LABORATOIRE PLASMA ET CONVERSION D’ENERGIE - UMR5213

Participants

Elisa BARISONE^*, Jean-Pierre BOEUF, Philippe BELENGUER, Jean-Luc BONNEVAL, Philippe GUILLOT^**, Thomas NELIS, Leanne PITCHFORD, Laurent THERESE^**, Abdellatif ZAHRI

^* EMPA, Thun, Ch ; ^** CUFR JF Champollion, Albi

Background

Glow discharge Optical Emission and Mass Spectrometry have become commonly used tools for both bulk and surface and interface analysis solid materials. Commercial instruments are available from several large players on the analytical instrument market. When Walter Grimm in 1967 designed his glow discharge spectrometer using a hollow anode tube he mainly thought about bulk analysis of solid conducting materials. However, spark emission spectroscopy and X-ray spectroscopy are the dominants techniques satisfying the needs of the customers in this market. Since then Glow Discharge Optical Emission Spectroscopy (GD-OES) has become a widely used technique for rapid depth profile analysis of surfaces, thin films and coatings where Mass Spectrometry (GD-MS) has for long time found its main application in bulk analysis of trace elements and monitoring the shallow depth distributions of traces, , . Since high-resolution time of flight mass spectrometers (ToF-MS) became available offering simultaneous measurement, the technique is now applied in some leading institutes for Compositional Depth Profiling (CDP) , of thin films. Glow discharges, as a tool for surface and interface analysis, have gained further interest since the introduction of the radio frequency glow discharge (rf-GD). For electrically conducting material, both direct current (dc) and rf show very similar performance in both depth resolution and sensitivity. Non-conducting material, however, can be analysed only with rf-powered discharges. Even on thin non-conducting layers dc-discharges do not operate smoothly. Without entering into the discussion of the subtle and interesting differences between dc and rf discharges, one may say that dc discharges are easier to operate as long as purely conducting materials are analysed, but rf discharges are required as soon as non-conducting materials and layers need to be studied. The first depth profiles reported with GD-OES were of GaAs thin films and stains on steel sheets in the early 1970s. For many years, the technique then developed largely in the steel and automotive industries where it found applications for surface segregation, surface treatments (such as carburising and nitriding), oxidation and passivation, metallic coatings, and polymer coatings. In recent years, the technique has spread into other industries. Hard coatings, produced by either chemical vapour deposition (CVD) or physical vapour deposition (PVD), have become a major application area. GD-CDP is also used in the semiconductor industry for the analysis of the boron phosphorus silicon glass (BPSG) layers on silicon wafers. GD-CDP is not the only technique capable of rapid depth profiling. Its main competitors are dynamic SIMS (Secondary Ion Mass Spectrometry) and SEM (scanning electron microscopy with x-ray analysis). However, GD is cheaper and faster than dynamic SIMS. The results are more easily quantified. It is faster and has better depth resolution than SEM with X-ray analysis. But as all techniques it has some limitations too, the principal being not to be capable of doing micro-spot analysis. Additionally non flat samples, such as screws and tubes are not suitable for direct mounting on the GD source and require accessories to be developed. Porous materials such as foams and many plasma sprayed ceramic coatings are difficult to handle, because they are not vacuum tight. GD-CDP is a comparative technique, i.e. calibration against known reference materials is required for quantitative analysis. Various schemes for quantitative analysis in both RF (radio frequency) and DC (direct current) powered GD sources have been around for some years now. , While they are now gaining some maturity, through international inter-laboratory tests, they are still improving and still require further improvements. Various GD Spectrometers are commercially available. They mostly differ in the detection system, whereas the principle operation of the atomisation and excitation/ionisation source shows only small variations, the major difference being the choice between rf and excitation for discharge ignition. The detector systems can be split into two major groups optical spectrometers, measuring the characteristic emission spectrum of the sputtered analyte material and mass spectrometers detecting the ions produced in the discharge. The optical spectrometers include smaller solid state detector based systems for specific applications and fast large high resolution spectrometer for versatile high end use. GD-MS is today only as a high end instrument available, using a high resolution double focussing magnetic sector mass spectrometer.

