SOLVING THE MYTHS OF RF PLASMA PROCESSING. DAVID LARNER REPORTS ON THE PROBLEMS OF MEASURING THE POWER USED IN RF PLASMA EQUIPMENT.
European Semiconductor, May 2000
Where would the semiconductor manufacturing industry be without RF plasma systems, back in the dark ages where line widths could be measured with a wooden ruler, that's where. However, those who use RF systems for the generation of plasma, or the acceleration of electrons, will know that such systems exhibit some pretty mysterious characteristics.
Aurion Anlagentechnik in Germany specialises in plasma delivery and monitoring systems, and is no stranger to the problems that can arise from such systems. Roland Gesche, the general manager of Aurion Anlagentechnik, explained some of the enigmatic problems to European Semiconductor and sheds some light on how to solve them.
According to Gesche, there are two major causes of problems in RF plasma equipment that need to be understood - poor performance of standard RF components, and shielding and grounding issues.
Let's begin with one of the most basic pieces of RF equipment, the power measurement systems of a typical RF generator, the reflectometer. Gesche says that some systems available on the market have disadvantages that can cause major problems, showing effects that hide the underlying reasons very effectively. Once recognised, these problems can usually be simply fixed.
Figure 1 shows the typical layout of an RF system using a standard generator for plasma processes. The signal is generated by a quartz-controlled oscillator. A power stage is followed by a low-pass filter and sometimes a low cost and often unreliable reflectometer that delivers signals for forward and reflected power. The forward power (Pv) signal is used to control the amplifier gain for power regulation. The reflectometer is usually directly linked to the output connection of the generator. A coaxial cable connects the generator to the matchbox, which transforms the cathode impedance to the standard impedance, usually 50W.
The filter is intended to suppress harmonics generated by non-linearities or switching distortions of the final power stage. This in turn reduces the distortion of the generator to a value below the permitted maximum when measured at a dummy load, as long as no plasma load is connected.
This specification originally came from the telecommunication equipment, where distortion is a major problem regarding the signal quality of the information modulated on the carrier. At the same time, it allows the calibration of the reflectometer, which is often a simple bridge circuit with integrated demodulation through diodes. This circuit has poor linearity and dynamics. Furthermore, the current transformer used to generate the input signal from the high power line often has a high pass characteristic while the voltage dividers show a mainly flat response. This causes significant measurement errors at frequencies above the specified frequency, such as harmonics (multiples of the base frequency). So actually it is the reflectometer that needs this filter, not the process.
NO PROBLEM WITHOUT PLASMA
A plasma load is a highly nonlinear impedance. Its diode characteristic gave the name to the first sputtering cathodes - the DC self-bias generated by this diode makes RF sputtering work.
But besides the DC offset, multiple harmonics of the base frequency are generated in the plasma at significant levels by repeated frequency multiplication, in addition to a broadband noise.
Figure 2 shows a typical frequency spectrum of a magnetron sputtering cathode operated at a RF source of 13.56MHz. Peaks at all multiples of 13.56MHz up to about 500MHz show the whole harmonic spectrum. The maximum is at a frequency of 67,8MHz, which is the 5th harmonic. Levels in the region of the fundamental frequency occur up to 81.36MHz, which is the 6th harmonic. Values of up to 500MHz, that are about 40dB lower, can still be clearly identified. In this chart voltage levels are shown that cannot be directly converted to power levels, but the base effect is significant. This has serious consequences.
- Electromagnetic radiation does not occur only at the base frequency:
Shielding has to be designed for more than 100MHz; radiation measurements have to be done on the entire frequency range, the use of industrial frequencies gives only minor advantages, mainly that a standardised frequency makes it possible to produce low cost generators.
- The generator output is exposed to noise and harmonics at significant levels. Figure 3 shows how erroneous power measurements can arise if this effect is not considered when the reflectometer is designed.
