Recent Advances in Hollow Cathode Technology for Plasma-Enhanced ALD—Plasma Surface Modiﬁcations for Aluminum and Stainless-Steel Cathodes

: Recent designs have allowed hollow cathode gas plasma sources to be adopted for use in plasma-enhanced atomic layer deposition with the beneﬁt of lower oxygen contamination for non-oxide ﬁlms (a brief review of this is provided). From a design perspective, the cathode metal is of particular interest since—for a given set of conditions—the metal work function should determine the density of electron emission that drives the hollow cathode effect. However, we found that relatively rapid surface modiﬁcation of the metal cathodes in the ﬁrst hour or more of operation has a stronger inﬂuence. Langmuir probe measurements and hollow cathode electrical characteristics were used to study nitrogen and oxygen plasma surface modiﬁcation of aluminum and stainless-steel hollow cathodes. It was found that the nitridation and oxidation of these metal cathodes resulted in higher plasma densities, in some cases by more than an order of magnitude, and a wider range of pressure operation. Moreover, it was initially thought that the use of aluminum cathodes would not be practical for gas plasma applications, as aluminum is extremely soft and susceptible to sputtering; however, it was found that oxide and nitride modiﬁcation of the surface could protect the cathodes from such problems, possibly making them viable.


Introduction
Hollow cathode gas plasma sources have recently emerged as an alternative to inductively coupled plasma (ICP) and microwave plasma (MP) sources for plasma-enhanced atomic layer deposition (PE-ALD) and for plasma-based chemical vapor deposition (CVD). To date, there have been over fifty publications in these related areas . Prior to this, most commercial hollow cathode devices used for film deposition were sputter sources [59,60]; however, these new gas plasma sources operate at much lower power densities and represent a rapidly growing area of development. Table 1, below, provides a listing of some of the materials, devices, and other implementations of these hollow cathode gas sources.
Among other advantages, the rise in the use of these plasma sources is mainly because they are an effective solution to a very subtle oxygen contamination effect caused by the etching of dielectric liners used in legacy plasma sources. This is a problem which is not widely known and is often underestimated but has become of increasing import for the development of nitride and other non-oxide films, and it is being "re-discovered" by many researchers. Hollow cathode gas plasma sources were specifically developed to address this contamination issue and are part of a long history of development in this little-known field. To understand the relevance of these plasma sources, it is important to understand the extent of the problem they solve. Therefore, in this paper, we provide a brief overview of the existing literature related to this problem, which has not been the subject of previous review. As we have designed and built many of the hollow cathode gas plasma sources in use, we are in a unique position to provide this overview. Table 1. Known publications where hollow cathode gas plasmas have been employed.
For ICP and MP sources, there has been a long recognized but not widely known issue of oxygen contamination from the plasma eroding the dielectric windows used to transmit the radiofrequency (RF) or microwave radiation into the vacuum system where the plasma is generated (e.g., [61]). Other sources of oxygen contamination, of course, exist and can include poor vacuum conditions, lack of gas purification, and post-growth ambient exposure; however, a base level of oxygen contamination can be caused by this lesser-known oxygen contamination source [62]. Johnson et al. were perhaps the first to show this parasitic effect, which they observed during the deposition of amorphous silicon layers using a remote microwave hydrogen plasma source [61]. In that case, 1% oxygen contamination was seen when using a quartz liner, though replacing it with an alumina liner reduced the problem for that particular material system. Similar results have more recently been observed with a planar ICP deposition system [63].
Though Johnson et al. [61] identified the effect as far back as 1989, it is only in more recent times that this contamination has become problematic. In past decades, the ICP and MW legacy sources have been used mainly for silicon oxide and silicon nitride deposition. The early work of Lucovsky et al. is demonstrative of this [64]. Obviously, for silicon oxide deposition, oxygen contamination was neither a concern nor, to a certain extent, was it for silicon nitride, which was mainly used for its dielectric properties or as a diffusion barrier. The defect modeling paper by Lucovsky and Lin [65] suggests that oxygen in silicon nitride may contribute to hole trapping, but they asserted that the oxygen concentration in silicon nitride grown by ICP-based remote plasma deposition was too low for this to happen. Though the oxygen contamination was reported as being lower than provided by prior growth techniques, later secondary ion mass spectroscopy (SIMS) results [66] showed that percentage amounts of oxygen were still present in these films. At the time, this may well have been lower than the oxygen contamination by previous growth methods, but the amounts reported were significant in terms of today's technology.
For the thicker films grown in past decades, this oxygen contamination could cause some variation in silicon nitride dielectric properties, but it was often only a problem if it affected the reproducibility of a process. The industry has, however, progressed since then. In more recent years, the requirement for thinner semiconductor layers has seen substantial growth in the atomic layer deposition (ALD) community. In 2008, a Berenberg Equity Research report listed ALD as close to zero percent of the semiconductor deposition equipment market, growing to 7% of the market share by 2012 to approximately USD 500 million [67]. More up to date projections have reported that the market was USD 1.48 billion in 2017 with projected growth to USD 9.5 billion by 2025 [68]. Along with this growing requirement for thinner layers, new materials have also been introduced into the silicon space as well as other parts of the wider semiconductor industry. For these new materials, oxygen contamination can be highly problematic.
Gallium nitride was one of the earlier materials to experience issues with high levels of oxygen contamination due to the etching of ICP and MP windows. This oxygen incorporation led to high background carrier levels in the material [69][70][71]. On page 54 of their 1999 review of defects in GaN, Pearton et al. identified "oxygen impurities leached from the quartz containment vessel often employed in N2 plasma sources" as being a major source of oxygen contamination for GaN research prior to that date [71]. For plasma-assisted molecular beam epitaxy (MBE), this problem was eventually circumvented by using pyrolytic boron nitride ICP windows in an ultra-high vacuum (UHV) environment, a solution, which, as explained below, is less effective in the high or medium vacuum systems used for many plasma-based chemical vapor deposition (CVD) processes. In a very early study in 1981, Matsushita et al. showed a 3.5% oxygen content for plasma CVD deposited GaN using an ICP source [69]. Sato et al. [70] reported high carrier concentrations of 1-3 × 10 19 cm −3 for their undoped material grown with a microwave plasma source, a value that is approximately three orders of magnitude higher than is typical of today's metalorganic CVD (MOCVD)-grown GaN. They also reported seeing "some" oxygen with X-ray photoelectron spectroscopy (XPS) and SIMS. It was later confirmed that oxygen is a shallow donor in GaN [71,72] and is soluble to a concentration of approximately 10 19 cm −3 [72], beyond that level, the oxygen segregates to form extended defects. In fact, for oxygen concentrations of >1%, Butcher et al. [73] observed a correlation between oxygen contamination during film growth and grain size for GaN grown on c-plane sapphire substrates with the segregated oxides decorating grain boundaries.
The influence of oxygen contamination in GaN from a microwave plasma using both quartz and fused alumina (sapphire) dielectric windows was studied in 2002 by Butcher et al. [74]. They found that for GaN growth in a remote plasma high vacuum system with a fused alumina tube and nitrogen plasma, the tube could be passivated over a 66 h period so that there was a reduction in oxygen content to approximately a third of that observed when using a quartz tube. However, both hydrogen and ammonia plasmas were found to strip away the AlN layer that formed on the window if they were subsequently used. Later research showed that in a UHV environment from approximately 10 −7 to 10 −8 Torr, both quartz and fused alumina tubes could be passivated using a nitrogen plasma treatment allowing low-temperature epitaxial growth on templates with oxygen levels close to those of commercial MOCVD-grown material [75]. Again, hydrogen or ammonia plasmas would strip this passivation, and exposure to air would hydrolyze the surface so that re-passivation was required. In fact, at 10 −8 Torr, it was found that after 2 weeks or more of disuse, the background water vapor in the system would also hydrolyze the window surface so that a short period of reconditioning (a few hours) was again required. Similar results were observed with a pyrolytic boron nitride liner at that pressure, for which hydrolysis would form a hydroxide layer that required plasma conditioning for removal. With the better UHV base vacuum pressure of an MBE system (<10 −9 Torr), these issues are generally not a factor; however, in the higher base pressure of a high vacuum system (>10 −6 Torr), the hydrolysis of tube dielectrics due to the background water vapor can take place on a much shorter time scale, so that other solutions must be sought.
