Volume 100, Number 4, July–August 1995

Journal of Research of the National Institute of Standards and Technology

[J. Res. Natl. Inst. Stand. Technol. 100, 427 (1995)]

An Inductively Coupled Plasma Source

for the Gaseous Electronics Conference
RF Reference Cell

Volume 100 Number 4 July–August 1995

Paul A. Miller and In order to extend the operating range of tion that was clearly evident from optical
Gregory A. Hebner the GEC RF Reference Cell, we devel- emission intensity. A dim mode occurred at
oped an inductively coupled plasma source low power and a bright mode at high
Sandia National Laboratories, that replaced the standard parallel-plate power. The transition between modes had
Albuquerque, NM 87185-1423 upper-electrode assembly. Voltage and cur- hysteresis. After many hours of high-
rent probes, Langmuir probes, and an 80 power operation, films formed on elec-
GHz interferometer provided information trodes and walls of one Cell. These de-
Kenneth E. Greenberg and on plasmas formed in argon, chlorine, posits affected the dim-to-bright mode
Paul D. Pochan and nitrogen at pressures from 0.1 Pa to transition, and also apparently caused
3 Pa. For powers deposited in the plasma generation of hot electrons and increased
Department of Chemical and from 20 W to 300 W, the source produced the plasma potential.
Nuclear Engineering, peak electron densities between 1010/cm3
and 1012/cm3 and electron temperatures near Key words: GEC RF Reference Cell;
University of New Mexico, 4 eV. The electron density peaked on high density plasmas; inductively coupled
Albuquerque, NM 87131 axis with typical full-width at half maxi- plasmas; Langmuir probes.
mum of 7 cm to 9 cm. Discharges in
and chlorine and nitrogen had bimodal opera- Accepted: March 27, 1995

Ben P. Aragon
Applied Physics, Inc.,
Albuquerque, NM 87110

1. Introduction
In order to extend the range of operation of the GEC ductively coupled source and three other laboratories are
RF Reference Cell [1], we have developed an inductively preparing to operate similar sources in Reference Cells.
coupled source that replaces the standard parallel-plate Parts for the initial inductively coupled sources were
upper-electrode assembly. This source modification fabricated in 1993 by a local machine shop from
generates plasmas with higher electron densities and sketches produced at Sandia. More recently, a vacuum-
lower pressures (e.g., 1012/cm3 at 300 W and 1 Pa) than component manufacturer has offered a set of parts
are obtainable with the original parallel-plate version of based on updated Sandia sketches. In order to minimize
the Cell. The new operating regime is more relevant to rf power loss, a custom-built manually tuned
new generations of industrial plasma tools being devel- impedance-matching network consisting of two air-
oped by the microelectronics industry. At present, three dielectric variable capacitors was installed immediately
Reference Cells at Sandia are operational with the in- on top of the Reference Cell. The total cost of the

427
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

source, including matching network but excluding rf through a silica window. The top assembly of the source
generator, depends on component selection and on pro- installs in a modified 33.65 mm (131/4 in) flange that
curement means, but is less than $5000. mates to the Reference-Cell chamber. The source re-
As with the original Cell configuration, we intended places the standard upper electrode assembly. The
the source to have satisfactory technical performance, to lower-electrode extension is a disk that rests on top of
be inexpensive, and to have excellent diagnostic access. the standard lower electrode. Bias (dc or rf) can be
We feel that the present inductively coupled source applied to the lower electrode through the connections
meets those goals and opens new areas for further stud- that are normally employed to power the standard paral-
ies in discharge operation. The source hardware is de- lel-plate version of the Cell. A spacer ring in the upper-
scribed in Sec. 2 and our techniques for implementing electrode assembly sets the gap between upper and
Langmuir probes are described in Sec. 3. Section 4 lower electrodes. The antenna-coupling window pro-
presents argon data and Sec. 5 presents chlorine and vides axial optical access to the plasma. Radial diagnos-
nitrogen data. Effects due to films deposited on the tic access is similar to that of the original parallel-plate
electrodes are discussed in Sec. 6. version of the Cell. The radial dimensions of the source
were chosen as a compromise: plasma modelers wished
to have electrodes with infinite radial extent (or a nearby
2. Source Configuration and Features solid cylindrical wall), and experimenters wanted large
openings for probes and microwave beams. In the
The electrode region of the inductively coupled present work, the extent of the plasma beyond the 16.5
source [2] is shown in Fig. 1. The antenna is a five-turn cm diameter electrode assembly, which was an initial
planar coil of 3 mm (1/8 in) diameter copper refrigerator concern, was found to be acceptably small.
tubing (0.75 mm thick walls) that couples to the plasma

Fig. 1. Inductively coupled plasma source with important dimensions in millimeters. The
standard body of the Reference Cell (248 mm ID) surrounded this hardware. Gas was
injected and evacuated through side ports. The flux excluder (dashed) reduced resistive
losses moderately.

