Eliminating the hysteresis effect for reactive sputtering processes
T. Nyberg, S. Berg, U. Helmersson, and K. Hartig
Citation: Applied Physics Letters 86, 164106 (2005); doi: 10.1063/1.1906333
View online: http://dx.doi.org/10.1063/1.1906333
View Table of Contents: http://scitation.aip.org/content/aip/journal/apl/86/16?ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
Particle beam experiments for the analysis of reactive sputtering processes in metals and polymer surfaces
Rev. Sci. Instrum. 84, 103303 (2013); 10.1063/1.4826066
Hysteresis behavior during reactive magnetron sputtering of Al 2 O 3 using a rotating cylindrical magnetron
J. Vac. Sci. Technol. A 24, 934 (2006); 10.1116/1.2198870
Process stabilization and increase of the deposition rate in reactive sputtering of metal oxides and oxynitrides
Appl. Phys. Lett. 88, 161504 (2006); 10.1063/1.2196048
Model relating process variables to film electrical properties for reactively sputtered tantalum oxide thin films
J. Appl. Phys. 93, 3596 (2003); 10.1063/1.1543650
Dynamic simulations of pulsed reactive sputtering processes
J. Vac. Sci. Technol. A 18, 503 (2000); 10.1116/1.582216
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Sun, 23 Nov 2014 17:45:25
Eliminating the hysteresis effect for reactive sputtering processes
T. Nyberg and S. Berga͒
The Angstrom Laboratory, Uppsala University, Box 534, SE-751 21 Uppsala, Sweden
U. Helmersson
Department of Physics, Linköping University, SE-581 83 Linköping, Sweden
K. Hartig
Cardinal CG, Spring Green, Wisconsin 53588
͑Received 10 December 2004; accepted 7 March 2005; published online 15 April 2005͒
Reactive sputter processes frequently exhibit stability problems. The cause of this is that these
processes normally exhibit hysteresis effects in the processing curves. Eliminating the hysteresis
would significantly simplify the use of reactive sputtering processes. So far the only known way of
eliminating the hysteresis is to increase the pumping speed to unrealistically high values. By an
increased understanding of the process we have realized a fully realistic technique to eliminate the
hysteresis for reactive sputtering processes. By simply reducing the size of the target sputter erosion
zone below a critical value, simulations predicted that hysteresis should be eliminated. This has been
experimentally verified for reactive sputtering of Al in an Ar/O2 atmosphere. The fundamental
explanation to this behavior as well as the experimental verification are presented. © 2005
American Institute of Physics. ͓DOI: 10.1063/1.1906333͔
The erosion rate of a sputtering target depends primarily
on the applied target ion current and the sputtering yield.
During reactive sputter deposition the supply of the reactive
gas will change the chemical composition of the target surface
thereby altering the sputter erosion rate. Due to the latter
also the voltage-current characteristics of the sputtering glow
discharge plasma will change. These are complicating factors
in controlling reactive sputtering processes. In fact most reactive
sputtering processes respond in a hysteresis manner to
input processing parameters.1,2
So far the only known way of
eliminating this hysteresis has been to increase the pumping
speed to unrealistically high values. Therefore this method
for eliminating the hysteresis is hardly not used at all.
A way to design a system where this hysteresis is eliminated
has been identified. By theoretical process modeling,
using the Berg model,3,4
we have found that this is indeed
possible by utilizing a small sputter erosion area. In this article,
we present the cause of the hysteresis elimination as
well as experimental verification of the phenomena.
The experiments were performed in a cylindrical
vacuum chamber, 44 cm in diameter and 70 cm in height. A
15 cm diameter Al plate was used as a sputtering target
mounted onto a water cooled magnetron in the top of the
chamber. The target current was supplied using a pulsed dc
arc surpression power supply from company ENI. Prior to
the measurements the chamber was evacuated by a turbomolecular
pump to a base pressure of about 10−6
Torr. The turbomolecular
pump was throttled to obtain the desired pumping
speed. Ar gas was introduced into the chamber to a
pressure of 20 mTorr and the reactive gas O2 was introduced
using a thermal mass flow regulator. The processing pressure
was recorded using an x-t plotter connected to the dc output
from a capacitance manometer. The partial pressure of the
reactive gas was determined by substracting the constant Ar
pressure from the processing pressure.