Most ALD systems are high or medium vacuum systems with low deposition rates that allow more time for parasitic oxygen species to be absorbed. Oxygen contamination may therefore be considerably more problematic when depositing non-oxide materials. A quick survey of PE-ALD grown nitride films, provided in Table 2, indicates widespread oxygen contamination. It is notable, too, that oxygen (and other elemental) impurity concentrations are often reported using X-ray photoelectron spectroscopy (XPS) data for PE-ALD films, rather than more sensitive techniques, such as secondary ion mass spectroscopy (SIMS), that are typically used for thicker films grown with higher temperature deposition techniques. Of course, there can be several sources of oxygen contamination for any given deposition system, lack of gas purification, high background vacuum leakage, small throughput pumps, the lack of a load lock, and post-growth ambient oxidation being examples; however, most PE-ALD equipment vendors currently use quartz or fused alumina tubes with their ICP sources; thus, some level of contamination can be expected from them. Goto, Shibahara, and Yokoyama were probably the first to identify oxygen contamination from the sputtering of a quartz plasma source for a PE-ALD process; in 1996, they observed 7.6% oxygen contamination in silicon nitride using a microwave plasma source, which they attributed to etching of the quartz liner [76]. It is worth noting that the slightly lower oxygen contamination seen in the TiN deposited by Brennan et al. [80], shown in Table 2 above, may have been because they used a sapphire tube rather than quartz. More recently, Krylov et al. [89] were able to demonstrate that severe oxygen contamination was evident from a quartz ICP tube for the PE-ALD deposition of TiN, with oxygen values ranging from 2.5% to approximately 45%, dependent on the plasma gas used. However, when using a sapphire tube for their ICP source, a reduction in oxygen was seen, down to 1-5% dependent, again, on the plasma gas used. These results are consistent with the earlier observations of Johnson [61] and Butcher et al. [74].
There is a recent paper by Muneshwar and Cadien [90] which examined oxygen in PE-ALD-deposited ZrN using an ICP source, where the ZrN layer was capped in situ with 1 nm of AlN and then 1 nm of Al 2 O 3 . From XPS data, the authors concluded that there was no oxygen in their capped ZrN layer, a result that seems to be in conflict with past observations of oxygen contamination from ICP sources; however, as mentioned above, in some very specific circumstances, the ICP dielectric can be conditioned to lower oxygen contamination [74,75]. Whether this was an instance of that conditioning, or not, is difficult to ascertain, particularly since the inelastic mean free path of photoelectrons in Al 2 O 3 was more recently studied by Powell [91] and assuming AlN is similar, the 2 nm capping layer used by Muneshwar and Cadien was at the escape depth for the O1s signal from their ZrN. Therefore, a low percentage of residual oxygen in their capped ZrN probably would not have been detectable with XPS. The strong oxygen signal from the overlying Al 2 O 3 may have also masked other weaker signals. The escape depth for the Zr3d level was closer to 2.5 nm, so the Zr signal could still be seen. The results are therefore indeterminate; SIMS measurements would have been a useful confirmation.
Rayner et al. [62] used UHV conditions to reduce the oxygen concentration of their ICP-grown TiN, AlN, and silicon nitride layers to less than 1%. They attributed the residual oxygen in their films to the plasma etching of their ICP source's quartz lining. Some degree of plasma conditioning of the liner may have been achieved in that case as per previous reports for quartz and sapphire plasma liners in UHV environments [74,75], though that possibility was not monitored by Rayner et al. However, it is notable that Ozgit-Akgun et al. [1] reported the same oxygen content in their AlN deposited using only a high vacuum system but with a hollow cathode plasma source. The effectiveness of hollow cathode sources seems evident from this.
Interest in the use of hollow cathodes for PE-ALD came about because of some low oxygen contamination results for GaN films grown using a relatively low-temperature means of deposition based on migration-enhanced epitaxy [42,44]. This is a pulsed method for deposition of GaN at temperatures below 670 • C, much lower than typical MOCVD growth temperatures of over 950 • C. However, ALD-like smoothness could be obtained with root mean square atomic force microscopy roughness measurements of below 1 nm [42,44,55] and, more importantly, when grown on GaN templates, oxygen contamination levels of 10 16 cm −3 were measured by SIMS [42], similar to the levels of commercial MOCVDgrown material. Migration-enhanced epitaxy with a hollow cathode plasma source has also shown some excellent results for InN [41,42,47,48] and InGaN [49]. For instance, some of the sharpest InN film photoluminescence ever reported was independently measured by McGill University [47], while some excellent quality epitaxial material has also been achieved [48]. These results came to the early attention of the research group of Nikolaus Dietz of Georgia State University, who upgraded a low-pressure CVD system to operate with a hollow cathode source [51,53,54,56,58]. From there, a group at Bilkent University learned about the low oxygen contamination from these hollow cathode sources.
The Bilkent University group had been growing GaN and AlN in a commercial PE-ALD system but were experiencing the high levels of oxygen seen elsewhere when using an ICP-based plasma source. They had put a load lock on their system, a turbopump, and gas purifiers and had thoroughly leak checked their system but were still experiencing high levels of oxygen contamination, which they eventually traced back to the quartz liner of their ICP source [1]. The liner had actually etched through over time. More recently, Krylov et al. confirmed this source of contamination for the same brand of system [89], though of course the problem is general to quartz ICP sources and, to a lesser extent, those with sapphire ICP sources. Swapping to a hollow cathode design, the Bilkent group were able to take their GaN films from ≥4.7% oxygen contamination to 0.14% as measured using SIMS. Similar results were achieved with their AlN [1], a result that is even lower than has been achieved with sapphire liners. For GaN, the material improvement allowed the Bilkent group to subsequently develop an operational GaN-based thin film transistor grown at the low temperature of 200 • C [2]. The adoption of hollow cathode plasma sources for PE-ALD began from there.
It is worth mentioning some of the results from the University of Texas, Dallas, where some excellent quality, low oxygen content silicon nitride layers have been grown using a hollow cathode plasma source [22,23,27,31]. The layers, grown at 300 • C, are among the best silicon nitride ever grown at low temperature, with current leakage at 2 MV/cm of 1-2 nA/cm 2 and a breakdown voltage of approximately 12 MV/cm [22]. The film density was an impressive 2.9 g/cm 3 [22,23], which allowed a high resistance to etching with 100:1 HF solution of 0.8 nm/min [23]. Some of the above results indicate other advantages of hollow cathode sources (aside from reduced oxygen contamination) including high radical production and low plasma damage; however, the current paper examined the use of aluminum hollow cathode material for the development of large area, gas plasma sources up to 300 mm in diameter as well as the surface modification of aluminum and stainless-steel plasma sources; in the sections below, we present those results.

Materials and Methods
Langmuir probe measurements were made with aluminum test sections used for the modeling of a nominal 12" (300 mm) diameter hollow cathode plasma source. Electrical measurements were subsequently made for a 12" cathode and these were compared with measurements for a 316 stainless-steel Meaglow Series 50 (Thunder Bay, ON, Canada) source (similar to the Meaglow 3 and 3/8" UHV Series hollow cathode plasma sources) and some past published data [44].
