Plasma Sources Sci. Technol. 8 (1999) R21–R44. Printed in the UK PII: S0963-0252(99)02255-0 REVIEW ARTICLE Cold-cathode discharges and breakdown in argon: surface and gas phase production of secondary electrons A V Phelps† and Z Lj Petrovi憇 † JILA, University of Colorado and National Institute of Standards and Technology, Boulder, CO 80309-0440, USA ‡ Institute of Physics, PO Box 57, Belgrade, Yugoslavia Received 16 October 1998, in final form 24 February 1999 Abstract. We review the data and models describing the production of the electrons, termed secondary electrons, that initiate the secondary and subsequent feedback avalanches required for the growth of current during breakdown and for the maintenance of low-current, cold-cathode discharges in argon. First we correlate measurements of the production of secondary electrons at metallic cathodes, i.e. the yields of electrons induced by Ar+ ions, fast Ar atoms, metastable atoms and vuv photons. The yields of electrons per ion, fast atom and photon vary greatly with particle energy and surface condition. Then models of electron, ion, fast atom, excited atom and photon transport and kinetics are fitted to electrical-breakdown and low-current, discharge-maintenance data to determine the contributions of various cathode-directed species to the secondary electron production. Our model explains measured breakdown and low-current discharge voltages for Ar over a very wide range of electric field to gas density ratios E/n, i.e. 15 Td to 100 kTd. We review corrections for nonequilibrium electron motion near the cathode that apply to our local-field model of these discharges. Analytic expressions for the cross sections and reaction coefficients used by this and related models are summarized. 1. Introduction This paper is concerned with the processes responsible for the production of the initial secondary or feedback electrons required for the growth of current at electrical breakdown and for the maintenance of cold-cathode discharges in Ar. These electrons are produced in collisions of Ar ions, fast Ar atoms, metastable atoms or photons with the cathode or in ionizing collisions of fast atoms or ions with the neutral Ar atoms in the gas phase. We review the published data for these processes and select the data that lead to consistency between our model and measured breakdown voltages and low-current, discharge-maintenance voltages for a wide range of operating conditions. Although our analyses make use of data obtained at low current densities, the assembled data are also appropriate for use in models of cold-cathode discharges at the higher current densities found in the normal and abnormal glow modes. Computer models of low-pressure electrical discharges in Ar have become a very useful tool for understanding and predicting the properties of discharges used for plasma processing [1–4], surface sputtering [5, 6], plasma displays [7, 8], lighting [9, 10], switching [11, 12], scintillation 0963-0252/99/030021+24$19.50 © 1999 IOP Publishing Ltd detectors [13] and lasers [14, 15]. In order for these models to be quantitatively useful it is necessary for the modeller to use realistic cross sections for electron, ion, atom and photon collisions with the gas and to use realistic probabilities for electron, ion, atom and photon interactions with electrode surfaces. The cross sections and rate coefficients required for the modelling of the gas-phase portion of discharges in Ar are relatively well known. In particular, many people have addressed the question of electron–Ar cross sections [3, 16–24]. Some of the cross section sets in use are consistent with and some badly inconsistent with simple swarm experiments [25]. The published cross section data for Ar+ –Ar and Ar–Ar elastic and inelastic collisions have been reviewed [26]. The accuracy of ion–atom cross sections by modellers has been considered recently [27], and recommendations made for avoiding common errors in their application. Photon absorption data and experimentally verified resonance-photon scattering treatments appropriate to models of discharges in Ar are available in the literature [28, 29]. The role of secondary electron avalanches in the electrical breakdown of gases and in low-current stationary discharges has been discussed in detail in numerous texts R21 A V Phelps and Z Lj Petrović and reviews [30–36]. We will include all generations of avalanches beyond the first under the terminology of secondary or feedback avalanches. The processes that have been considered in the literature for the production of