428
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

Data were taken using three different inductively cou- and operation of the matching network that will be the
pled sources on three different Cells. As discussed be- subject of a separate publication. We are developing an
low, the main performance differences among Cells automated matching network that uses computer-con-
were due to different histories leading to different levels trolled stepping motors to adjust vacuum-insulated vari-
of electrode contamination. The data reported are for able capacitors.
the nominally clean condition of electrodes. Except as A watt meter was used at the input to the matching
noted below, data were taken with the center turn of the network and the network was adjusted to minimize re-
antenna coil at high voltage and the outer turn grounded. flected power. Reflected power was kept below 1 % of
Operation in the opposite configuration was generally incident power. We present data as functions of ‘‘plasma
similar. As mentioned below, the rf oscillations in power’’. Plasma power refers to the total input power
plasma potential were small and we have not investi- minus I 2Reff losses, where Reff (typically 0.6 V) was due
gated use of an electrostatic shield between the coil and to resistance in the antenna and associated hardware. Reff
plasma. was determined by equating the measured input power
Gas was injected through one side port and exhausted to I 2Reff during matched-impedance operation under
by a throttled turbomolecular pump connected to an- high-vacuum conditions without plasma present. Chilled
other side port. Because the pressures were low and the water flowing through the hollow antenna prevented
pumping rates were low, the asymmetry of the gas injec- overheating of the antenna during extended operation in
tion and exhaust was not thought to be a significant this configuration. Water cooling was used in all our
issue. Flow rates were 3.7 mmol/s to 14.9 mmol/s operations, though it may not be needed in many cases.
(5 sccm to 20 sccm) and no flow-rate dependence of The resistive losses were due in part to currents induced
results was noted. in the stainless-steel cylinder (which became hot to the
A capacitively coupled voltage probe and an induc- touch) that surrounded the cool antenna coil. In one test,
tively coupled current probe were used to measure a copper sleeve was placed between the antenna and the
voltage V across, and current I through the antenna coil. surrounding cylinder in the location indicated in Fig. 1
Construction and operation of probes were as described by the dashed rectangle. The sleeve reduced Reff from
in Ref. [1]. Antennas used to date have been wound by 0.57 V to 0.49 V. We expect that significant resistive
hand with outside diameter of approximately 10 cm and losses also arose due to currents induced in the electrode
with approximately equal spacing between turns. Vec- extension of the lower electrode, though we did not
tor-impedance-meter measurements of several antennas attempt to measure those losses. Consequently, the resis-
gave inductance values L=1.2 mH610 %. The capacitive tive losses could be reduced by use of aluminum in the
voltage probe was calibrated in situ (without plasma) upper and lower electrode assemblies rather than stain-
with reference to a commercial resistive voltage probe. less steel.
The inductive current probe was subsequently calibrated Accounting for Reff caused some uncertainty in our
(without plasma) using the measured L and equating the computation of plasma power. If the plasma signifi-
current to V /vL , where v is the radian frequency of cantly altered the current distribution in the stainless
operation (v=2p313.56 MHz). steel from that occurring in the vacuum case, then Reff
While the source has been operated with a commer- would change in the presence of plasma and our com-
cial impedance-matching network (pi-network configu- puted losses (typically 20 % to 30 % of input power)
ration) delivering rf power via coaxial cable, better per- would change. Further work would be needed to ac-
formance resulted from incorporating a manually tuned count more accurately for resistive losses.
capacitive matching network located directly on top of Figure 1 shows the traversal path for radial scans of a
the Cell. The source relies on large circulating rf current tuned Langmuir probe. That path was chosen to be
(to 25 A) in the antenna. Resistive losses can be large if coincident with the axis of the microwave beam from an
inductors are used in the matching network or if cables 80 GHz interferometer that was used for electron den-
are used to connect the network to the coil. Conse- sity measurements. The Langmuir probe work is de-
quently, the network should avoid use of inductors and scribed below. The interferometer, which was described
output cables. Our matching network employed two air- in detail in Ref. [3], employed dielectric lenses as shown
dielectric variable capacitors with plate spacing of 2 in Fig. 2 to form a microwave beam that passed through
mm. These were suitable for operation at our altitude large ports on opposite sides of the Cell chamber and
(1600 m) at antenna voltages above 2 kV. However, for midway between the upper and lower electrodes. The
some discharge conditions and gases, high-power opera- beam diameter was 1 cm in the electrode region. A
tion resulted in excessive voltage across the capacitors line-integrated electron density of 1.731011/cm2 gave a
which caused breakdown between capacitor plates. phase shift of 18, which was approximately the mini-
There are several considerations and options in design mum detectable phase shift for the interferometer.