The small aluminum hollow cathode test sections were 2.5" in diameter and were used to model the operation of larger 12" (~300 mm) diameter sources that were eventually built for use. Most of the parameters for the test sections scale with area; these test sections were of the "large area" design described below and were run in a specially built vacuum test chamber that was connected to the plasma source as shown schematically in Figure 1 below. The plasma source held the test sections in a 3" diameter nipple, which was connected to the test chamber. The chamber itself had a 4.25" inside diameter and was 5 inches deep. For the testing of the 12" source, a larger purpose-made vacuum chamber was used but only for the electrical measurements. some excellent quality, low oxygen content silicon nitride layers have been grown using a hollow cathode plasma source [22,23,27,31]. The layers, grown at 300 °C, are among the best silicon nitride ever grown at low temperature, with current leakage at 2 MV/cm of 1-2 nA/cm 2 and a breakdown voltage of approximately 12 MV/cm [22]. The film density was an impressive 2.9 g/cm 3 [22,23], which allowed a high resistance to etching with 100:1 HF solution of 0.8 nm/min [23]. Some of the above results indicate other advantages of hollow cathode sources (aside from reduced oxygen contamination) including high radical production and low plasma damage; however, the current paper examined the use of aluminum hollow cathode material for the development of large area, gas plasma sources up to 300 mm in diameter as well as the surface modification of aluminum and stainless-steel plasma sources; in the sections below, we present those results.

Materials and Methods
Langmuir probe measurements were made with aluminum test sections used for the modeling of a nominal 12″ (300 mm) diameter hollow cathode plasma source. Electrical measurements were subsequently made for a 12" cathode and these were compared with measurements for a 316 stainless-steel Meaglow Series 50 (Thunder Bay, ON, Canada) source (similar to the Meaglow 3 and 3/8" UHV Series hollow cathode plasma sources) and some past published data [44].
The small aluminum hollow cathode test sections were 2.5" in diameter and were used to model the operation of larger 12" (~300 mm) diameter sources that were eventually built for use. Most of the parameters for the test sections scale with area; these test sections were of the "large area" design described below and were run in a specially built vacuum test chamber that was connected to the plasma source as shown schematically in Figure 1 below. The plasma source held the test sections in a 3" diameter nipple, which was connected to the test chamber. The chamber itself had a 4.25" inside diameter and was 5 inches deep. For the testing of the 12" source, a larger purpose-made vacuum chamber was used but only for the electrical measurements. The cathode holes of the small test sections had different dimensions related to the wall separations where the hollow cathode effect took place. The effect is dependent on the cathode sheath width, which itself is dependent on gas pressure and type [60]. Larger dimensions work better for the longer mean free path of plasma species at lower pressure, Figure 1. Test chamber used for the 2.5" aluminum cathodes; a Langmuir probe is visible from the side. A schematic of the test setup is also shown.
The cathode holes of the small test sections had different dimensions related to the wall separations where the hollow cathode effect took place. The effect is dependent on the cathode sheath width, which itself is dependent on gas pressure and type [60]. Larger dimensions work better for the longer mean free path of plasma species at lower pressure, and smaller dimensions work better at higher pressures. A number of test sections were examined to achieve the minimum required operating range of 300 mTorr to 8 Torr, though for this report, our interest related to the modification of the aluminum by the plasma and how that affected operational parameters.
The plasma sources were powered with a Seren (Vineland, NJ, USA), 300 W, R301 RF generator operating at 13. 56 MHz. An older Seren AT6M auto-matching network with a Seren MC2 controller was used to match the RF to the plasma source. The match was modified with the addition of some extra inductance to cope with the capacitive nature of the hollow cathode. Nitrogen and oxygen gases were used for the measurements with the For higher pressures, the test systems were pumped by a two-stage Edwards E2M18 (Burgess Hill, UK) rotary pump with fomblin oil; at lower pressures, where backflow from the rotary pump became evident (<120 mTorr), a Pfieffer (Aßlar, Germany) HiPace 300 turbopump was used, backed by the rotary pump.
The hollow cathode test sections were examined visually after being used for plasma generation with nitrogen and oxygen gases. The test chambers each had a one-inchthick, borosilicate view port on the top (see Figure 1) so that the plasma and hollow cathode could be observed while the plasma was running. Langmuir probe measurements were also carried out for the test sections; the purpose of those measurements was not absolute accuracy, they were only indicative measurements used to help determine the range of operation for different gas pressures; therefore, for simplicity, the measurements were performed using an uncompensated single-wire probe as described in previous references [44,55]. Oksuz, Soberon, and Ellingboe [92] revisited the use of uncompensated Langmuir probes with RF fields using this methodology and determined that with a 13.56 MHz RF field, the floating potential and the plasma potential will appear in more negative positions than they should; accurate electron temperatures can be obtained; however, the electron densities will be underestimated [92][93][94]. These limitations are to be noted, though, general data trends can still be observed. For the measurements carried out here, we employed a passive compensation filter at the output of the Langmuir probe (to eliminate the RF signal), as shown in Butcher et al. [44] and in Figure 1, so although the probe was uncompensated, the measurement system was considered partially compensated [92]. The probes themselves were made in-house and were of the cylindrical type, typically 0.7 mm in diameter with a 3-5 mm exposed tip length with ceramic casing and an outer stainless shield. In the oxygen plasmas, the probes could oxidize under the ion bombardment conditions experienced at high negative bias and were replaced when this occurred.
The measurement pressure ranged from the collision-less sheath regime to the quasicollisional sheath regime [95]. The electron temperature (Te) was determined from the transition region between the floating potential and the saturation potential, where the I-V characteristic was logarithmic [96]. A Maxwellian characteristic can be assumed for at least the limited part of the curve that abides by a logarithmic change, though closer to saturation, the curve often showed signs of non-Maxwellian behavior, in part, due to the formation of a quasi-collisional sheath resulting from some of the intermediate pressures used [95]. The electron density (n e ) was determined from the ion saturation current (I ion sat ) as per our prior publication using Langmuir probes [44,55] and as described in the recent review by Merlino [96] according to the equation: where q is the electron charge, A s is the area of the probe, and M is the molecular mass of the ions produced by the plasma. As mentioned above, this methodology is only indicative, and more accurate methods have been recently applied by a US Naval Research group for hollow cathodes examined for PE-ALD use [97]. The electrical supply characterization followed the procedure given by Butcher et al. [44]. The gas pressure of oxygen or nitrogen was changed to specific values and then the applied RF power was increased while measuring the peak-to-peak voltage at the various RF power levels. The measurements were conducted by hand so that the state of the plasma source could be visually confirmed through the top viewport for each setting. To obtain the peak-to-peak voltage measurement, a part of the signal from the match box was teed off to a voltage divider with a 1 MΩ and 1 kΩ resistor, so that the voltage waveform was divided down by 1001:1. An AlloSun (Zhangzhou, China) EM1230 oscilloscope was then used to measure the peak-to-peak RF voltage so that the value applied to the plasma source could be calculated from the measurement. The oscilloscope voltage reading was calibrated beforehand. Coupling of the plasma to parts of the chamber other than the plasma source could cause a slight (<10% of total signal) beating of the RF waveform because of multiple reflected components, particularly at lower pressures; therefore, for simplicity, the peak-to-peak voltage was measured as the maximum envelope of the waveform.
For comparison, some electrical measurements are also presented for a Meaglow Series 50, 316 stainless-steel plasma source. The design of these sources has in the past been limited to operation above 100 mTorr due to the dielectric heating of the feedthroughs and electrical breakdown in those regions. However, for the measurements presented here, the design was modified so that the Series 50 unit had extra dielectric insulation added near the feedthrough to allow operation at lower pressures.
At the beginning of this project, we were unsure that aluminum would be a viable material for hollow cathode gas source construction, the issue being that aluminum is a very soft material with a low heat of sublimation. Materials with low heat of sublimation sputter more readily [98,99]; thus, there were concerns that the aluminum would sputter very quickly when in use. 6061 aluminum alloy was used for the test samples, which is an aviation-grade alloy with 95.8-98.6% aluminum, small amounts of silicon and magnesium, and trace amounts of other elements. The advantage of aluminum is that, like titanium, it is more compatible with silicon-based processes. Some small amount of sputtering can be expected with all plasma sources; however, many transition metals, such as iron and copper, are incompatible with silicon even at parts per billion levels. This is because they are fast diffusers in that material and can introduce deep level defects into the silicon bandgap [100]. We have previously used titanium for hollow cathode construction, largely for TiN deposition; however, this is our first report of aluminum use.