the initiating secondary electrons include electron emission from the cathode induced by positive ions, by fast atoms, by photons and by metastable atoms and molecules; collisional ionization by fast ions and atoms produced by earlier avalanches; collisional ionization by electrons backscattered from the anode and photoionization of the gas by photons from earlier avalanches. We will not consider photoionization of the Ar and will not review the data on the backscattering of electrons at the anode. For convenience, we will usually drop the qualifier ‘avalanche initiating’ and use the terminology ‘secondary electrons’. The emission of secondary electrons as the result of collisions of ion beams with metallic surfaces has been reviewed by a number of authors [37–44]. These references contain little recent effort devoted to gas-covered, practical discharge surfaces. Almost all of the recent beam work [42, 43] has been concerned with what we will call clean surfaces, with electron emission induced by multiply charged ions and with multiple electron emission induced by a single ion. Some measurements of electrons released as the result of incident neutral atom beams have been reported [45–51]. In view of the evidence for large fluxes of energetic atoms incident on the cathode [5, 52–55] of discharges it is important to model their contributions to electron production at the cathode and in the gas phase of Ar discharges. Because of our interest in cathode surfaces found in practical devices, we consider the yields of electrons at metal surfaces that are exposed to pure gases or gases that may result in oxidation or other contaminating reactions. We will find that at Ar+ energies above about 0.5 eV (E/n > 250 Td) the electron yields for cleaned metals in very pure Ar are close to the yields obtained for cleaned metals in ultra-high vacuum. Here E/n is the electric field to gas density ratio and 1 Td = 10−21 V m2 . We review data showing that the electron yields for metal surfaces exposed to oxidizing and/or other contaminating gases, i.e. for ‘dirty surfaces’, are very different than for clean metals and very pure Ar. Because of the sputtering that necessarily accompanies a cold-cathode discharge, we will attempt to correlate the very limited data on the effects of sputtering on the production of secondary electrons at metal surfaces. Opinions regarding the role of heavy-particle collisions in the production of electrons in the electrical breakdown of gases have varied from Townsend’s original proposal [30] that ion–atom collisions are the only source of negatively charged carriers (electrons) to the almost complete rejection of this idea in favour of ion-, metastable- and photon-induced electron ejection from the cathode as the source of secondary electrons [32, 33, 35]. The role of heavy-particle-induced ionization has been most extensively investigated in the case of breakdown [56–60] of hydrogen and the cathode fall region [61] in low-pressure hydrogen discharges. A few investigators have indicated the importance of such a process in the cathode fall [5, 61–63]. and in the breakdown [53] of low-pressure Ar. In particular, Neu [61] has performed an analysis similar to that of section 5, but for conditions appropriate to a discharge with a well developed cathode fall. R22 Section 2 of this paper reviews and correlates measurements of electron emission by ion and atom bombardment of clean and dirty metal surfaces using beam techniques. Measurements of effective electron yields per ion for clean and dirty surfaces obtained using swarm techniques, including low-current discharges, are summarized in section 3. Relevant published photoelectric yield data are correlated and fitted in section 4. In section 5 models of breakdown in the rare gases are extended and applied to experiment so as to indicate the roles of ions, fast atoms, photons and metastables in electron production at the cathode and in the gas phase. In appendix A we review two corrections for nonequilibrium effects near the cathode that are applied to our fluid models of electrons in discharges. The specific parameters used in our model of breakdown data for Ar are given in appendix B. In this paper we will sometimes use the terminology of breakdown to refer to the conditions for both quasi-static gas breakdown and steady-state, lowcurrent gas discharges [64]. 