Two magnetic configurations were used for the magnetron:
A standard, weakly unbalanced magnetic assembly and
a small, magnetic assembly, that also could be rotated in a
circular path inside the magnetron housing. The circulation
allowed the two magnetic setups to run at the same maximum
power with respect to target cooling. These two magnetic
configurations gave rise to erosion tracks on the target
with an approximate diameter of 9 and 2.5 cm, respectively.
The smaller erosion track is estimated to be around 20 times
smaller.
The Berg model for the reactive sputtering process describes
in a simplified way the behavior of the reactive sputtering
process under steady state conditions.3,4
The model
assumes a uniform partial pressure ͑P͒ of the reactive gas in
the chamber. It is also assumed that all material is uniformly
sputter eroded from the target erosion area ͑At͒. Typical calculated
curve for P as a function of supplied reactive gas
flow ͑Q͒ is shown in Fig. 1. The solid S-shaped curve is only
obtained experimentally if the process is carried out with a
fast feedback control system of P. Without a feedback control
system and using Q as the control parameter the system
a͒
Author to whom all correspondence should be addressed; electronic mail:
soren.berg@angstrom.uu.se
FIG. 1. Calculated reactive gas partial pressure P vs supply of the reactive
gas Q. The avalanche positions B-C and D-A define the hysteresis width of
the process also marked by dashed lines and shadowed area.
APPLIED PHYSICS LETTERS 86, 164106 ͑2005͒
0003-6951/2005/86͑16͒/164106/3/$22.50 © 2005 American Institute of Physics86, 164106-1
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Sun, 23 Nov 2014 17:45:25
will avalanche from B to C for increasing values of Q and
avalanche from D to A for decreasing values of Q. The avalanche
positions B-C and D-A define the hysteresis width of
the process. This width is marked by the dashed lines and
shadowed area. As long as the solid line exhibits a S-shape
the process is defined ͑and behave͒ as a hysteresis process.
It has earlier been reported, by Serikawa and Okamoto,5
that the pumping speed ͑S͒ of the external vacuum pump in
the system is of critical importance to the shape of the processing
curves. Increasing S will decrease the width of the
hysteresis region and at a critical value for the pumping
speed the hysteresis will in fact be eliminated. Unfortunately
the pumping speed needed to eliminate the hysteresis is normally
far too high to be realized
In all planar sputtering systems the area of the target
sputter erosion zone At is smaller than the deposited area
͑Ac͒. In the calculations of Fig. 1 it was assumed that At and
Ac were 150 and 2500 cm2
, respectively. In Fig. 2 calculated
target erosion rate ͑R͒ versus Q curves for different values of
At are plotted when the target ion current is 0.5 A. Here we
clearly see that the calculations predict that the S shape of
the curves can be affected and eliminated by decreasing At to
small values. Consequently the calculations predict that it is
possible to obtain hysteresis free reactive sputtering simply
by decreasing the target size.6
Experience has shown that the
preferred experimental processing points are A1,A2, and A3
for the curves in Fig. 2. Notice that the total sputter erosion
rate is greater for the small target ͑A3͒ than for the large
target ͑A1͒ for the same target current. Calculations also predict
͑not shown here͒ that decreasing the sputter erosion area
will result in an increase of the difference between the compositions
of the target surface and the growing film, in such
a way that the target becomes more and more metallic and
the growing film becomes more and more compound rich
͑closer to stoichiometric compound͒. This feature, together
with the nonhysteresis behavior, is close to the ideal reactive
sputtering process.