At this stage it is worthwhile describing the design used for the plasma source test sections. Our earliest hollow cathode design, shown in Figure 2, employed many shallow holes over a large area-a strategy that has been used by other designers in the past. However, as shown in Figure 3, the difficulty with this design is uneven striking of the plasma in the cathode holes. Backflow of metalorganic into the cathode, although not problematic with radiofrequency (RF) excitation in terms of the ability to strike a plasma, is problematic from the point of view of the exact conditions under which striking may occur so that there may be some variability among holes with different deposits. This problem was overcome with a plasma cleaning regime between runs and by operating in a narrow pressure range close to the optimum design pressure; however, these solutions were non-ideal in terms of the conditions available for plasma operation. The characterization of one of these sources has been reported elsewhere [44,55].
Coatings 2021, 11, x FOR PEER REVIEW 9 of 27 of this ICP design. Backflow is only a problem with materials that can be conductive when deposited, though of course almost all metalorganics fall in that category. The deposition of conductive oxides and nitrides can also be problematic. With a steady flow of plasma gas through the small tube diameter, there is minimal backflow of species that can damage the tube [64]. This strategy results in plasma source vacuum connections that are generally quite small, NW-50 fittings, and ConFlat fittings that are 4 and 5/8" or 3 and 3/8" are most common, though custom sizes abound. For plasma source replacements with hollow cathodes, these size restrictions meant there could only be a few cathode holes as per the middle example in Figure 2. For that barrel cathode, operating at 300 W, the applied RF power was approximately the same per total hole volume compared to the earlier generation hollow cathode, shown to the left of the figure and described in previous publications [44,55] when using 600 W. The idea with the barrel design was that with fewer, deeper holes, the statistics for striking a plasma in all holes is multiplied so that consistent striking of the entire hollow cathode is possible over a much wider pressure range. Figure 2. Schematics of the three types of hollow cathode design that we employed. The earliest designs were simple metal discs with many shallow holes. A 4.5" diameter example is shown on the left (the other images are at the same scale). The next-generation design included the Series 50 units (shown in the center), which had fewer but deeper holes, with a step in the holes to extend the pressure range. A cross-section through an example unit is shown above. The improved large area design we used here, shown on the right, had a series of annular channels that joined a series of hollow cathode holes. Figure 2. Schematics of the three types of hollow cathode design that we employed. The earliest designs were simple metal discs with many shallow holes. A 4.5" diameter example is shown on the left (the other images are at the same scale). The next-generation design included the Series 50 units (shown in the center), which had fewer but deeper holes, with a step in the holes to extend the pressure range. A cross-section through an example unit is shown above. The improved large area design we used here, shown on the right, had a series of annular channels that joined a series of hollow cathode holes.

Figure 2.
Schematics of the three types of hollow cathode design that we employed. The earliest designs were simple metal discs with many shallow holes. A 4.5" diameter example is shown on the left (the other images are at the same scale). The next-generation design included the Series 50 units (shown in the center), which had fewer but deeper holes, with a step in the holes to extend the pressure range. A cross-section through an example unit is shown above. The improved large area design we used here, shown on the right, had a series of annular channels that joined a series of hollow cathode holes.
(a) (b) The limitation of this second design of plasma sources is that-like the ICP and MP sources they replaced-the plasma was being generated in a small area and then the plasma species were delivered to a much bigger area for film growth, effectively diluting them over that larger area. One of the long-held promises of hollow cathode technology Our next hollow cathode plasma source design was built to overcome the uneven plasma striking problem of the earlier sources. These barrel-shaped cathodes were developed largely as replacements for ICP sources, and they include the Series 50 plasma sources for which measurements are provided below. For many ICP sources, the plasma is generated in a small 1" diameter tube of quartz or sapphire. The reason for this small diameter is that the backflow of metalorganic into the tubes can result in deposits that block the electromagnetic excitation and cause localized heating that can result in catastrophic tube failure; this was discussed by Lucovsky et al. [64], who were early advocates of this ICP design. Backflow is only a problem with materials that can be conductive when deposited, though of course almost all metalorganics fall in that category. The deposition of conductive oxides and nitrides can also be problematic. With a steady flow of plasma gas through the small tube diameter, there is minimal backflow of species that can damage the tube [64]. This strategy results in plasma source vacuum connections that are generally quite small, NW-50 fittings, and ConFlat fittings that are 4 and 5/8" or 3 and 3/8" are most common, though custom sizes abound. For plasma source replacements with hollow cathodes, these size restrictions meant there could only be a few cathode holes as per the middle example in Figure 2. For that barrel cathode, operating at 300 W, the applied RF power was approximately the same per total hole volume compared to the earlier generation hollow cathode, shown to the left of the figure and described in previous publications [44,55] when using 600 W. The idea with the barrel design was that with fewer, deeper holes, the statistics for striking a plasma in all holes is multiplied so that consistent striking of the entire hollow cathode is possible over a much wider pressure range.
The limitation of this second design of plasma sources is that-like the ICP and MP sources they replaced-the plasma was being generated in a small area and then the plasma species were delivered to a much bigger area for film growth, effectively diluting them over that larger area. One of the long-held promises of hollow cathode technology is the potential to scale the source to a large area so that the plasma is generated over the same area as the substrate. Our earlier sources were used for this, but the limitation of uneven plasma ignition, which could be coped with using a rigorous plasma cleaning regime and a limited operating pressure, was not ideal. Our newer large area plasma source design was introduced to overcome these limitations. For this design, a number of circular channels, themselves able to operate as hollow cathodes, join sections of the cathode holes so that the plasma striking in one hole is effectively distributed to all holes connected by the channels, provided sufficient RF power is supplied. Again, this raises the probability of all holes striking at once. Metalorganic backflow does not appear to be overly problematic with this design, which was used for the aluminum test sections described.

Aluminum Test Sections
The 2.5" diameter test sections were all made to model the design of larger 12" diameter aluminum hollow cathodes. The first of these test sections, shown in Figure 4, was initially run in 425 mTorr of nitrogen (100 sccm gas flow) at the full 300 W of power of our Seren RF generator. Twenty minutes of plasma exposure at this power resulted in the rapid buildup of a black outer covering that produced large flakes of material that broke away from the cathode; such flakes could fall on a substrate during a deposition run. A longer plasma exposure time of 105 min resulted in severe pitting of the cathode, see Figure 4b, which was a type of visible erosion of the aluminum indicative of sputtering. For most materials there is a threshold ion energy for sputtering; for materials with a low energy of sublimation, such as aluminum, that threshold energy is lower [98,99], so it was unsurprising that sputter conditions were so easily reached. This would be an unacceptable result during operation of a gas plasma source; however, the test sections scale by area to the larger 12" diameter hollow cathodes they were modeled for, which we only wanted to run at a maximum power of 3000 W. Therefore, the operation of the small diameter test piece at 300 W equated to 23.04 times this for the 12" diameter, or 6912 W of power. We therefore realized that we should scale back the operating power of the test cathodes to the equivalent of 3000 W for the 12" diameter design, that would be approximately 130 W of power for the 2.5" diameter test sections. the channels, provided sufficient RF power is supplied. Again, this raises the probability of all holes striking at once. Metalorganic backflow does not appear to be overly problematic with this design, which was used for the aluminum test sections described.