2. Argon ion and atom beam results In this section we consider measurements of the yield of electrons γ as the result of the bombardment of metallic surfaces with monoenergetic Ar+ ions γi , Ar atoms γa and Ar metastables γm at the low and moderate particle energies found in most discharges, i.e. below about 1000 eV. 2.1. Beam results for clean metals In this section we review the experimental results for what are called clean metal surfaces in the gas discharge field. For high-boiling-point metals this means that the surface has been heated to ≈2000 K in a very good vacuum and that the measurements are made with the surface at room temperature and under high vacuum. Such a heat treatment is often called ‘flashing.’ We are not concerned with what might be done by more thorough cleaning [65], because there appear to be no gas discharge experiments with such surfaces. Figure 1 shows experimental electron yields for Ar+ ions [46, 47, 66–71] and Ar atoms [46, 47] incident on various clean metal surfaces as a function of particle energy. We have shown essentially all of the yield data for clean metallic surfaces that we have found for energies below about 1 keV. At higher ion energies, these data are only a sample that available [37–40, 42–44]. A short summary of the cleaning procedure is given with the reference. Note that the two sets of values shown for Mo from the experiments of Vance [70] were obtained with two different high-temperature cleaning processes. For most surfaces the electron yield per ion shown in figure 1 is nearly independent of ion energy below about 500 eV. This constant yield is attributed to an Auger process called potential ejection [37, 42–44, 68]. The energy dependent portion of the yields at above several keV is called kinetic ejection and is less clearly understood [37–40, 42, 43]. Thresholds for kinetic ejection are associated with the onset of multiple electron ejection [43, 72]. The electron yields per ion measured [73–75] for most metals are about 0.1 at 1 keV. At energies near 1 keV Oechsner [73] finds a periodic Cold-cathode discharges and breakdown in argon 10 10 Electron yield per ion or atom Electron yield per ion or atom dirty metals 1 Ar+ 10-1 Ar 10-2 1 Ar+ 10-1 clean metals 10-2 10-3 Ar 10-3 10 102 103 10-4 10 104 Ion or atom energy (eV) + Figure 1. Electron yields for Ar and Ar beams incident on various clean metal surfaces versus particle energy. The solid symbols are for Ar+ and the open symbols are for Ar. The symbols, metals and references are: !, W, [68]; +,#, Mo, [46]; % , , Mo, [47]; ", Mo, [70]; , Mo, [66]; #, Au, [71]; ×, Cu, $ [67]; $, Pt, [69] and %, Ta, [69]. The curves drawn through representative values will be used in our model. • correlation of yield with atomic number [73]. The low, energy dependent yields at low energies for Ta may result from problems in cleaning [65, 69, 76]. In contrast to the near constant yields obtained for Ar+ ions at below 1 keV, the electron yield per fast Ar atom varies rapidly with energy with an effective threshold of roughly 500 eV. At high energies the yield per fast Ar atom approaches that for Ar+ ions. The overall energy dependence is that expected for kinetic ejection by a projectile with no available internal energy [37]. This behaviour is similar to that for ions with low ionization potentials, but comparable mass, which show a strong energy dependence and high threshold energies [37, 42, 43, 72]. Beam experiments in which the target is cleaned by ion sputtering have been reported [39, 44, 71, 73, 74, 77]. Experiments using sputtering followed by yield measurements at ion energies below 4 keV are shown for Au targets [71] in figure 1. Small but repeatable changes in electron yield are found [77] for 100 keV H+ ions when sputtered copper is annealed at near 600 K. This is approximately the temperature at which some forms of damage produced by sputtering are observed to annihilate [39]. How these results carry over to the Ar+ ions and the much lower energies of interest for discharges is not clear. For further discussion of sputtering effects see section 3.3. Measurements of γi using moderate current discharges are sometimes interpreted as beam experiments with clean surfaces. The crosses in figure 1 show very early results by Güntherschulze [67] in which the copper cathode was first cleaned by electron bombardment heating and the Ar+ ions were obtained by operating in very low-pressure discharge. Here the uncertainties include [37] the energy of the ions reaching the cathode from the discharge. In spite of the uncertainties, these results are consistent with more recent data. Recently, Böhm and Perrin [78] have measured γi for 102 103 104 Ion or atom energy (eV) Figure 2. Electron yields for Ar+ and Ar beams