429
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

Fig. 3. Four different Langmuir probes for measuring eedfs with

important dimensions in millimeters. The tips all were 0.63 mm di-
ameter wire. The 1.5 mm diameter insulator was alumina; the larger
Fig. 2. Focusing of the 80 GHz microwave-interferometer beam insulators were Pyrex.
through large side ports to measure line-integrated electron density.
The beam diameter was approximately 1 cm in the interelectrode For the plasma parameters of interest, mean free
region.
paths were much larger than the probe diameter and the
Debye length was much less than the probe diameter,
and, consequently, simple one-dimensional probe theory
3. Probe Techniques was deemed adequate. However, axial gradient scale
3.1 Langmuir Probe lengths in the plasma were approximately 1 cm which
was less than typical mean free paths. Consequently,
Langmuir probe measurements can be misinterpreted
distribution functions could have been anisotropic, lead-
easily and a description of our approach is warranted. In
ing to uncertainty in interpretation of probe data.
general, we heeded the admonitions of Godyak [4]. For
The probe was biased negatively 20 V to 40 V with
our work, the regime of interest was densities of 1010
respect to the plasma potential to maintain constant
electrons/cm3 to 1012 electrons/cm3, effective bulk elec-
probe surface conditions. The probe was pulsed positive
tron temperatures of 1 eV to 10 eV, and gas pressures of
by triangular pulses with duration #1 ms at a rate of 10
0.5 Pa to 3 Pa (4 mTorr to 23 mTorr). The corresponding
Hz to 20 Hz. We found, as reported by Godyak [4], that
range of Debye lengths is 0.01 mm to 0.15 mm. The
the probe curves displayed hysteresis when pulse dura-
probe electrical element was 0.63 mm diameter nickel-
tions of 20 ms to 50 ms were used. This implied the
chrome wire. This diameter was chosen to be much
presence of changing probe surface conditions during
larger than the typical Debye length of interest so that
those slower pulses.
probe-area corrections could be neglected. The probe-
The probe’s bias circuit included a tunable L -C shunt
wire material did not appear to interact chemically with
circuit [5] that resonated at the plasma-excitation fre-
any of the discharge-gas species. Four different ge-
quency (13.56 MHz) to allow the probe to follow the
ometries of probes were used. They are shown approxi-
high-frequency oscillations of the plasma potential and
mately to scale in Fig. 3. The succession of smaller
thereby to reduce distortion of current-voltage (I -V )
probe tips and tip holders were tested to measure the
curves. Figure 4 shows a schematic of the bias circuitry.
extent to which the probes perturbed the plasma. Probe
With switch S1 closed, V1 indicated the probe voltage
A, the largest, had a wire surface area of 0.2 cm2. Geo-
and the difference between V1 and V2 indicated the
metric shadowing of the wire by the 3 mm diameter
probe current. Series resistor Rs (22 V to 1000 V) was
Pyrex wire holder reduced the solid angle available for
adjusted for optimum signal levels at different plasma
collection, and the effective area, by 7 %. This wire
densities. The circuit’s resonant frequency was tuned to
holder was in turn supported by a 6 mm diameter Pyrex
maximize the measured dc component of the probe’s
tube that provided the main mechanical support for the
floating potential, which was measured by a digital volt-
60 cm long probe assembly. The probe assembly was
meter (DVM) with switch S1 open. At lower plasma
attached to a manually operated x -y -z translation stage
densities, the tuning resonance was sharp; at higher
which was attached to one of the 6.98 mm (23/4 in)
densities the resonance was broad. This behavior was
flanges on the Reference Cell. The stage had a 15 cm
expected because the impedance of the probe-to-plasma
translation capability along its z axis. The probe-wire
junction decreased as the sheath thickness decreased
axis was oriented radially in the Cell, parallel to the
(with increasing plasma density). As a numerical exam-
lower electrode’s surface. Probes B, C, and D were
ple, a 4 eV plasma with 8.831010/cm3 electron density
supported in the same manner as was probe A. Data
has a Debye length of 0.05 mm. With probe A (0.2 cm2
comparing the results from the probes are presented
below.
430
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