Aluminum Test Sections
The 2.5" diameter test sections were all made to model the design of larger 12" diameter aluminum hollow cathodes. The first of these test sections, shown in Figure 4, was initially run in 425 mTorr of nitrogen (100 sccm gas flow) at the full 300 W of power of our Seren RF generator. Twenty minutes of plasma exposure at this power resulted in the rapid buildup of a black outer covering that produced large flakes of material that broke away from the cathode; such flakes could fall on a substrate during a deposition run. A longer plasma exposure time of 105 min resulted in severe pitting of the cathode, see Figure 4b, which was a type of visible erosion of the aluminum indicative of sputtering. For most materials there is a threshold ion energy for sputtering; for materials with a low energy of sublimation, such as aluminum, that threshold energy is lower [98,99], so it was unsurprising that sputter conditions were so easily reached. This would be an unacceptable result during operation of a gas plasma source; however, the test sections scale by area to the larger 12" diameter hollow cathodes they were modeled for, which we only wanted to run at a maximum power of 3000 W. Therefore, the operation of the small diameter test piece at 300 W equated to 23.04 times this for the 12" diameter, or 6912 W of power. We therefore realized that we should scale back the operating power of the test cathodes to the equivalent of 3000 W for the 12" diameter design, that would be approximately 130 W of power for the 2.5" diameter test sections.  The same cathode was sand blasted to remove the black coating and to reduce the pitting and was then run at 100 W RF power in nitrogen at 350 mTorr (100 sccm gas flow) for 30 min. Visible inspection of the cathode after this run revealed a bronze color coating on the surface of the cathode, probably indicating the beginning of surface nitridation of The same cathode was sand blasted to remove the black coating and to reduce the pitting and was then run at 100 W RF power in nitrogen at 350 mTorr (100 sccm gas flow) for 30 min. Visible inspection of the cathode after this run revealed a bronze color coating on the surface of the cathode, probably indicating the beginning of surface nitridation of the aluminum as observed by others when using nitrogen plasmas [101]. No flaking was observed. The cathode was then run for a further 2 h at 150 W with no sign of degradation. The RF power was then turned up to 200 W and run for two more hours, again without any signs of degradation (see Figure 5). the aluminum as observed by others when using nitrogen plasmas [101]. No flaking was observed. The cathode was then run for a further 2 h at 150 W with no sign of degradation. The RF power was then turned up to 200 W and run for two more hours, again without any signs of degradation (see Figure 5).  Figure 6 is a simplified schematic showing some of the processes that lead to surface nitridation of the metal cathodes. A more detailed diagram of hollow cathode species generation can be found in a publication by Bardos (Figure 2) [59]. To generate the plasma, thermionic electrons are emitted from the cathode walls, these collide with neutral species creating electron-ion pairs. The nitrogen ions generated in this manner are attracted back to the walls of the cathode by the negative potential of the sheath region (shown in Figure  6). If these ions hit the cathode walls with sufficient energy, sputtering can occur; however, if their energy is below the sputter threshold, they may react with the cathode wall surface and even implant into that surface, creating a nitride surface layer.   Figure 6 is a simplified schematic showing some of the processes that lead to surface nitridation of the metal cathodes. A more detailed diagram of hollow cathode species generation can be found in a publication by Bardos (Figure 2) [59]. To generate the plasma, thermionic electrons are emitted from the cathode walls, these collide with neutral species creating electron-ion pairs. The nitrogen ions generated in this manner are attracted back to the walls of the cathode by the negative potential of the sheath region (shown in Figure 6). If these ions hit the cathode walls with sufficient energy, sputtering can occur; however, if their energy is below the sputter threshold, they may react with the cathode wall surface and even implant into that surface, creating a nitride surface layer.
Coatings 2021, 11, x FOR PEER REVIEW 11 of 27 the aluminum as observed by others when using nitrogen plasmas [101]. No flaking was observed. The cathode was then run for a further 2 h at 150 W with no sign of degradation. The RF power was then turned up to 200 W and run for two more hours, again without any signs of degradation (see Figure 5).
(a) (b)  Figure 6 is a simplified schematic showing some of the processes that lead to surface nitridation of the metal cathodes. A more detailed diagram of hollow cathode species generation can be found in a publication by Bardos (Figure 2) [59]. To generate the plasma, thermionic electrons are emitted from the cathode walls, these collide with neutral species creating electron-ion pairs. The nitrogen ions generated in this manner are attracted back to the walls of the cathode by the negative potential of the sheath region (shown in Figure  6). If these ions hit the cathode walls with sufficient energy, sputtering can occur; however, if their energy is below the sputter threshold, they may react with the cathode wall surface and even implant into that surface, creating a nitride surface layer.  Similar results were observed when using oxygen, though in that case, a grey surface layer was observed. The results indicate that at 300 W, the ion energies increased to beyond the sputter threshold for aluminum, causing severe and rapid damage to the aluminum cathode. However, at lower RF energies of 100-200 W, an outer coating of aluminum nitride/aluminum oxide seemed to evolve (for nitrogen and oxygen plasmas, respectively), discoloring the cathode but without any pitting or flaking effects. Both aluminum nitride and aluminum oxide are quite hard materials compared to aluminum, so they may add some resistance to sputtering of the surface of the aluminum cathode; this was tested by subsequently increasing the RF power to 300 W for the cathode that had previously been coated at 100-200 W of RF power by exposure to a nitrogen plasma. After 30 min of operation in a nitrogen plasma, at 300 W of power, the bronze color of the cathode had intensified, but there was no sign of the flaky black layer seen when exposing a fresh aluminum surface to nitrogen plasma with this RF power, neither was any pitting evident. This suggests that the hard aluminum nitride layer slowly evolved at lower power could protect the underlying aluminum from sputtering.
The outer nitride coating of the aluminum cathodes had another benefit: between the first electron density measurement of these cathodes, when the initial nitridation of the metal surface was taking place, and subsequent measurements (by which time the cathodes had some degree of nitridation), a substantial increase in electron density was observed. For example, at a distance of 35 mm and a pressure of 0.125 Torr, the initial measurement for one fresh aluminum test unit was 7.2 × 10 10 cm −3 with an electron temperature of 1.31 eV, but the second measurement under the same conditions, with a by then at least partly nitrided cathode, was 3.0 × 10 11 cm −3 with a lower electron temperature of 0.361 eV. We studied this effect more carefully by initially subjecting a fresh aluminum cathode to twenty minutes of nitrogen plasma at 100 W of power. The cathode built up some of the bronze-like nitride layer previously seen with this treatment. We subsequently subjected the cathode to nitrogen plasma at 300 W while taking a series of Langmuir probe measurements, each of which took approximately 20 min to perform. Figure 7 shows the current versus voltage trace obtained for the first 300 W run. It can be seen in Figure 7 that there was a voltage difference at 0 V between the initial positive voltage measurements and the subsequent negative voltage measurements, both of which started from 0 V. We interpret the offset in these Langmuir probe measurements as being due to the further nitridation of the aluminum cathode during this run, with most of the nitridation occurring during the earlier positive voltage measurements. It can be seen that the zero offset in this case was over an order of magnitude of current between the positive and negative going voltage measurements, which would equate to a similar change in electron density.

Pressure Range Langmuir Probe Measurements
One of the requirements for the final version of this plasma source configuration was  A slight but much reduced voltage offset was seen for the second Langmuir probe measurement, but by the third measurement, no substantive offset could be recognized. Table 3 shows the electron density and electron temperature measurements for each of the runs. The probe measurements were made at a 5 cm distance from the cathode at 400 mTorr with a 100 sccm nitrogen flow rate.

Pressure Range Langmuir Probe Measurements
One of the requirements for the final version of this plasma source configuration was operation between 0.3 and 8 Torr. To achieve this, there were three-dimensional steps within the hollow cathode test pieces that could be configured to operate in different pressure ranges: there were a series of holes, and these were connected by annular rings (channels) that had a step from one annulus width to another. Figure 2, above, shows an example of this configuration (right-hand image). Nitrogen plasma-based Langmuir probe measurements for one of the aluminum test pieces are provided in Figure 8 and show the operation of this source over a pressure range from 0.125 to 8 Torr. In fact, this plasma source could go down to the 77 mTorr base pressure of the rotary pump used but was not tested below that point.

Pressure Range Langmuir Probe Measurements
One of the requirements for the final version of this plasma source configuration was operation between 0.3 and 8 Torr. To achieve this, there were three-dimensional steps within the hollow cathode test pieces that could be configured to operate in different pressure ranges: there were a series of holes, and these were connected by annular rings (channels) that had a step from one annulus width to another. Figure 2, above, shows an example of this configuration (right-hand image). Nitrogen plasma-based Langmuir probe measurements for one of the aluminum test pieces are provided in Figure 8 and show the operation of this source over a pressure range from 0.125 to 8 Torr. In fact, this plasma source could go down to the 77 mTorr base pressure of the rotary pump used but was not tested below that point.  For the pressure versus electron density data, we would expect three pressure regions with high electron density values that correspond to each of the three hollow cathode dimensions where the hollow cathode effect was generated. Only two peaks were visible: one around 125 mTorr and another at approximately 2 Torr. The third was possibly outside of the pressure range measured or may have overlapped one of the other peaks (as one of the channels had the same separation as the holes, though channels have a slightly different behavior to holes due to their different geometry). It should also be noted that different gases can have different regions of optimum operation. Figure 8 also shows that away from the optimum pressure regions determined by the generation of the hollow cathode effect, the effect still operated, though less efficiently. Other types of plasma source (ICP, MW) also show considerable variation in electron density with pressure.