It's worth remembering that the wavelengths of frequencies above 100MHz are shorter than 3m and reach dimensions that are often beyond the control of industrial equipment. Furthermore, most high power components used, such as capacitors and coils, have resonance frequencies between 20MHz and 100MHz that are caused by parasitic components - so capacitors can be inductive and coils might be capacitive. This makes the behaviour of the system regarding the harmonic spectrum very unpredictable.
The result of these unwanted parasitic effects can lead to power measurement errors of as much as 20% from generators made by different suppliers (forward or delivered power does nor matter, the absolute error typically exceeds the reflected power level). Calibration of the power measurement is often good under laboratory conditions with a dummy load, but this gives no criteria for values measured under real world plasma and process conditions. All is fine as long as there is no plasma and the generator is connected to a dummy load and adjusted to nominal power.
However, these errors are major problems for process stability, reproducibility and portability, especially due to the fact that there is also a non-reproducible, chaotic portion.
ENERGY CRISIS SOLVED?
As shown in Figure 1, the generator is usually power controlled. If a measurement error occurs, the reading remains exactly at the desired level. Adjust the real submitted power changes and the process engineer is left wondering about his results. But the information usually displayed on plasma processing equipment will not help him to identify the reason for the problem. An even worse scenario is that the reflected power reading is also affected. In this instance, the reflected power cannot be minimised adequately either automatically nor manually, but at least there is an indication that something is wrong.
Often this kind of problem is treated by changing the RF coaxial cable length. In a 50W system, the cable length should not cause any changes if a proper impedance matching is achieved. In reality however, changes can be observed. These changes are caused by the different cable lengths for the non matched harmonics. In some cases errors can be reduced by optimising the cable lengths in order to minimise harmonic impact on the reflectometer.
In practice, this is a long and difficult experimental task that can only be done intuitively because there are no rules available. In systems with multiple generators (sputter/bias) and multiple cathodes, it usually becomes a never ending story which starts again after every hardware change of repair. The reflected power reading that is shown in Watts, appears to be a real physical reading. But most users have experienced matching conditions under which the reflected power exceeds the forward power, resulting in delivered power coming from the plasma itself delivered to the generator. So is this the solution to the world's energy problems? Well unfortunately, not.
THE REFLECTED POWER
If the reflected power cannot be minimised even manually, and is stuck at levels of a few percent, there are three possible reasons, according to Gesche. First, one variable element of the matchbox may have reached its limit value and therefore the circuit design cannot match the actual load. This could be because there is something wrong with the cathode, such as a short or open circuit. Secondly, there could be amplitude modulations on the RF generated by power control oscillations of the generator. This sometimes occurs if a generator is used too far below its nominal power level.
Finally, a reflected power reading is affected by harmonics generated in the plasma. In the latter case, matching can be achieved, but measurement makes this impossible.
What is needed is a method that measures power without being affected by harmonics. One possibility is to use a directional coupler instead of a reflectometer, as shown in Figure 4. The RF output signals of the coupler can be filtered before demodulation, so the harmonics are suppressed.
This solution would be preferable, but it depends on the product programme of the RF generator supplier. It is possible to add an additional selective directional coupler or an impedance probe outside of the generator, but it would be difficult and tricky to integrate this into the generator power control.
For example, an interruption of the forward power measurement line would cause the generator to deliver a transient of maximum power to the system that could cause severe damage.
Aurion Anlagentechnik has developed a second solution that can be used with every generator independently from the supplier, and that can also be retrofitted into old systems (Figure 5). An additional power filter is placed between generator and matchbox. It is designed to work as a 1:1 impedance transformer at a frequency of 13.56MHz and has low pass characteristics.
So nothing changes regarding the base frequency, but harmonics are blocked from the reflectometer by the insertion loss of the filter. Experiments have shown good performances and instant problem solving for very difficult cases of the discussed problem.
As the saying goes: 'Yesterdays magic is today's technology', and in this case at least one class of mystic RF problem can be solved through knowledge about the lack of usual RF components, analysing effect, and good old fashioned reasoning.