circuit model for the bias circuitry that included correc-

tions for parasitic inductance and capacitance which are
not indicated in Fig. 4. The model was developed and
tested with the probe replaced by resistive loads as well
as open- and short-circuit loads. The first and second
derivatives of the probe I -V curve were obtained by
fitting cubic polynomials to the data at numerous points
and by evaluating the analytic derivatives of the polyno-
mials. The range of fitting was varied from, typically,
1 V near the plasma potential to several volts in the tail
of the distribution. This provided the eedf and identified
the plasma potential Vp (d2I /dV 2=0 and dI /dV is a max-
imum at V=Vp). The computed second derivative usually
peaked at 1 eV to 2 eV below Vp, which is a measure of
Fig. 4. Probe-bias circuit. When the resonant circuit was being tuned, the validity of our techniques [4]. However, this does not
switch S1 was set to the open position and a digital voltmeter (DVM) indicate the presence of such an error in the values
was used to indicate floating potential, which was maximized by obtained for Vp. Our use of cubic polynomials limited
adjusting the variable capacitor. During probe measurements, V1 indi-
the accuracy of eedf computations near Vp because the
cated probe voltage and (V2-V1)/Rs indicated probe current. The
value of Rs was increased for measurements of low currents (low discontinuity in d2I /dV 2 was not treated precisely. Fur-
electron densities). ther algorithm development probably could reduce the
numerical contribution to the energy of the peak. For
area), the probe-to-plasma vacuum capacitance across example, if the cubic-polynomial fit were restricted to a
such a sheath is calculated to be 3.5 pF which has a single-sided fit with V<Vp, then the fit would have more
reactance at 13.56 MHz of 3.3 kV, whereas the noise near the peak but the discontinuity could be ap-
impedance of the L -C circuit was measured using a proached more closely without affecting the fit. Alter-
vector impedance meter to be 31 kV at resonance. Thus natively, one could replace the cubic polynomial form
the effect of the plasma-potential oscillations at 13.56 with a more complicated form that included an explicit
MHz on the I -V curves would be reduced by approxi- break at Vp.
mately a factor of 10 (=34.3/3.3) for this example. This Electron density and effective bulk electron tempera-
estimate is conservative, however, because it neglects ture were obtained from probe data in two different
the effect of conduction current in the sheath on reduc- ways. First , a Maxwellian distribution was fitted to the
ing the probe-to-plasma sheath impedance. Probe data eedf over the range 4 eV to 15 eV (typically) to obtain
showed that, for a case closely matching this numerical a temperature and the eedf was integrated directly to
example, the differential impedance of the I -V charac- obtain an electron density. The derivative signals were
teristic was actually between 60 V and 500 V for more frequently noisy and, thus, the fitted temperatures varied
than 90 % of the probe sweep. Consequently, the effect erratically by 0.5 eV. Second , using the value of Vp from
on probe A of the plasma potential oscillations at 13.56 the differentiated signal, a Maxwellian distribution was
MHz would be reduced by a factor exceeding 50. Mea- fitted to the original I -V curve in the region V<Vp, with
surement of the plasma oscillations is described in Sec. a constant displacement allowed for the ion current. Val-
3.2. ues of electron density and temperature were then calcu-
The probe electrical data (V1 and V2) were recorded lated from the two free parameters of that fitted
using a two-channel digital oscilloscope. The oscillo- Maxwellian.
scope technique was very convenient for us and pro- The first data-analysis technique gave electron densi-
vided the necessary high-speed data acquisition to re- ties that were 10 % to 15 % lower than the second
solve the triangular pulses on a single-shot basis. technique, presumably because of the error in the eedf
However, it did not allow resolution of the tail of the below the peak at 1 eV to 2 eV. The electron tempera-
electron energy distribution function (eedf) because of tures obtained by the two techniques had no consistent
the limited resolution and range of the oscilloscope’s 8 differences. Consequently, we report below the results
bit analog-to-digital converter. This could be improved of the Maxwellian fits to the original I -V curves (the
upon in future work by simultaneously employing two ‘‘second’’ technique, which had lower noise), which
oscilloscopes set at different sensitivities [6] or by using were obtained using Vp from the differentiated data.
higher-resolution recorders. Except as discussed in Sec. 6, the Maxwellian fits to the
The I-V characteristics of the Langmuir probe were data were subjectively quite satisfactory.
derived from the oscilloscope data using an equivalent-