The graph in Figure 8 demonstrates many of the difficulties found when attempting to obtain useful Langmuir probe measurements, particularly at lower pressures where the plasma could couple with ports in the measurement chamber due to the electrical fields concentrating in those volumes, creating regions of higher-density plasma where the coupling can originate. These coupled plasmas interfere with Langmuir probe measurements making the results difficult to interpret. Gas flow also contributed to this difficulty; like many ALD systems, the test system used here did not have independent gas pressure control, so the gas flow controlled the pressure. Though the hollow cathodes themselves were pressure-dependent devices that operated largely independent of gas flow, the delivery of plasma species downstream of the plasma generation region to a Langmuir probe was dependent on gas flow. Additionally, for a given fixed mass flow of gas provided by a mass flow controller, the actual velocity of gas species would be slower for higher pressures. At higher pressures, a greater number of gas collisions between plasma species and neutral species would cause electrons and ions to lose energy, with an increasing probability of recombination. Therefore, increased gas pressure would result in lower electron densities because of the increased number of collisions due to the higher density of species and their slower velocity. Because of these difficulties, the Langmuir probe measurements have limited value and will vary from chamber to chamber and process to process.
In many contexts, direct power supply measurements at the source have greater value than Langmuir probe measurements that are remote from the source; however, with the larger chamber used here, coupling effects at low pressure were minimized so some analysis of the results could be made. For instance, for the 10 mm distance, there were two data points at 125 mTorr; the lower value was the first measurement with the fresh aluminum surface for which a value of 1.6 × 10 11 cm −3 was found for the electron density; however, for subsequent measurements, the cathode was nitrided so that the electron density for this dimensional configuration increased by three times the value to 4.8 × 10 11 cm −3 . Despite this, the 10 mm distance only had higher electron density values than the 35 mm distance at 125 mTorr. At other pressures, the 35 mm distance showed higher values. This was probably due to the positioning of the Langmuir probe in relation to the metal features of the cathode (walls versus holes) which influenced the measurements at the closer distance quite strongly. In terms of film growth, if the plasma sources were at that 10 mm distance, considerable non-uniformity would probably result. At a greater distance, the plasma expanded out from the hollow cathode holes so that it was more uniform. For the transition from 35 to 50 mm, the expected decrease in electron density was evident. At a greater distance of 50 mm from the plasma source, many of the features seen closer dissipated, particularly at higher pressures, due to the increased number of collisions of the electrons and ions generated; this would probably result in a high flux of neutral radicals, though these are not detected by a Langmuir probe (optical emission spectroscopy measurements by a Georgia State University group have shown a high density of neutral radicals downstream of a hollow cathode plasma source [57]).

Electrical Source Measurements
Because of the difficulties inherent to Langmuir probe measurements at a distance from the source, it was decided to measure the properties of the source directly at the generation point using a combination of electrical characteristics and visual observation. This was done for a nominal 12" large-area aluminum hollow cathode source built using the experience gained from the smaller test sections but also for a 316 stainless-steel Meaglow Series 50 plasma source, for comparison.
As described in the above experimental section, the measurements entailed the collection of peak-to-peak voltage data versus the applied RF power at various pressures using nitrogen for the 12" source and both oxygen and nitrogen for the Series 50 source. Visual checking (in a darkened room) was used to determine whether a plasma was present or not and whether the plasma was lit in all cathode holes. The measurements started from 1 W of applied RF power; however, it should be noted that this was too low for the match box to react to the signal and adjust itself for proper matching, so the tune and load of the auto-match basically stayed at the last settings used. For an applied power of 2 W and more, the match box could react and match the signal, if somewhat slowly at these lower powers. Figure 9 shows the visually observed RF power levels for the striking of a plasma at various nitrogen gas pressures with the 12" aluminum hollow cathode plasma source. In addition, the pressure at which a uniform plasma was observed for the cathode is shown, because for some pressures a higher current/power was required to maintain the plasma over the entire cathode. For the two lowest pressures measured, it was not clear that there was a hollow cathode effect evident. At those pressures, the plasma appeared to be very diffuse throughout the chamber to the maximum applied power of 300 W. Interestingly, the graph shows two dips in the fully on power for the source at approximately 200 mTorr and 2000 mTorr. The dimensions of this source are the same as the test unit in Figure 8, for which the Langmuir probe measurements indicated peaks of electron density at similar pressures; thus, the easier turn on seen here may relate to the optimum pressures for the hollow cathode effect related to those two dimensions.  For Figure 9, it is best to remember that the 12" plasma source design was meant to operate with a maximum power of 3000 W, so the plasma strike powers seen here are very low. Comparisons can be made with similar data for the barrel-shaped Series 50 stainlesssteel plasma source, the data for which is presented in Figures 10 and 11 for nitrogen and oxygen, respectively. Figures 10 and 11 also show data for the earlier large-area plasma design used previously. The data for that were extracted from Reference [44], but they only show when the cathode was initially struck, not that all hollow cathode holes were fully lit with plasma-as that information was not collected at the time. Regardless, it is clearly evident that a much broader range of operation is now possible with the modified Series 50 unit. Similarly, the new large-area design also has a wider range of operation than the old design as shown by comparison with Figure 9. In both cases, the wider pressure range is related to the use of more than one hole dimension for the cathode. . Comparing the RF power needed a) to strike a nitrogen plasma at various pressures for the 12" aluminum hollow cathode source and b) to provide a uniform nitrogen plasma from the source. In the optimum pressure range for the 12" aluminum source, the final version of which is shown in Figure 12, it could be fully struck at values of 3 W, quite a low power, and that low striking power was still the same low power at the lowest pressure measured for that source, i.e., 23 mTorr. In its optimum pressure range, the Series 50 unit could be struck at 5 W of power. However, for the Series 50 plasma source, at both the high and low pressure extremes of operation, the minimum RF power needed to strike the plasma increased as shown in Figure 10 for nitrogen and Figure 11 for oxygen. In the optimum pressure range, all the cathode holes also struck at the same time; however, at the pressure Pressure (mTorr)  There were some differences between the oxygen results and the nitrogen results for the Series 50 unit. Other plasma gases could, of course, show some further variation and, in fact, a second fresh Series 50 hollow cathode plasma source could not be struck at 10 Torr with nitrogen when initially attempting to run the plasma at that pressure. It was only after conditioning the hollow cathode with a nitrogen plasma at a lower pressure of 1 Torr for over an hour that a plasma could be struck at 10 Torr. It is probable that the surface of the stainless steel took on a nitride layer during this conditioning, plasma nitriding of 316 stainless steel is a well-known process [102]. How this would enhance the ability of the plasma source to strike at higher pressure is unclear, though in the discussions section below, we speculate as to the possible reason for this change.
Pressure (mTorr)  In the optimum pressure range for the 12" aluminum source, the final version of which is shown in Figure 12, it could be fully struck at values of 3 W, quite a low power, and that low striking power was still the same low power at the lowest pressure measured for that source, i.e., 23 mTorr. In its optimum pressure range, the Series 50 unit could be struck at 5 W of power. However, for the Series 50 plasma source, at both the high and low pressure extremes of operation, the minimum RF power needed to strike the plasma increased as shown in Figure 10 for nitrogen and Figure 11 for oxygen. In the optimum pressure range, all the cathode holes also struck at the same time; however, at the pressure extremes, the plasma source would initially only partially strike. Higher power was needed to ignite the plasma in all cathode holes; this is also shown in the figures. generation source with nitrogen and b) the modified Series 50 plasma source with oxygen. The data series c) shows when the plasma was fully lit for the modified Series 50 source with oxygen.