incident on various dirty metal surfaces versus particle energy. The open symbols are for Ar+ and the solid symbols are for Ar. The symbols, metals and references are: $ % , Pt, [69]; $, Ta, [69]; #, !, Au, [49]; , Cu, [75]; ♦, #, Cu, [45]; (, ", Ta, [80]; ×, W, [79]; %, brass, [81]; , unknown, [50] and , CuBe, [51]. The solid curves are plots of the analytical yield expressions for dirty surfaces for Ar+ and Ar, while the dashed curves are the representative yield curves for clean surfaces for Ar+ ions and Ar atoms from figure 1. ◦ • ions from an Ar plasma-processing discharge incident on Cu that had been cleaned by exposure to a H2 discharge. The ions are not mass identified, but are expected to be Ar+ with mean energies less than 60 eV. Their γi values of about 0.2 are about twice the beam results shown in figure 1. 2.2. Beam results for dirty surfaces Next we show in figure 2 the experimental values of the electron yield γi for Ar+ ions [45, 49, 69, 75, 79, 80] and γa for Ar atoms [45, 49–51, 80] incident on metal surfaces with varying degrees of surface exposure to oxygen [69], to water [50], to ambient gas [75], or to unspecified contamination [51]. We will refer to these surfaces as ‘dirty’, although the terms ‘practical’ or ‘laboratory’ surfaces are sometimes used. We have shown all of the absolute yield data that we have found for energies below about 1 keV. At higher ion energies, these data are only a sample of that available [37–40, 42, 43]. At energies above about 500 eV the differences in yields among metals are small compared to the differences from the clean metal values. The solid curves of figure 2 show fits to the experimental beam data that will be used in section 5.3 for comparison with swarm data. The dashed curves are averages through the experimental data of figure 1 and show the large changes in yield that typically occur when clean surfaces become oxidized or otherwise contaminated. The measurements for Ar+ (open points) in figure 2 show that for low energies (<150 eV) there is more than a two order of magnitude spread in the yields. Some of the low-energy data [45, 81] show relatively large yields, such as those observed on exposure of the metal [68, 69] to gases like H2 , N2 and CO, while data [50, 69, 79] obtained following exposure to O2 or H2 O show very small yields. Much of the R23 A V Phelps and Z Lj Petrović Ar+ data at energies below 100 eV suggest the relatively weak dependence on ion energy characteristic of potential (Auger) [37, 42, 43, 68] ejection. Parker [69] (Pt and Ta) and Hofer [82] (W) observed the change in γi during exposure of the surface to oxygen. We have not found similar direct demonstrations of the increase in γi with increasing contamination expected at Ar+ energies above 300 eV. One does not know whether to attribute the energy independent or weakly energy-dependent portions of the yield to patches of clean surfaces, to reduced penetration of the Auger electrons through the contaminants, to energydependent collisional ionization of material on the surface or to other mechanisms. The measured electron yields per incident fast Ar atom γa shown by the solid points of figure 2 have much the same energy dependence as for fast Ar incident on clean metals, but are shifted downward in energy by about a factor of ten so that the yields at a given energy are much larger. Haugsjaa et al [49] found the yields at high energies to be somewhat larger for fast Ar than for Ar+ . At very low atom energies they found structure in the yield curves for untreated surfaces that was interpreted as the result of the collisional ionization of two different adsorbed atoms. Thus, the electron yields (∼0.01 at 100 eV) are roughly consistent with the product of the surface density for a few monolayers of contaminants (∼1018 m−2 ) and an ionization cross section for fast Ar colliding with atoms (10−20 m−2 for Ar target). On the other hand, Kadota and Kaneko [51] did not find structure in the energy dependence of yield data at low Ar atom energies for contaminated Cu–Be. They found that for their fast atom source and incident Ar atom energies below 95 eV most of the electron yield appears to be the result of Ar metastable atoms produced in collisions of metastable Ar+ ions with Ar atoms. of the gas-phase process. The only potentially relevant data sets we have found are that for Ar metastables [86] colliding with O2 , where the cross section increases by an order of magnitude between 10 and 500 eV and that for Ar metastables [87] colliding with Na, where the cross section decreases