431
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

3.2 Capacitive Voltage Probe of the fundamental. Values of rf voltage and current are
reported in this paper as zero-to-peak values, i.e., one-
The rf components of the oscillating plasma potential
half the peak-to-peak values. Data are shown for two
were measured directly using a capacitive probe im-
different Cells which had Reff=0.36 V and 0.52 V, and
mersed in the center of an argon plasma. The probe
L=1.15 mH and 1.22 mH, respectively. This 30 % differ-
sensing element (<1 cm2 area) was attached to the end
ence in resistances was the biggest seen throughout sev-
of the center conductor of a semirigid coaxial cable. The
eral iterations of antennas and Cells and no specific
sense element and the attached cable were enclosed in 6
cause was identified for the large variation in this partic-
mm diameter Pyrex tubing. The cable was terminated in
ular case. However, as evident from Fig. 5, the differ-
50 V at a digital oscilloscope and the signal was inte-
ence in Reff did not cause major differences in coil
grated to provide an rf voltage measurement. The probe
parameters or plasma power. Usually, Reff was between
was calibrated by tightly wrapping the tip of the Pyrex
0.5 V and 0.7 V. The phase angles in Fig. 5 were
tube with aluminum foil and by exciting the foil with a
calculated to make the voltage and current values consis-
calibration signal from a signal generator. The relation-
tent with the plasma power, as defined in Sec. 2. The
ship of the measured excitation-signal amplitude to the
probe response provided a calibration constant.
With the outside of the induction coil powered (center
grounded) over the range 30 W to 170 W, the rf compo-
nents of the plasma potential were found to be 1.5 V to
2 V amplitude at 13.56 MHz, 0.5 V at the second har-
monic, and less than 0.08 V at higher harmonics. The
harmonic amplitudes decreased with increasing power.
With the center of the coil powered (outer turn
grounded), the oscillating potentials were approximately
twice as large as with the outside powered. Because
these oscillating potentials were so small, the signal
distortion for the tuned Langmuir probe was expected to
be fairly small and predominantly due to the second
harmonic, except perhaps at the lowest densities. The
distortion due to the second harmonic would be reduced
if the Langmuir-probe-bias circuit were modified by
adding a resonance at the second harmonic. Since, in
either configuration of coil excitation, the plasma-po-
tential oscillations were only a few volts, we did not
pursue development and testing of an electrostatic shield
for the inductively coupled source.
At rf input powers above 170 W, the Pyrex tube sur-
rounding the capacitive-probe tip was heated by the
plasma sufficiently to cause expansion and flow of insu-
lating material in the tip and to cause melting of soft
solder in the tip [melting point 455 K (360 F)]. This
severe thermal environment might prove troublesome
for some rf filters that have been used previously in the
tips of Langmuir probes by other workers. We avoided
the use of solder or adhesives in the tips of our Langmuir
probes, which were used at powers to 300 W.

4. Argon Discharges
Argon discharges were generally very stable. Figure 5
shows voltage and current data at the fundamental fre- Fig. 5. Antenna parameters for 2 Pa (15 mTorr) argon discharges in
quency for the antenna coil for discharges in 2 Pa (15 two different Reference Cells. The upper curves (dashed) are for a
mTorr) argon. Harmonics amplitudes were less than 1 % source with unusually low Reff=0.36 V. The solid curve is for a source
with Reff=0.52 V. The phase was calculated from V , I , and power data.

432
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

phase angles were very close to 900 and, consequently,

they could not be measured with sufficient accuracy for
reliable power computation by our standard data acquisi-
tion techniques [1]. The coupling of the antenna to the
plasma was sensitive to any spacing between the an-
tenna and the coupling window. In one test, in order to
maintain constant plasma power, the required antenna
current increased by 50 % when the antenna was moved
approximately 6 mm away from the window.
Figures 6 and 7 show voltage and current data at the
fundamental frequency for the bias electrode for 2 Pa
argon discharges for several rf antenna input powers.
Measured V and I values have been converted to values

Fig. 7. Relation of dc bias on the lower electrode to applied rf power

and voltage for 2 Pa argon discharges.

at the electrode for Figs. 6 and 7 by use of standard

capacitive and inductive corrections [1]. The rf bias was
provided by a tee connected to the main source of rf
power supplying the antenna coil. The bias amplitude
was adjusted by varying a series coupling capacitor.
Relative phase between the antenna and bias signal was
neither determined nor controlled. Figure 6 shows that
the electrode-to-plasma impedance decreased in magni-
tude and became more resistive as bias power increased.
Figure 7 shows how the resulting dc bias was related to
the rf bias power and voltage. Caution should be exer-
cised in operating with high bias voltage. In one case,
extended operation with high bias voltage sputtered
enough metal from the bottom electrode onto the cou-
pling window to make it opaque (mirror-like). The
metal coating attenuated the power delivered from the
antenna coil to the plasma and the window was difficult
to clean. All other data reported in this paper were taken
with the lower electrode grounded.
Figure 8 shows results from radial scans of Langmuir
probes A, B, C, and D in 1.33 Pa (10 mTorr) argon
discharges with 150 W total input power (118 W plasma
power) with the probes 15 mm above the lower electrode
(see Fig. 1). Refer back to Fig. 3 for probe geometries.
As the probe became smaller, the implied electron den-
Fig. 6. RF parameters of the lower electrode vs applied rf bias for 2 sities increased, but the temperatures and plasma poten-
Pa argon discharges. The curves are labeled with rf input power to the
coil antenna.
tials remained fairly constant. There was no clear reason

433
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

Fig. 8. Radial profiles of electron density and temperature and of plasma potential for the four
Langmuir probes of Fig. 3. Data are from 1.33 Pa argon discharges with 118 W plasma power.
The axial position of the probes was 15 mm above the lower electrode. The radial position of
the chamber wall and the extent of the lower electrode are indicted in the bottom graph.