There were some differences between the oxygen results and the nitrogen results for the Series 50 unit. Other plasma gases could, of course, show some further variation and, in fact, a second fresh Series 50 hollow cathode plasma source could not be struck at 10 Torr with nitrogen when initially attempting to run the plasma at that pressure. It was only after conditioning the hollow cathode with a nitrogen plasma at a lower pressure of 1 Torr for over an hour that a plasma could be struck at 10 Torr. It is probable that the surface of the stainless steel took on a nitride layer during this conditioning, plasma nitriding of 316 stainless steel is a well-known process [102]. How this would enhance the ability of the plasma source to strike at higher pressure is unclear, though in the discussions section below, we speculate as to the possible reason for this change.  There were some differences between the oxygen results and the nitrogen results for the Series 50 unit. Other plasma gases could, of course, show some further variation and, in fact, a second fresh Series 50 hollow cathode plasma source could not be struck at 10 Torr with nitrogen when initially attempting to run the plasma at that pressure. It was only after conditioning the hollow cathode with a nitrogen plasma at a lower pressure of 1 Torr for over an hour that a plasma could be struck at 10 Torr. It is probable that the surface of the stainless steel took on a nitride layer during this conditioning, plasma nitriding of 316 stainless steel is a well-known process [102]. How this would enhance the ability of the plasma source to strike at higher pressure is unclear, though in the discussions section below, we speculate as to the possible reason for this change.

Discussion
From the results in Table 3, there was an obvious increase in electron density the longer that the nitrogen plasma was run as well as a significant decrease in electron temperature. This can be explained by reference to Figure 13 below. The figure shows a schematic diagram of the energy levels for a surface in vacuum; the example shown is for a semiconductor with conductance band edge (E c ), valence band edge (E v ), and Fermi level (E F ). The near surface vacuum energy is denoted as E vac , and further away from the surface, the E vac will vary with the applied field. E EA is the electron affinity, given as E EA = E vac − E C . In the case of a metal, the Fermi level will exist in the conduction band, since there is no band gap. W, the work function of the material, is the minimum energy needed to move an electron from the surface of a material to the vacuum. Aluminum has a metal work function of 4.06-4.26 eV [103]; however, aluminum nitride can have a negative electron affinity [104,105]. In other words, for electrons to be ejected from the bare aluminum surface, they need to overcome an energy barrier of 4.06-4.26 eV; however, for aluminum nitride, electrons in the conduction band will be ejected into vacuum with no energy barrier. Of course, assuming the Fermi levels of the aluminum and aluminum nitride align and that the Fermi level of the 6.0 eV band gap aluminum nitride is near midgap, then electrons tunneling through the insulating AlN layer (which occurs readily under RF conditions) would be ejected with a lower work function than for a bare aluminum surface by virtue of the negative electron affinity of the latter. This could well explain the increase in electron density for the nitrided aluminum surface: as the nitride layer increases in coverage a higher density of electrons were emitted from the nitride than for the bare aluminum, and it also explains the decrease in electron temperature for emitted electrons, as the energy barrier for emission becomes lower with nitridation. nitride, electrons in the conduction band will be ejected into vacuum with no energy barrier. Of course, assuming the Fermi levels of the aluminum and aluminum nitride align and that the Fermi level of the 6.0 eV band gap aluminum nitride is near mid-gap, then electrons tunneling through the insulating AlN layer (which occurs readily under RF conditions) would be ejected with a lower work function than for a bare aluminum surface by virtue of the negative electron affinity of the latter. This could well explain the increase in electron density for the nitrided aluminum surface: as the nitride layer increases in coverage a higher density of electrons were emitted from the nitride than for the bare aluminum, and it also explains the decrease in electron temperature for emitted electrons, as the energy barrier for emission becomes lower with nitridation. It is to be noted that metal surfaces can also act as catalytic recombination sites for many other plasma species, so that a nitride or oxide hollow cathode surface may have an increased flux of those species compared to a bare metal surface. For instance, for an oxygen plasma, the recombination coefficient of oxygen atoms into oxygen molecules is catalyzed strongly by metal surfaces compared to oxide surfaces as shown in Table 2 by Benavides et al. [106] and Table 1   It is to be noted that metal surfaces can also act as catalytic recombination sites for many other plasma species, so that a nitride or oxide hollow cathode surface may have an increased flux of those species compared to a bare metal surface. For instance, for an oxygen plasma, the recombination coefficient of oxygen atoms into oxygen molecules is catalyzed strongly by metal surfaces compared to oxide surfaces as shown in Table 2 by Benavides et al. [106] and Table 1 by Mozetic et al. [107]. Figure 13 can also be used to explain the inability of a fresh 316 stainless-steel Series 50 plasma source to strike a plasma at 10 Torr, whereas after nitriding, it could. In the case of stainless steel, the metal work function has been shown to increase from 4.5 to 5.8 eV when the material is nitrided [108]. The higher work function of the nitrided material would seem to indicate that even higher voltages should be needed to strike a plasma for a hollow cathode "conditioned" with a nitrogen plasma. However, the work function only indicates part of the process needed to strike a plasma. As pressure increases, higher RF power is needed to strike the plasma as shown in Figures 9-11. Figure 14 compares the raw peak-to-peak voltage versus RF power characteristics for 1 and 10 Torr of pressure with nitrogen for the Series 50 plasma source. The cathode was nitrided at lower power for some time before either of these measurements were taken, so the surface nitriding was comparable. The figure indicates the peak-to-peak RF voltage needed to strike the plasma in both cases, and for the 10 Torr case the required voltage was 2.7 times higher. However, the work function of the metal would be overcome at similarly applied voltages in both cases; thus, from these results, it is clear that other processes are affecting the situation at the higher pressure.
for some time before either of these measurements were taken, so the surface nitriding was comparable. The figure indicates the peak-to-peak RF voltage needed to strike the plasma in both cases, and for the 10 Torr case the required voltage was 2.7 times higher. However, the work function of the metal would be overcome at similarly applied voltages in both cases; thus, from these results, it is clear that other processes are affecting the situation at the higher pressure. In both cases, just below the strike voltage, where there is no plasma, electrons were still flowing from the cathode metal into the gas as evidenced by the non-zero RF power used to reach these high voltages; however, greater voltages are needed to accelerate the electrons to energies where they have sufficient energy to ionize the gas molecules. At the higher pressure, the mean free path of the electrons was reduced, so that the electrons collided more frequently with gas molecules. The loss of energy from these collisions means that the electrons had less chance of reaching the energy needed to ionize the gas molecules through impact ionization. Higher voltages, well above those needed to overcome the work function, must be applied to accelerate the electrons to sufficient energy with their lower mean free path. The early work of Townsend described this process [109]. The higher work function of the nitrided stainless-steel surfaces may actually help with this situation, since the electrons must be emitted from the cathode surface with higher initial energy; in that case, the extra applied voltage needed to accelerate the electrons to an energy sufficient to cause ionization would be less. In both cases, just below the strike voltage, where there is no plasma, electrons were still flowing from the cathode metal into the gas as evidenced by the non-zero RF power used to reach these high voltages; however, greater voltages are needed to accelerate the electrons to energies where they have sufficient energy to ionize the gas molecules. At the higher pressure, the mean free path of the electrons was reduced, so that the electrons collided more frequently with gas molecules. The loss of energy from these collisions means that the electrons had less chance of reaching the energy needed to ionize the gas molecules through impact ionization. Higher voltages, well above those needed to overcome the work function, must be applied to accelerate the electrons to sufficient energy with their lower mean free path. The early work of Townsend described this process [109]. The higher work function of the nitrided stainless-steel surfaces may actually help with this situation, since the electrons must be emitted from the cathode surface with higher initial energy; in that case, the extra applied voltage needed to accelerate the electrons to an energy sufficient to cause ionization would be less.