by an order of magnitude between 0.01 and 10 eV. As for fast Ar atoms, the electron yields for Ar metastables are very roughly consistent with the probability of an ionizing collision of an Ar metastable with a column density of contaminant equal to that of a monolayer of gas. 3. Swarm experiment results In this section we review and analyse the available data for the effective electron yield per Ar ion as determined from pre-breakdown, breakdown and low-current discharge measurements. These will be collectively referred to as swarm measurements. For most of the data discussed in this section we adopt the long-standing convention [32, 33] of expressing the results of breakdown experiments as an effective yield per ion γeff . This convention can be regarded as a convenient bookkeeping procedure because, as we will see, the Ar+ ion is often not the principal source of electron emission from the cathode of an Ar discharge and is never dominant for Ar at low E/n for any surface of interest here. The convenience of γeff results from the simplicity of the model used to interpret breakdown and low-current discharge data, from the relatively high accuracy attributed to the ionization coefficient data used in the data reduction, from the rough correlation between the γeff ion and the true electron yields per incident particle and the relatively high sensitivity of γeff to changes in electron yield mechanisms compared to the sensitivity of the breakdown voltage. 3.1. Interpretation of swarm results 2.3. Metastable beam results Beam measurements of electron emission induced by the impact of Ar metastables on metallic surfaces following a wide range of surface treatments show an apparently uncorrelated large range of electron yields [83, 84]. Hagstrum [68] showed theoretically that electron yield per Ar metastable γm for clean metals, such W and Mo, should be the same as for the ions, i.e. the metastable is ionized near the surface and the ion ejects an electron by the Auger process. A review of experimental data [83] gives values for near thermal impact energies from 0.02 to 0.4, although measured yields for freshly flashed surfaces are close to the values for Ar+ . The measured [85] energy dependence for Ar metastables incident on Cu–Be exposed to O2 at incident energies from 250 to 600 eV is the same as that shown for ground state atoms in figure 2. Data for CuBeO contaminated by alkali vapours [86] shows a similar yield at energies above 500 eV, but is much higher at the lower energies. The change in electron yield with Ar metastable kinetic energy for other clean or dirty surfaces is unknown. A significant increase in yield with surface temperature has been observed [84]. If the production of electrons by metastable impact on metallic surfaces is regarded as a form of Penning ionization [83], then it may be instructive to note the energy dependence R24 The model used to obtain γeff from measurement of the steady-state current resulting from photoelectrons released from the cathode requires the fitting of an expression for the current as a function of the electrode spacing or gas density at fixed E/n. This procedure has been investigated by many authors [30–33, 35, 88, 89]. The model assumes that only Ar+ ions produce electrons at the cathode and that only electrons ionize Ar to produce electrons. The current i for electrode separations d greater than d0 is i= i0 fes exp[αei (d − d0 )] . 1 − γeff {exp[αei (d − d0 )] − 1} (1) Here i0 is the photoelectric current in vacuum, αei is the spatial (Townsend) ionization coefficient, exp[αei (d − d0 )] is the electron multiplication in crossing the gap and exp[αei (d − d0 )] − 1 is the number of ions arriving at the cathode per electron leaving the cathode. The delay distance d0 and the electron-escape fraction fes are empirical corrections applied to these local-field, fluid model results in order to account for electron nonequilibrium effects near the cathode. See the discussion in appendix A. Note that γeff includes fes , e.g., for metastable-dominated electron emission γeff = fes γm , where γm is the electron yield per metastable in vacuum. Cold-cathode discharges and breakdown in argon γeff = {exp[αei (d − d0 )] − 1}−1 . (2) When applying equation (2) for conditions such that the ionization coefficient for ion–atom collisions αii is comparable with αei , the interpretation of γeff is complicated by departures from the usual exponential growth of the current versus distance [91]. Failure to allow for the possibility of comparable electron- and ion-induced ionization