General Objectives

The users of the glow discharge request instruments showing both high repeatability, accuracy and versatility. The repeatability of the analytical result is generally expected to be better the 1% in the relative uncertainty. In collaboration with the users of the technique and supported by one of the instrument manufacuteres, Horiba Jobin Yvon, in Longjumeau France we are working on different aspects of the technique in order to help the users and manufacturers to meet their goal of providing reliable material analysis and analytical tools. In the following article we focus on our work concernting the rf-power transfer into the plasma, being the Achilis heel of the rf-GD technique.

Plasma power measurement and control

In both GD-MS and GD-OES the number of ions or photons collected is linked to the chemical composition of the solid sample to be analyzed. The efficiency obviously depends on the collection and detection efficiency, but also on the plasma characteristics, namely the plasma voltage (average energy of electrons and discharge gas ions) and discharge current, influencing the average number of ions and electrons. Two parameters must therefore be controlled to assure reliable elemental analysis. The difficulty of measuring and controlling two independent parameters is one of the major draw-back of RF-GD-OES and RF-GD-MS compared to DC operation. It is well known that only a part of the RF power emitted by the RF generator is actually used to maintain the plasma. A significant part is lost in the RF power supply circuit, the amount of power lost can be of the order of 25% of the emitted power. It was also found that the amount of the power dissipation in the supply circuit increases with the square of the rf-voltage supplied to the sample depends on the sample size and of course on the quality of the rf-supply circuit. The discharge source used in most commercially available GD spectrometers is a 4 mm diameter cylindrical copper electrode facing a plane electrode. The plane electrode is the sample itself and the gap space between the two electrodes is maintained by a toroidal o-ring. The 13.56 MHz radiofrequency voltage is applied on the back of the sample with a cylindrical applicator through a blocking capacitor. An automatic impedance matching system insures the coupling between the generator and the source. A Pirani gauge is used to monitor the pressure inside the reactor. Compared to discharge reactors typically used in the semiconductor manufacturing, this glow discharge source is rather small and therefore presents its specific challenges.

Existing techniques for rf-gd-plasma power measurements

Fabienne Canpont and later Kim Marshall have used a technique known as “subtraction method” or “blind power subtraction” to estimate the power lost in the rf circuit and to correct for this loss to improve the analytical performance of the rf-gdoes systems. This method consists of measuring the (blind) power needed to achieve a given voltage at the sample, without actually igniting the plasma. This blind power depends strongly on the size of the sample used. The blind power must therefore be re-measured each time the sample has been change or moved, i.e. prior to each analysis. Another draw back of this correction method, is that it does not allow the impedance matching during operation and therefore requires rf-generators able to support high levels of reflected powers, which excludes many of the shelf instruments.

Highlights

Integral Plasma power measurement

Fig1. Integral power device

One problem of major concern, affecting the analytical performance of the spectrometer, is measuring the discharge current in radiofrequency discharges and removing the capacitive current due to the reactor capacitance. Applying a radiofrequency voltage of the form V= Vrf cos (ωt), to the back of the sample, gives rise to a capacitive current (Cs dV/dt), where Cs is the system capacitance. This current is significantly higher than the total discharge current. The method for suppressing this capacitive current in the GD-OES instrument is based on the technique perfected on the GEC (Gaseous Electronics Conference) reference cell. We briefly describe this technique here, more detailed information can be found in the paper by Hargis7. The technique consists in adjusting a second capacitor, parallel to the reactor to the same value as the reactor capacitor, when the discharge is not ignited. The capacitance is generally 10’s of picofarads. The accuracy of this technique depends on the possibility to perfectly equalize the two currents, we were able to adjust the two currents within less than a 1 % difference

A diagram of the plasma reactor and electrical measurement system is shown figure 1.

Fig. 2 rf-voltage and capacitive current

Figure 2 shows the current and voltage measured on the two lines (reactor and variable capacitor) over a radiofrequency cycle. We can see on this figure that the measured currents are capacitive currents ; the phase shift between the currents and the voltage is close to 90 degrees. We notice that the two currents are very similar, therefore the total discharge current will be only a small part of the total current. The amplitude of the total current is about 1.2 A. The current on the two electrodes of the plasma reactor has to be identical over a radiofrequency cycle. The surface of the two electrodes being different, the voltage exhibits a large bias, the voltage is shifted to negative values. The voltage and the current are measured on the back of the sample, the small electrode. The measured voltage is of the form V(t) = Vrf cos(ωt) + Vbias , under our conditions we find Vrf = -353 V, Vbias =-260 V. That means that the sample acts as a cathode for most of the cycle. It is during the cathodic part, when the sample is on negative potential, the analyted material is sputtered. The cylindrical electrode is a cathode when the potential, at the sample is positive (only during a short time of the cycle). As the positive voltage amplitude is low, no sputtering is observed during this part of the cycle.