Calculations of the gettering of the reactive gas inside
the processing chamber ͑QG͒ and the throughput ͑QP͒ of
reactive gas to the external vacuum pump ͑pumping speed S͒
as a function of reactive gas partial pressure may illustrate
the mechanism for eliminating the hysteresis by decreasing
the target area. Calculation results are shown in Fig. 3. The
total supply of the reactive gas ͑Q͒ is the sum of the throughput
to the external pump QP and the gettering in the chamber
QG. The straight line, corresponding to the throughput to the
pump QP will be unaffected by a change in target size. The
effect of decreasing the target size is shown to decrease the
maximum negative slope of the QG versus P curve. Below a
critical target size this negative slope will become smaller
than the positive slope of the straight line for the pumping
speed. Consequently the resulting Q versus P curve will
have no negative slope and the process will not exhibit any
hysteresis. This defines the proposed principle to generate
hysteresis free reactive sputtering processes. It should be understood
that this effect is not caused by an increase of the
target current- or power-density. Increasing the power to the
target will increase the sputter erosion rate at the target area
as well as the deposition rate at the deposited areas. This will
not, however, change the ratio of the reactive gas gettering at
these two areas. The increased power implies more gettering
of reactive gas QG, which pushes the QG versus P curve
upward but at the same time the increased power density
pushes the curve to the right. The combined effect is that the
shape of the curve ͑negative slope͒ is unaffected and the
hysteresis will therefore not be eliminated by an increase of
the target power. However, decreasing the target surface ͑at
constant power͒ will only marginally push the Q versus P
curve upward. Here the total power is constant while the
power density is increased. Consequently, the QG versus P
curve is pushed to the right which gives a reduced negative
slope and subsequently a reduced hysteresis. In this way it is
possible to obtain favorable hysteresis free processing con-
dition.
Experiments were carried out to verify these theoretical
findings. In Fig. 4 are shown the measured P as a function of
Q for the large and the small targets, respectively. The curves
FIG. 2. Calculated sputter erosion rates vs supply of the reactive gas for
different target ͑=effective sputter erosion͒ areas. The unit of the erosion
rate is sputter eroded particles/s. The indicated points A1 , A2, and A3 correspond
to preferred experimental processing points.
FIG. 3. Calculated curves Q vs P, QP vs P, and QG vs P for a small and a
large area target.
FIG. 4. Experimental results for reactive gas partial pressure P vs supply of
the reactive gas Q for a small and a large area target.
164106-2 Nyberg et al. Appl. Phys. Lett. 86, 164106 ͑2005͒
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Sun, 23 Nov 2014 17:45:25
were taken under identical processing conditions ͑pumping
speed, target current, target material, etc.͒. It is clearly indicated
that the hysteresis is eliminated for the small sized
target. Also the deposition rate was found to be higher for the
smaller sized target. These results strongly support the theoretical
predictions.
The focused power dissipation caused by the small target
erosion zone may be evenly distributed over the whole target
surface by moving the magnets inside the target assembly.
An additional experiment showed that the process was hysteresis
free also under such moving magnets conditions
͑0.4 rounds/s͒.
It has been shown that it is possible to obtain hysteresis
free reactive sputtering from a small sized planar target. Experimental
results verify the theoretical predictions of the
behavior of such a process. To distribute the power dissipation
and to allow for a high rate deposition a system with a
small moving magnet may be used. With this technique it is
possible to obtain hysteresis free reactive sputtering process
acting on a large target area.
1
I. Safi, Surf. Coat. Technol. 127, 203 ͑2000͒.
2
S. Berg, T. Nyberg, H.-O. Blom, and C. Nender, Handbook of Thin Film
Process Technology ͑Institute of Physics Publishing, Bristol, UK, 1998͒.
3
S. Berg, H.-O. Blom, T. Larsson, C. Nender, J. Vac. Sci. Technol. A 5, 202
͑1987͒.
4
S. Berg, T. Larsson, C. Nender, and H.-O. Blom, J. Appl. Phys. 63, 887
͑1988͒.
5
T. Serikawa and A. Okamoto, Thin Solid Films 101, 1 ͑1983͒.
6
S. Berg and T. Nyberg, Patent No. WO 03/006703A1 ͑2001͒.
164106-3 Nyberg et al. Appl. Phys. Lett. 86, 164106 ͑2005͒
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Sun, 23 Nov 2014 17:45:25