eight tests) full fac-

Fig. 8. Effects on grinding cycle time.

torial design is used for the experiments. Detailed

description of factorial design can be found in many text-
books such as the one by DeVor et al. [7]. The matrix
of the experiments is shown in Table 1 and the variable
levels are listed in Table 2. These tests are conducted in
a random order.
Five output variables are observed: grinding force;
spindle motor current; cycle time; surface roughness;
and grinding marks.
Under each test condition, three wafers are ground.
Grinding force, surface roughness and cycle time data
Fig. 7. Comparison of grinding force and spindle motor current. are taken for all the three wafers. Only one data point
400 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404

Table 5
Grinding cycle time data

Wheel speed Chuck speed Feed-rate Grinding cycle time (s)

Wafer 1 Wafer 2 Wafer 3

⫺ ⫺ ⫺ 85 94 108
+ ⫺ ⫺ 105 114 112
⫺ + ⫺ 98 103 107
+ + ⫺ 84 81 94
⫺ ⫺ + 63 60 57
+ ⫺ + 49 63 54
⫺ + + 59 53 57
+ + + 51 51 51

is taken for motor current and grinding marks due to the removal on the ground surface. Then the wafer is
following reasons. The motor currents for the three inspected under a Magic Mirror (Model YIS-200SP-4,
wafers are very consistent and it is very complex and HOLOGENiX, 15301 Connector Lane, Huntington
expensive to prepare the samples for the evaluation of Beach, CA). The picture does not automatically give any
grinding marks by means of a Magic Mirror. quantitative description about the grinding marks. To
The grinder records the grinding force automatically. obtain a quantitative measure for grinding marks, all the
The grinding force measured is the interaction force Magic Mirror pictures are compared and each picture is
between the grinding wheel and the wafer in the direc- assigned a number subjectively according to the severity
tion parallel to the spindle axis. It is also the direction of the grinding marks. For example, the grinding marks
perpendicular to the wafer surface. The maximum force in Fig. 4(a) are hardly visible and thus receive a severity
during the entire grinding cycle is used for analysis. The number of 0. The grinding marks in Fig. 4(b) are severe
monitor of the grinder displays the spindle motor current and therefore number 6 is assigned to the picture.
during grinding and the maximum motor current value
is recorded manually. Grinding cycle time is the time of
actual grinding. It does not include the time for the wheel
to approach the wafer surface and the spark-out time. 3. Results and discussion
Surface roughness of the ground surface is measured
along a direction approximately perpendicular to the In this section, the test results are presented for each
grinding lines. The instrument used is a Tencor P-2 sur- of the output variables. The software called Design-
face profiler (KLA-Tencor, One Technology Drive, Mil- Expert (Version 5, Stat-Ease Corporation, Minneapolis,
pitas, CA). The scan length is 100 µm and scan speed MN) is used to process the data. After identifying the
is 5 µm/s for the measurement. The measurement is done significant effects, the analysis of variance (ANOVA) is
at the same X–Y coordinates for each wafer. performed for each output variable. The details of these
Magic Mirror pictures are used to evaluate the grind- analyses will not be presented here. This section will
ing marks. One wafer from each test condition undergoes give the geometric representations of the significant
a same polishing process with same amount of polishing effects along with some discussion.

Table 6
Surface roughness data

Wheel speed Chuck speed Feed-rate Surface roughness Ra (nm)

Wafer 1 Wafer 2 Wafer 3

⫺ ⫺ ⫺ 13.0 12.5 11.2

+ ⫺ ⫺ 13.0 10.3 10.7
⫺ + ⫺ 16.4 14.3 11.8
+ + ⫺ 18.5 16.8 16.8
⫺ ⫺ + 9.5 11.1 12.2
+ ⫺ + 10.4 11.7 11.3
⫺ + + 16.6 18.5 16.9
+ + + 20.3 17.7 20.6
 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404 401

grinding force; while at the high level of feed-rate, the

increase in chuck speed will decrease the grinding force.

3.2. Results on spindle motor current

Table 4 shows the results on spindle motor current.

The maximum value for the current is very consistent
over three wafers ground. Therefore only one value per
grinding condition is used for ANOVA analysis.
Only the main effect of wheel speed is significant.
Lower wheel speed will cause larger motor current (see
Fig. 6).
Comparing Fig. 5 with Fig. 6, it is easy to see that
the spindle motor current does not have the same
response as grinding force when the process variables
such as wheel speed change their levels. Fig. 7 shows
both grinding force and motor current data when con-
tinuously grinding 35 wafers under a same grinding con-
dition. It can be seen that grinding force is much more
sensitive than the motor current. For instance, the grind-
ing force increases over 50% from the first wafer to the
last wafer ground while the current increases only 5%.