For nitrogen, the lowest pressure at which a plasma could be struck with the modified Series 50 source was 24 mTorr. The peak-to-peak voltage versus RF power characteristic at this pressure is presented in Figure 15. That figure shows many of the characteristics well known for DC hollow cathodes using voltage versus current characteristics. In the RF case, the phase between current and voltage is not as easily measured, so the RF power, which is strongly dependent on the current, was substituted for ease of measurement. The voltage initially increased sharply with RF power, in this case to 850 V peak to peak. At this point, it was observed that no plasma was present, so that electrons were being injected from the hollow cathode metal surface into the gas, providing a current, but the electrons were causing insufficient ionization to be observed or to sustain a continuous plasma. Then, when the power was increased further, the plasma struck and the voltage dropped as the plasma had greater electrical conductivity than the non-ionized gas, and a greater current was drawn from the power supply. With further increases in power the peak-to-peak voltage increased at a much lower rate because a great deal of current was produced in the hollow cathode holes. Greater detail of the DC characteristics can be found in a paper by Muhl and Perez [60]. They commented that the drop in voltage is a characteristic of the hollow cathode effect, but is not always seen, and indeed for some of the characteristics that we measured at higher pressures, we observed this as well, for instance, the 1 Torr characteristic visible in Figure 14 had only a very slight peak-to-peak voltage drop of 20 V, which is not readily discerned in that figure.
Coatings 2021, 11, x FOR PEER REVIEW 21 of 27 Figure 15. The peak-to-peak voltage versus RF power characteristic measured at 24 mTorr with nitrogen for the modified Series 50 hollow cathode source. The insert shows a magnification of the lower power part of the characteristic.
In regard to the difficulties of obtaining useful Langmuir probe measurements, this was related to the delivery of plasma species downstream of the plasma source, which can be a very complicated process. The flux of a plasma species that actually reaches a substrate for a given power input to a hollow cathode is dependent on the gas type but also on distance from the source, the pressure, and the gas flow rate. As mentioned above, collisions of the plasma species with other neutral species lowers the energy of the ions and electrons and of the excited neutral species, so lower pressures, shorter distances, and higher gas flow all result in a reduction in collisions, while higher pressures, longer distances, and low gas flow result in an increase in collisions. It turns out that for a fixed substrate to plasma source distance, gas pressure is the dominant effect as discussed by Butcher [110]. Of course, the delivery of a greater proportion of active species at low pressure has to be offset against the possibility of plasma damage from the energetic species reaching the substrate, these effects tend to be quite different for different material systems; in fact, some plasma exposure can be beneficial for some materials. Butcher et al. [41] studied plasma damage effects for InN, which results in the preferential loss of nitrogen. At low pressures, they found that indium metal was the dominant component of the films grown and that stoichiometry variation by metalorganic supply made little difference, whereas when the pressure was higher, the plasma damage was limited and the InN could be grown without any free metal. For films grown by PE-ALD, we often find that the growth per cycle increases when using a hollow cathode with the same recipe as an ICP source (an example is given in Reference [29]). We believe this is because there is less plasma damage with the hollow cathode source. Plasma damage often manifests itself by the etching of deposited species, though other forms of damage can occur [111]. For the hollow cathode, since the plasma species are generated in the relatively small volume of the cathode holes, there is limited ion and electron acceleration; this means that there is less generation of the higher energy species that can cause damage.
It should be noted that many of the plasma variables discussed above will be peculiar to particular chambers, so that the measurements presented here are only of limited indicative value. In the end, local measurements within the system to be used are recommended. Empirical film growth and film quality results are often better indicators of the benefit of a particular plasma source configuration. The lower pressure limits of the Series 50 hollow cathode are dimensionally limited, as again explained by the early work of Townsend [109]. As the mean free path increases, there is less likelihood of electrons impacting with enough gas molecules to sustain a plasma within the current hollow cathode structure, larger dimensions would be needed. At 24 mTorr, the mean free path for electrons in nitrogen was 3 mm at room temperature, which still allowed a plasma to be struck. At 17 mTorr, the mean free path in nitrogen was 4.2 mm, and the applied peak-to-peak voltage increased to 2000 V at 40 W without striking a plasma, at which point the matching box was unable to tune. Large electron currents were present, as evidenced by the RF power setting, and these can aid with striking a plasma at higher pressure, but these currents were insufficient at 17 mTorr. For oxygen, a lower limit of 13 mTorr-corresponding to a mean free path of 5.8 mm-was found to be able to strike the plasma. At 9.7 mTorr (7.8 mm mean free path), even with 5000 peak-to-peak V applied at 175 W, no plasma could be struck, though breakdown did occur in the matching box.
In regard to the difficulties of obtaining useful Langmuir probe measurements, this was related to the delivery of plasma species downstream of the plasma source, which can be a very complicated process. The flux of a plasma species that actually reaches a substrate for a given power input to a hollow cathode is dependent on the gas type but also on distance from the source, the pressure, and the gas flow rate. As mentioned above, collisions of the plasma species with other neutral species lowers the energy of the ions and electrons and of the excited neutral species, so lower pressures, shorter distances, and higher gas flow all result in a reduction in collisions, while higher pressures, longer distances, and low gas flow result in an increase in collisions. It turns out that for a fixed substrate to plasma source distance, gas pressure is the dominant effect as discussed by Butcher [110]. Of course, the delivery of a greater proportion of active species at low pressure has to be offset against the possibility of plasma damage from the energetic species reaching the substrate, these effects tend to be quite different for different material systems; in fact, some plasma exposure can be beneficial for some materials. Butcher et al. [41] studied plasma damage effects for InN, which results in the preferential loss of nitrogen. At low pressures, they found that indium metal was the dominant component of the films grown and that stoichiometry variation by metalorganic supply made little difference, whereas when the pressure was higher, the plasma damage was limited and the InN could be grown without any free metal. For films grown by PE-ALD, we often find that the growth per cycle increases when using a hollow cathode with the same recipe as an ICP source (an example is given in Reference [29]). We believe this is because there is less plasma damage with the hollow cathode source. Plasma damage often manifests itself by the etching of deposited species, though other forms of damage can occur [111]. For the hollow cathode, since the plasma species are generated in the relatively small volume of the cathode holes, there is limited ion and electron acceleration; this means that there is less generation of the higher energy species that can cause damage.
It should be noted that many of the plasma variables discussed above will be peculiar to particular chambers, so that the measurements presented here are only of limited indicative value. In the end, local measurements within the system to be used are recommended. Empirical film growth and film quality results are often better indicators of the benefit of a particular plasma source configuration.

Conclusions
Initial results for hollow cathode gas plasma sources made of aluminum were presented. Like our past titanium sources, these aluminum plasma sources should be more compatible with silicon technology than stainless-steel hollow cathodes.
Because of its softness, we had concerns that an aluminum hollow cathode might have problems with excess sputtering; however, for operation with nitrogen, it was found that as long as the power was initially kept below the sputter threshold for the aluminum, a hard nitride layer would build up very quickly, in less than an hour on the bare aluminum surface. This layer can protect the underlying aluminum from the effects of sputtering and had the added advantage of an increase in the electron density compared to the naked aluminum metal. Similar results were seen for oxygen. A 12" diameter plasma source operating between 23 mTorr and 8 Torr with nitrogen plasma was demonstrated based on the modeling results of smaller test pieces.
For comparison, results were also provided for a stainless-steel Meaglow Series 50 hollow cathode plasma source. An updated version of the design-typically used as an ICP replacement-incorporated more insulation near the feedthroughs where dielectric heating effects had limited the operation to over 100 mTorr. With the modification, the source could be operated at lower pressures, down to 24 mTorr for nitrogen and 18 mTorr for oxygen. Moreover, it was found that operation up to 10 Torr was possible for the stainless-steel source but only after conditioning the plasma source with a nitrogen or oxygen plasma. We have speculated that the increased work function of the oxide and nitride surfaces, compared to the bare metal, is beneficial in producing higher energy electrons at this higher pressure so that lower applied RF voltages are needed for plasma ignition. This subtlety of operation had not been noted previously.