coefficients has led to much debate and misleading conclusions in the literature [30, 31]. The model used to obtain secondary electron yields from transient measurements of the current resulting from a pulse of electrons released from the cathode or in the gap at voltages below that required for breakdown varies with the dominant electron production process at the cathode and therefore with time scale of the experiments. In general, the techniques involve separating the initial electron avalanche from the delayed electron and/or ion current. The results of these experiments are expressed in terms of the electron yields for individual processes, e.g., γi . Hornbeck [92] and Varney [93] were concerned with electrons produced by Ar+ and Ar+2 , while Engstrom and Huxford [94] and Molnar [95] were primarily concerned with electron production by metastable Ar atoms. See the original papers for the details. Note that these techniques do not appear to have been applied to what we call dirty metal cathodes. Also, transient techniques do not appear to have been used to separate the contributions of ion-induced electron emission from the cathode and ion–atom collisional ionization to avalanche initiation. A variant of the transient technique uses optical pumping to change the relative populations of the metastable and resonance states. The only application of this approach to Ar appears to be Molnar’s use of discharge illumination to determine the relative electron yields for metastables and photons [95]. See section 4 for results. Laser-induced pumping has been used for determining dominant cathode processes for Ne discharges, but no electron yield data were obtained [96]. An area of recent interest in which equation (1) is applicable is that of scintillation detectors [97] that usually operate at very low E/n and high pressures. Of particular interest here are experiments [98, 99] with pure Ar at E/n <10 Td, p ' 760 Torr and d ≈ 2 mm. For these conditions αei d * 1, d0 * d and equation (1) can be written as fes fes i = . (3) = i0 [1 − γeff αei d] [1 − γph fes αph d] Here γph and αph are the photoelectric yield and effective excitation coefficient for the dominant photon as discussed 10 Electron yield per ion The model used to convert electrical breakdown data to γeff data is the steady-state electron avalanche model known as the Townsend model in numerous texts [31–34, 89]. As discussed in some of our earlier papers [64, 90], a second source of such data is the maintenance voltage for a low-current discharge. Here the catastrophic current growth characteristic of breakdown is prevented by a large external circuit resistor. The breakdown and low-current maintenance condition is obtained by setting the denominator of equation (1) equal to zero. Solving for γeff gives 1 10-1 10-2 10-3 10 102 103 104 105 E/n (Td) Figure 3. Electron yields per Ar+ ion for various clean metal surfaces versus E/n as obtained from pre-breakdown swarm experiments. The data indicated by the solid points were obtained after heating (flashing) the cathode to a high temperature, while the open points show data obtained after cleaning by sputtering. The procedures for determining the yields from swarm measurements are reviewed in the text. The symbols, metals and references are: #, Mo, [95]; !, Mo, [93]; ", Mo, [100] and $ % , Cu, [88]. The dashed curve shows the γi values for Ar+ calculated from the solid curve of from figure 1. The solid curve shows the γeff vales predicted by our model in section 5.2 and includes all electron production processes. in section 5.1, i.e. the Ar2 continuum II near 127 nm. The second form of equation (3) shows that the high-pressure breakdown condition, i.e. the denominator equals zero, allows one to determine the product of γph and αph without knowing αei . As discussed in appendix B, the total excitation coefficient αph /n is very nearly equal to E/n divided by the excitation energy over a wide E/n range. Therefore, the voltage required for breakdown is inversely proportional to γph . This behaviour is in contrast to the logarithmic dependence of breakdown voltage on γi at the higher E/n where the exponential terms of equation (1) are dominant. From equation (3) and our lack of independent knowledge of γeff , we conclude that determining αei from steady-state swarm data at very low E/n will be very difficult. 