Fig 3. discharge current

On figure 3, we present the total discharge current obtained by subtracting the 2 previous measured currents. The maximum amplitude is about 0.25 A, at half cycle. If we look at the shape of the discharge current, we notice an asymmetry during the first half and the second half of the cycle. The measured total current is the sum of 3 components, electron current, ion current, and displacement current. The displacement current, is related to the temporal variations of the electric field on the electrode and can be written as Id= ε0 S dE/dt, where ε0 is the vacuum permittivity, E the electric field on the electrode and S the electrode surface. The displacement current has an opposite sign during the first half and the second half of the rf cycle. Therefore the displacement will subtract during the first half cycle and add during the second half. We also notice a current peak appearing in mid cycle, which can be attributed to the electron current flowing towards the sample because at this time the electrode is an anode. Knowing the voltage and current time waveforms it is possible to calculate the power deposited in the plasma. The mean power can be written , where T is the rf period, V the applied voltage and It the total discharge current. In the case studied here the mean power calculated is 10.9 Watts to be compared to 15 Watts indicated by the generator. The power efficiency coupling is under our operating conditions around 72 %. This value is coherent with different measurements9-12. Figure 4 and Fgure 5 represent the frequency spectrum of the measured voltage and current. We present the spectrum up to 200 MHz. Compared to the Voltage, the current presents strong components of high harmonics (2.3% in 15th harmonic). For this reason a large number of data points per rf cycle must be acquired and used for the integral power determination. The difference between positive charges and negative charges is less than 0.05 %. We have mentioned previously that the discharge current is the sum of three terms corresponding to the electron, ion and displacement current. It is possible to extract the displacement current from the measured total current, same conduction currents during the sheath contraction and expansion. The sheath contraction and expansion is directly related to the time varying applied voltage but also to the charge transport in the sheath. This corresponds to the strong coupling between charge transport and electric field. We can finally separate the contribution of the particle and displacement current to the total discharge current. If we look at the particle current, we observe a positive current region, that is the ion current (if we neglect the secondary electron current, generally lower than 1/10 of the ion current), it is maximal when the electric field on the electrode is maximal, that is at the beginning of the radiofrequency cycle, the ions are accelerated toward the electrode. When the electric field decreases, the ion current decreases. When the sheath potential is lower than the electron energy, the electrons can overcome the sheath potential and reach the electrode, during that time we can see a large peak of electron current. As this time duration is short, the peak amplitude is high because the number of negative charges must equal to the number of positive charges during a radiofrequency cycle, no time averaged current. Under the chosen experimental conditions the displacement current is of the same order of magnitude than the conduction current, at lower pressure and higher voltage the displacement current will be larger then the conduction current. An analytcal description of the discharge and the results of the plasma modelling performed in this work will be presented in a different section of this web page.

CONCLUSION In this work we measured the total current and the voltage time variations for a radiofrequency discharge using a titanium sample at 13.56 MHz, 950 Pa and 10.9 Watts in argon. From these measurements and using adequate assumptions, supported by observations described in the literature, we are able to electrically characterize the discharge and to get information on the plasma. We found that the plasma is only extended a few millimeters from the sample surface, 2.6 mm, and has a rather high density. The electrical characterization of the discharge will help in building an electrical equivalent circuit including the generator, the matching box and the associated cables. A parametric study of the discharge coupled with this electrical equivalent circuit will be useful in controlling the electrical discharge parameters.

Effective power determination.