why the electron temperature from probe C was 0.3 eV in Figs. 10–14 have not been adjusted upwards with the
lower than from the other probes. The plasma potentials correction factor suggested by Fig. 9.
all were similar. Figure 9 shows a comparison of the We noted that, at powers of 150 W and higher, strong
line-integrated electron densities from the four probes low-frequency oscillations (e.g. <10 Hz, 10 % ampli-
with the value from the microwave interferometer tude) were present on probe signals when the probe was
(bridge). We conclude that all the probes perturb the extended beyond the center axis, deep into the plasma.
plasma significantly and that probe D gives closest to the We did not see (by eye) any flickering of the discharge
correct value for electron density. Subsequent data were light. In order to obtain probe traces when oscillations
taken with probe D, and the electron densities reported were present, the low-frequency oscillations were sim-

434
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

Fig. 9. Line-integrals of the data in Fig. 8 compared to microwave

bridge data. As the probe size was decreased, the probe values ap-
proached the microwave value.

ply averaged away using the averaging feature of the

digital oscilloscope. We did not observe any degradation
in the quality of the analytic fits to the averaged data due
to the oscillations. It is surprising that the probe data do
not show more spatial asymmetry due to the probe-in-
duced oscillations.
Figure 10 shows an axial scan, on axis at r=0, be-
tween the lower electrode and the coupling window, for
1.33 Pa argon discharges with 150 W input power. To
acquire these data, three versions of probe D were used.
Due to restrictions on the transverse travel of the probe Fig. 10. Axial variation of electron density and temperature and of
stage, and due to obstruction by the clamp ring that plasma potential. Three different probes, all type D, were used to
holds the coupling window, we had to use probes with obtain the axial scan.
bends in the 3 mm diameter Pyrex section near the tip.
The disagreements in Fig. 10 at the joining and overlap confining walls in the radial direction, which enhances
regions of the three probe curves provide a measure of diagnostic access, distinguishes the Reference-Cell ge-
the reproducibility of our techniques. The disagreement ometry from many other inductively coupled sources.
in densities was most probably due to an error in setting We have not evaluated in detail the decrease in electron
probe wire length. temperature at large radii, but we speculate that the
It is interesting to note the difference in the variation decrease may be caused by cooling of the electrons due
of bulk electron temperature in the axial and radial to volumetric expansion. This subjects merits further
directions. In the axial direction (Fig. 10), the tempera- consideration.
ture is quite constant over the extent of the plasma. In the Figure 11 presents radial scans for 1.33 Pa argon
radial direction (Fig. 8), the temperature drops smoothly discharges at various levels of plasma power. The peak
beyond the radius of the edge of the induction coil. In electron density increased with power and the radial
the axial direction the plasma is confined by abrupt profile broadened slightly. The electron temperatures
sheaths at the electrodes while in the radial direction the and plasma potentials show no clear variation with
plasma can stream away freely from the source region. power. A separate set of probe scans (not shown) that
The electrons are collisional with themselves, but nearly extended out to the chamber wall showed the continued
collisionless with the neutrals. The absence of nearby rapid decay of the plasma with increasing radius. Figure

435
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

Fig. 11. Variation with power of plasma parameters for 1.33 Pa argon discharges. For
reference, Gaussian profiles have been matched to the peaks of the 34 W curve and the
245 W curve.

12 shows a comparison of Langmuir-probe data and 5. Chlorine and Nitrogen Discharges

interferometer data. Line-integrated values of electron
density for the profiles in Figs. 8 and 11 are shown along Discharges in all gas species showed bimodal behav-
with interferometer data for 3 different argon pressures. ior with hysteresis in switching between modes, as has
As in Fig. 9, the probe-D values in Fig. 12 are approxi- been seen in other discharge systems. At low input
mately one-third below the interferometer values. power, most of the input power was dissipated by resis-
tive losses (I 2Reff) and the optical emission was weak.
This was termed the ‘‘dim’’ mode. In argon, a mild
transition to a brighter mode occurred at low power, and

436
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

Fig. 12. Integrated electron densities for argon discharges. The Lang-
muir probe data (type-D probe) for 1.33 Pa are approximately 30 % Fig. 14. Integrated electron density for 1.33 Pa chlorine and 2.67 Pa
nitrogen discharges. Electron density and power jumps were seen for
below the corresponding microwave data, as is the case in Fig. 9.
both gases when the discharges switched from the dim to bright mode.