3.2. Swarm results for clean metals The solid points of figure 3 show electron yields per Ar+ ion γi incident on various clean metal surfaces versus E/n from swarm experiments. These data were obtained using the pre-breakdown transient techniques [93, 95, 100]. Most of these experiments used cleaning of the electrodes by heating at high temperatures. They also used getters of chemically active metals sputtered or evaporated onto the tube walls to keep the gases clean [101]. These authors separated the ioninduced emission from the metastable and photon effects so that the yield values shown by the solid points in figure 3 are γi values. In these experiments the γi values are very nearly equal to the γeff values. The dashed line of figure 3 shows the γi results of beam experiments from figure 1 with the ion energy to R25 A V Phelps and Z Lj Petrović E/n conversion discussed in appendix B. Because of the relatively low ion and fast atom energies at the E/n of these experiments, i.e. ion temperatures [26] of less than 2 eV, the yield of electrons produced at these clean surfaces by fast atoms is negligible. See figure 1. The solid curve of figure 3 shows γeff values calculated using the model to be discussed in section 5 and the γi values from figure 1. The electron yields per ion γi determined from prebreakdown data for clean Mo surfaces for E/n from 300 to 1500 Td shown in figure 3 are in good agreement with the beam results shown in figure 1. Note that the three different experiments show a significant decrease in γi for E/n less than 250 Td, i.e. for calculated Ar+ ion energies less than about 0.5 eV. According to the model of section 5, this qualitative behaviour is consistent with the loss of Ar+ by conversion to Ar+2 ions and with a reduced electron yield per ion for Ar+2 . However, the E/n of the observed transition is higher than predicted by the model of section 5 by a factor of about 2.5 and the pressures are too low by roughly two orders of magnitude. An unexplored possibility is that the Ar+ ions have difficulty penetrating an adsorbed gas layer at low ion energies. The possibility that associative ionization in Ar∗ +Ar → Ar+2 + e collisions [102] becomes the dominant source of ions at these E/n seems ruled out by the estimates of the contribution of this process by Puech and Torchin [17]. Also, the highest-pressure mass spectrometer data available [103] appears to suggest a transition to Ar +2 at pressures significantly higher than those corresponding to breakdown at E/n near 250 Td, i.e. roughly 1.5 Torr. Electron yields per Ar metastable atom for clean metals have been determined by Molnar [95] from the transient measurements of pre-breakdown currents. For flashed Ta the yields per metastable vary from 0.0035 at 200 Td to 0.023 at 500 Td. For flashed Mo the yields increase from 0.02 to 0.06 for the same E/n range. At E/n ' 330 Td these values are in good agreement with the measured electron yields for Ar+ as predicted by theory [68]. 3.3. Swarm results for sputtered metal surfaces In this section we note a potentially significant correlation among γi values for various sputtered surfaces, i.e. for unannealed surfaces the γi values decrease with successive sputtering cycles. This topic takes on added importance with the recent interest in the modelling of sputtering discharges [5, 6] and other discharges operating at high current densities [2, 4, 9, 10]. The extreme γeff values determined by Kruithof [88] for Ar+ incident on Cu shown by the open points in figure 3 were obtained by analyses of pre-breakdown, steady-state current growth data after various intense sputtering treatments of the cathode using Ne discharges with no flashing. These data are shown as γi values because ion-induced electron emission is expected to be dominant for clean metals at these E/n. Of particular interest is the decrease in averaged γi values with successive sputtering periods. These results are to be compared with those of Varney [93] in which repeated cleaning of Mo by sputtering and flashing consistently gave the results shown by the solid inverted triangles in figure 3. These values equal those R26 obtained with flashing alone. Similarly, reproducible cathode fall voltages in Ne were obtained following discharge sputtering of Mo at high cathode temperatures [101]. The annealing temperature and duration required for a significant change in γi for heavily sputtered surfaces and Ar+ energies of interest for discharges is unknown. A small decrease in γi is observed [77] for 10 keV Ar+ incident on sputtered Cu when the surface is annealed at 200–300 ◦ C. The changes in γi with intense sputtering are qualitatively consistent with the changes observed [39, 104, 105] in surface morphology. 