3.3. Results on grinding cycle time

Table 5 and Fig. 8 show the results on grinding cycle

time. Feed-rate has the most significant effect on cycle
time. The higher the feed-rate, the shorter the cycle time.
The interaction between wheel speed and chuck speed
is significant. At the low level of chuck speed, increase
in wheel speed will increase the cycle time. While at the
high level of chuck speed, increase in wheel speed will
decrease the cycle time.

3.4. Results on surface roughness

The results on surface roughness are included in Table

6 and Fig. 9. The main effect of chuck speed is signifi-
cant. Higher chuck speed produces rougher surface. The
interaction between chuck speed and feed-rate is also
significant. The effect of chuck speed on roughness is
stronger at the higher feed-rate level.

Fig. 9. Effects on surface roughness.

3.5. Results on grinding marks

Fig. 10 shows the Magic Mirror pictures for all the

test conditions. Also included are the test conditions and
3.1. Results on grinding force severity number of grinding marks. The Design-Expert
software does not identify any significant effects, prob-
The results on grinding force are shown in Table 3. ably due to the fact that the assignment of the severity
The main effects of wheel speed and feed-rate are sig- number for grinding marks is judgmental. This points
nificant. The increase in wheel speed or feed-rate will out the necessity of looking into some objective rather
increase the grinding force (see Fig. 5). than judgmental ways to quantify grinding marks.
As also shown in Fig. 5, the interaction between chuck All the main effects and interactions are graphically
speed and feed-rate is significant. At the low level of presented in Fig. 11. The figure shows obvious interac-
feed-rate, the increase in chuck speed will increase tions between the three variables.
402 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404

Fig. 10. Results of the grinding marks.

An important observation from Figs. 10 and 11 is the It is clear that the five process outputs studied here
following. With the same grinding wheel and the same respond differently to the change in the process vari-
grinder, altering the process variables (wheel speed, ables. For example, as feed-rate increases, the cycle time
chuck speed and feed-rate) can dramatically change the decreases (Fig. 8) and hence the throughput increases.
severity of grinding marks. There exists an optimum set However, an increase in feed-rate will increase grinding
of process variables that can produce wafers with the force (Fig. 5). Therefore, the optimum grinding con-
least severity of grinding marks. dition for one output is not necessarily good for other
 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404 403

Fig. 11. Effects on the grinding marks.

outputs. In other words, there are no such grinding con- Another important point obtained from this study is
ditions under which all five outputs can be optimized at that the interactions are significant for all the outputs
the same time. Therefore, it is important to prioritize the except for spindle motor current. Therefore, the optim-
requirements for the outputs. ized condition for any of the outputs (except spindle
404 Z.J. Pei, A. Strasbaugh / International Journal of Machine Tools & Manufacture 42 (2002) 395–404

current) cannot be achieved by changing one process 3. Process variables have significant effects on grinding
variable at a time. The variables have to be altered simul- marks. For a given grinding wheel and a given
taneously for optimization. grinder, grinding marks can be greatly reduced by
optimizing the process variables.
4. Compared to spindle motor current, grinding force is
4. Conclusions much more sensitive to changes in the grinding pro-
cess such as wheel status.
A three-factor two-level full factorial design is used
to conduct an experimental investigation into fine grind-
ing of silicon wafers. The main effects and the two-factor
interactions of wheel speed, chuck speed and feed-rate References
on the process outputs (grinding force, spindle motor
[1] A. Mozer, Plane silicon wafer technology, Eur. Semicond. April
current, cycle time, surface roughness and grinding (2000) 29–30.
marks) are obtained. [2] M.S. Bawa, E.F. Petro, H.M. Grimes, Fracture strength of large
The following conclusions can be drawn from this diameter silicon wafers, Semicond. Int. Nov. (1995) 115–118.
study: [3] T. Fukami, H. Masumura, K. Suzuki, H. Kudo, Method of manu-
facturing semiconductor mirror wafers, European Patent Appli-
cation, EP0782179A2, Bulletin 1997/27.
1. The interactions between wheel speed, chuck speed [4] H.K. Tonshoff, W.V. Schmieden, I. Inasaki, W. Konig, G. Spur,
and feed-rate are significant. Therefore, these process Abrasive machining of silicon, Ann. CIRP 39 (2) (1990) 621–630.
variables need to be changed simultaneously to obtain [5] R. Vandamme, Y. Xin, Z.J. Pei, Method of processing semicond-
the optimized output performances. uctor wafers, US Patent 6 114 245, September 5 (2000).
2. The five process outputs respond differently to the [6] Z.J. Pei, A. Strasbaugh, Fine grinding of silicon wafers, Int. J.
Mach. Tools Manufact. 41 (5) (2001) 659–672.
change in process variables. Therefore, these outputs [7] R.E. DeVor, T.H. Chang, J.W. Sutherland, Statistical quality
cannot be optimized at the same time. Compromise design and control, contemporary concepts and methods, Macmil-
and prioritization are needed for process optimization. lan, New York, 1992.