all the data reported above (Figs. 5–12) are for this
‘‘bright’’ mode. In chlorine and nitrogen, the dim mode
persisted above 100 W. As input power was increased
above 100 W, the discharge often changed suddenly and
dramatically, at 150 W to 250 W, to the bright mode in
which most of the input power went into the plasma and
the optical emission increased manyfold. For example,
in one test with 2.67 Pa of nitrogen, the dim mode
persisted up to 175 W input power (<1 W reflected), of
which 150 W was dissipated resistively and 25 W went
into the plasma. The antenna V and I were 2252 V and
21.47 A, with implied phase angle of 89.598. The opti-
cal signal from a broad-band silicon photodiode viewing
the plasma was 0.07 V. Once ignited, the bright mode
could be maintained down to 200 W input power (<1 W
reflected), of which 77.5 W was dissipated resistively
and 122.5 W went into the plasma, a five-fold increase
in plasma power over the dim mode. Antenna parame-
ters in this case were 1639 V, 15.44 A, and 89.098. The
optical signal jumped to 1.3 V in the bright mode, an
increase of a factor of 18.6. Stable operation was not
obtained in that test at input powers between 175 W and
200 W. At times, when the discharge was marginally
stable in the bright mode, insertion of the Langmuir
probe into the discharge caused a mode change or dis-
charge extinction.
We use the terms ‘‘dim’’ and ‘‘bright’’ to describe
the most obvious feature of these modes. It is commonly
accepted that these modes are E and H modes that are
dominated by capacitive and inductive coupling, respec-
Fig. 13. Radial profiles for 185 W discharge in 2.67 Pa chlorine from
tively. The mode change in the molecular gases could
probe D. Although the chlorine data had more variability than the
argon data, the overall shapes were quite similar. also be associated with a change from domination by
molecular neutrals and/or ions to atomic species, which

437
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

would cause a large change in kinetic rates. It should be viewed in ambient fluorescent light. Surface analysis by
noted that simple nonlinear differential equations, which Raman scattering, infrared absorption, and x-ray fluo-
could describe the interaction of plasma and the match- rescence indicated that the film contained predomi-
ing network, can give rise to strong bimodal behavior nantly silicon and oxygen. The Si and O appeared to be
[7]. Phenomenologically similar multi-mode behavior a form of silica deposited at high temperature, and prob-
has been reported in parallel-plate systems [8] where H ably were sputtered from the rf coupling window.
modes should be absent, and in ECR systems [9]. This We demonstrated two obvious effects due to the coat-
subject deserves further study, particularly to elaborate ings. The cause-and-effect relationships were estab-
the mechanism controlling the mode-transition point in lished by selectively replacing coated electrodes with
different geometries and gases. clean electrodes and by observing resulting changes in
Figure 13 shows radial profiles obtained with the Cell operation. First , in chlorine discharges, the coated
Langmuir probe in a 2.67 Pa chlorine discharge at Cell operated stably in the bright mode at 1.33 Pa,
plasma power of 185 W. Because the chlorine discharge whereas a clean Cell did not. For example, in a clean
was less stable than argon discharges and the electron Cell in the dim mode at 1.33 Pa, as rf input power was
density was lower, these data had higher noise levels. increased above approximately 130 W, the discharge
The shapes in Fig. 13 are remarkably similar to those for extinguished completely. In the clean Cell, the pressure
argon in Fig. 11. The electron density was several times had to be above 2 Pa to obtain stable bright-mode oper-
higher in argon than in chlorine at the same power, and ation in chlorine. Second , Langmuir probe data from
the plasma potential was somewhat higher in argon. argon discharges in the coated Cell showed elevated
Langmuir-probe data for a 2 Pa nitrogen discharge at plasma potential (to 45 V) and higher effective electron
167 W plasma power indicated similar radial profiles as temperatures (10 eV). However, unlike the lower tem-
for the other gases. The peak electron density was ap- peratures for clean Cells, these higher ‘‘temperatures’’
proximately 631010/cm3. The plasma potential was 24 V actually gave poor fits to the probe data. The I -V curves
peak and the peak electron temperature was between and computed eedfs for the coated Cell were clearly
5 eV and 6 eV. non-Maxwellian, with one or two large groups of hot
Figure 14 shows microwave interferometer data from electrons in addition to a low-energy group. It is puz-
1.33 Pa chlorine discharges and from 2.67 Pa nitrogen zling, however, that electron density and optical emis-
discharges in the dim and bright modes. In agreement sion were not grossly affected by the changed plasma
with the Langmuir probe data, the electron densities in conditions. Consequently, it is possible that the com-
nitrogen discharges were approximately a factor of three puted eefds with hot electrons are actually indicative of
lower than in chlorine discharges. It is remarkable that some unknown effect that affected the validity of the
the change in electron density between dim and bright Langmuir-probe data analysis.
modes was only a factor of three while the change in If copious hot electrons were actually to occur in a
optical emission was nearly a factor of 20. The one microelectronics production reactor, circuit-damage
isolated data point in Fig. 14 (labeled LP20) is the inte- mechanisms would be affected. The elevated plasma
grated value from the Langmuir-probe data in Fig. 13 potential would increase ion impact energies. Energetic
from a 2.67 Pa chlorine discharge. Because of the pres- electrons could cause damage either directly or indi-
sure difference, that Langmuir probe measurement is rectly by increasing production of ultraviolet radiation.
not directly comparable with the microwave chlorine Further work is needed to understand and validate the
measurements in Fig. 14. We did not take Langmuir- chemical and/or electrical mechanisms of the interac-
probe data in chlorine discharges at 1.33 Pa because, at tion of the coating with the discharge.
that point in the experiments, the probe and interferom-
eter were used with different Cells that operated differ-
ently for reasons discussed below. 7. Summary