Fig. 4 current probe

The above described method is ideal for research laboratory applications, determining the power actually dissipated in the discharge. Given its complexity and cost of required equipment, it is, however, not sensible to implement this solution in an instrument dedicated for routine analysis. We therefore developed a different approach, based on estimating the power consumed in the power supply circuit, represented by a single effective resistance. The approach was developed and tested in close collaboration with EMPA, Thun Switzerland. Most parts, such as cables, connectors and capacitors can be made in sufficiently high quality as to avoid measurable power losses. Inductive coils, however, operated at radio frequencies have a quality factor (Q=ωL/R) of the order of 250 and consequently a significant effective resistance. The effective resistance of the inductive coil used in the automatic impedance matching system is of the order of 1 Ohm. The second inductive coil used in this experimental set-up is smaller and has consequently a smaller effective resistance. Assuming the inductive coil presents the major contribution to the power consumption (except for the plasma), the power loss is determined from the electrical current ‘IL’ through the inductive coil L2 (eq.1) eq.1 The plasma power can be determined from the difference of the generated power and the ‘lost’ power. In the following we will use the term effective resistance ‘Reff’ rather then the equivalent resistance RL of the inductive coil L2, as other components, not identified in this work, may contribute to the measured power loss and its functional dependence on the electrical current measured at the exit impedance matching system.

We made a simple current probe was made in our laboratory using a ferrite ring having a cut off frequency of 40 MHz and an insulated copper wire for the current probe coil (Fig. 4). The voltage generated between the two ends of the coil was used as a measure of the current. Positioned at the exit of the automatic impedance matching system and protected from the strong electric field surrounding the rf-power supply cable, it measures the electrical current flowing through the coil L2. Two different methods were employed to verify the assumption that the power loss in the rf-power supply circuit can be reduced to the effect of one effective resistance. eq.2 The power loss depends on the stray capacity generated by the source including the sample and the voltage applied ‘Vrf’ to the sample[8]. Eq .2 describes the link between the displacement current IL, the stray capacity Cstray and the rf-voltage applied to the sample ‘Vrf’, including its frequency. Samples of different size and material were used for the experiment. The small samples were discs of 2 cm in diameter and 5 mm thickness. The large samples were exactly the same specimen but backed with a large rectangular (9 cm x 17 cm) aluminium plate. Low alloyed aluminium and steel were chosen because of their significantly different secondary electron emission yields. We compared the effective resistance approach to the integral method developed earlier. The effective resistance was calculated from blind power and capacitive current measurements without plasma ignition. When the system is run without igniting a plasma, the rf-voltage at the sample varies strongly with the sample size, the capacitive current measured by the current probe (Fig.4) is independent of the sample. This implies that the power consumption is directly linked to the capacitive current ‘IL’, but only indirectly, via the variable stray capacities, to the potential difference across the discharge source ‘VRF’. The effective resistance thus derived was then used to estimate the plasma power for a large range of discharge conditions. A comparison of the results with integral power measurements is shown in the following table. Small sample size

The values derived by both methods are in good agreement, within 1 W, independent of sample size, total rf power and source impedance.

To illustrate the relevance of the power transmission to analytical routine work we have analysed a computer hard disc, first as a whole piece, a disc of 9 cm diameter and then as a small sample (1 cm x 2 cm) cut out of the same specimen. These hard discs are often used as test sample in GDOES as they show a very reproducible layer structure and homogeneous elemental composition[10]. The nickel phosphate layer and the aluminium-magnesium alloy base material have significantly different secondary electron emission yields leading to different source impedance. The comparison of the intensity-time profiles are displayed in the following figure. The small sample yields higher intensities, and faster erosion. The dc bias voltage decreases for both samples from the NiP layer to the aluminium alloy base material. The average voltage is significantly higher for the small sample. The difference in the detected line intensities is much stronger for the NiP layer and rather small for the aluminium alloy. Increasing the generator power from 40 W to 44 W for the large sample is sufficient to compensate the reduced power transmission. A general calibration procedure, based on the constant emission yield approach, will necessary fail to supply accurate results. Both, a VDC correction or pressure regulation, used to compensate the effects of changes in the source caused by the varying secondary electron emission yield, must fail because the reason for the different auto-bias voltage is not the composition of the sample but its size. Analytical results depending on the sample size is clearly not satisfactory, unless the size of the sample was the be measured.

external links

Spectroscopy (website on spectroscopy for analytical application) ;

Glow Discharge (website on glow discharge optical emission spectroscopy for analytical application) ;

Collaboration

ERT DPHE, Albi ;

STREP EMDPA (Elemental and Molecular Depth Profile Analysis ) ;

RTN GLADNET (Analytical Glow Discharge Network) ;

LPGP, Paris ;

EMPA EMPA Material Science and Technology, Thun Switzerland ;

Horiba Jobin Yvon, Longjumeau ;

Unversity of Oviedo, Spain,