The inductively coupled source developed for the

6. Electrode-Coating Effects GEC RF Reference Cell successfully generated high-
density low-pressure plasmas. Langmuir-probe and mi-
As mentioned previously, we employed three different crowave diagnostics showed that electron densities ap-
Reference Cells with inductively coupled sources in this proaching 1012/cm3 were obtained in 1.33 Pa argon at
work. After extensive high-power operation in several 300 W input. The powers per unit volume and per unit
gases, one Cell developed visible coatings on the metal- area were similar to those obtained in industrial high-
lic electrode surfaces. The coatings were transparent density reactors while the unique geometry of the Ref-
and showed several cycles of interference fringes when erence Cell provided exceptional diagnostic access.

438
 Volume 100, Number 4, July–August 1995
Journal of Research of the National Institute of Standards and Technology

While the plasma was not radially confined to the elec- About the authors: Paul A. Miller and Gregory A. Heb-
trode region, only a few percent of the line-integrated ner are Members of Technical Staff in the Laser, Optics,
electron density extended beyond the edge of the elec- and Remote-Sensing Department at Sandia National
trodes. Oscillations in plasma potential were a few volts Laboratories. Kenneth E. Greenberg is a Research Asso-
in argon, suggesting that an electrostatic shield between ciate Professor in the Department of Chemical and Nu-
coil and plasma is not needed for many purposes. Dis- clear Engineering at the University of New Mexico. Paul
charges in argon were stable, but discharges in chlorine D. Pochan is a Research Assistant and student in the
and nitrogen displayed instability and multimode opera- Department of Chemical and Nuclear Engineering at
tion as have been reported in other systems. These insta- the University of New Mexico. Ben P. Aragon is an
bilities compromise plasma control and repeatability. Engineering Technician under contract to Sandia.
Further study is needed of discharge instabilities, multi-
mode operation, and effects of electrode coatings in the
Reference Cell.

Acknowledgments

We thank Lee Berry (Oak Ridge National Labora-

tory), Prof. David Graves (University of California,
Berkeley), and Prof. Mark Kushner (University of Illi-
nois, Urbana) for suggestions concerning the initial de-
sign of the inductively coupled source. We thank Manuel
J. Garcia and David R. Tallant (Sandia) for performing
analysis of the coatings on electrode surfaces. We are
grateful for the support for this work provided by the
U. S. Department of Energy under contract DE-AC04-
94AL85000 and by SEMATECH.

8. References

[1] P. J. Hargis, Jr., K. E. Greenberg, P. A. Miller, J. B. Gerardo,

J. R. Torczynski, M. E. Riley, G. A. Hebner, J. R. Roberts, J. K.
Olthoff, J. R. Whetstone, R. J. Van Brunt, M. A. Sobolewski,
H. M. Anderson, M. P. Splichal, J. L. Mock, P. Bletzinger, A.
Garscadden, R. A. Gottscho, G. Selwyn, M. Dalvie, J. E.
Heidenreich, J. W. Butterbaugh, M. L. Brake, M. L. Passow, J.
Pender, A. Lujan, M. E. Elta, D. B. Graves, H. H. Sawin, M. J.
Kushner, J. T. Verdeyen, R. Horwath, and T. R. Turner, Rev. Sci.
Instrum. 65, 140–154 (1994).
[2] Paul A. Miller, GEC RF Ref. Cell Newslett. 6, 2 (1994); P. A.
Miller, B. P. Aragon, K. E. Greenberg, and P. D. Pochan, Bull.
Am. Phys. Soc. 39, 1463 (1994).
[3] K. E. Greenberg and G. A. Hebner, J. Appl. Phys. 73, 8126–8133
(1993).
[4] V. A. Godyak, Plasma-Surface Interactions and Processing of
Materialas, O. Auciello et al., eds., NATO ASI Series E: Appl.
Sci. 176, 95–134 (1990).
[5] Ajit P. Paranjpe, James P. McVittie, and Sidney A. Self, J. Appl.
Phys. 67, 6718–6727 (1990).
[6] V. A. Godyak, private communication.
[7] For example, see the description of a forced Duffing oscillator by
E. Atlee Jackson, Perspectives of nonlinear dynamics, Vol. 1, Sec.
5.10, Cambridge University Press, NY (1992).
[8] Eray S. Aydil and Demetre J. Economou, J. Appl. Phys. 69,
109–114 (1991).
[9] Eray S. Aydil, Jeffrey A. Gregus, and Richard A. Gottscho, J. Vac.
Sci. Technol. A11, 2883–2892 (1993).

439