3.4. Swarm results for dirty metals The swarm results for dirty metal surfaces are presented in two different formats in order to make better contact with previous work. Firstly, we show in figure 4 measured breakdown [88, 106–112] and low-current discharge-maintenance voltages [90, 113, 114] as a function of the product of pressure and electrode spacing pd, in what is known as the Paschen curve [32, 33, 35, 88, 89]. Later, we reformulate these data so as to emphasize the production of secondary electrons. Generally speaking, the rise in breakdown voltage at high pd in figure 4 is attributed to a decrease in the ionization coefficient with decreasing E/n and decreasing electron energy. The rise at low pd is attributed to the difficulty in building up an electron avalanche when the number of collisions between electrons and the gas in crossing the gap is small. Some authors [32, 33] have attempted to relate these breakdown data directly to the electron yield at the cathode. The solid, dashed and chain curves of figures 4 and 5 are the results of the model presented in section 5 and will be discussed there. It should be kept in mind that for many of these experiments and all of our model results plotted, the electrode separation d is 1 cm, so that the numerical values of pd are equal to the Ar pressure p in Torr. In order to establish more closely the connection between breakdown data and electron production at the cathode, we show in figure 5 the effective yield of electrons per Ar+ ion reaching the cathode γeff . The points of figure 5 show breakdown results [88, 106–112] and low-current discharge results [90, 109, 113, 114] for γeff determined using equation (2). The pd values shown vary from 0.1 Torr cm at the highest [90, 111, 112] E/n to 2000 Torr cm at the lowest [107–109] E/n, corresponding to pressures from 0.1 Torr to about 760 Torr. As will be shown in section 5, the very large values of γeff at E/n ( 100 Td reflect the fact that at low E/n there are many vuv photons produced per ion by electron collisions with Ar. The large values of γeff at E/n ' 3000 Td result from the combination of an increasing electron yield per ion and the ionization produced by Ar+ ion and fast Ar atom collisions with Ar. 4. Photoelectric yields Figure 6 shows (open points) photoelectric yield γph versus wavelength for Au, Cu and Pt from the review by Weissler [115] for untreated and flashed metal surfaces. Also shown (solid and dotted points) are the results for untreated surfaces by Cairns and Samson [116] and others [115, 117, 118]. Cold-cathode discharges and breakdown in argon 1 105 Ar resonance lines Electrons per photon Breakdown voltage (V) 10-1 104 103 Ar2 band (continuum II) untreated 10-2 treated 10-3 10-4 102 10-2 10-1 1 102 10 103 104 10-5 10 102 Pressure × distance (Torr cm) Wavelength (nm) Figure 4. Breakdown and discharge maintenance voltages for Ar and various dirty metal surfaces versus pd, i.e. the Paschen curve for Ar. The symbols, metals and references are: $ % , Ni, [106]; , Cu, [88]; , brass, [107]; #, Cu, [108]; ♦, Ni, [109]; ∇, Au, [90]; ×, Steel, [168]; +, Cu, [169]; !, Au, [110]; #, SS, [169] and ", SS, [112]. Low-current discharge maintenance voltages are shown by , Au, [90]. !, Cu, [113]. (, Cu, [114]. The dashed and solid curves show the results of application of the full model of section 5. The chain curve shows the predictions when only Ar+ ions contribute to the production of secondary electrons. ◦ • Effective electron yield per ion 10 1 Figure 6. Photoelectric yield versus photon wavelength. The solid and open points are for untreated and flashed surfaces from [115]. The points with central dots are for untreated surfaces from [116]. The circles, diamonds and squares are for Au, Cu and Pt, respectively. The inverted triangles are for Cu films from [119]. The crosses are for Cu surfaces treated by laser irradiation from [120]. have been used to ‘activate’ (clean?) metallic surfaces and to determine the photoelectric yields shown by crosses (×) [121]. We have fitted an empirical expression to the γph data for heat-treated Au surfaces, i.e. γph = 10-1 10-2 10-3 10 102 103 104 105 103 106 E/n (Td) Figure 5. Effective electron yields per Ar+ ion incident on various dirty metal surfaces versus E/n. These γeff are calculated from breakdown and discharge operating voltages assuming that only Ar+ ions produce secondary electrons. The values are often very different from the true electron yield per ion. The symbols, cathode metals and references for the breakdown data are: $ % , Ni, [106]; , Cu, [88]; , brass, [107]; #, Cu, [108]; ♦, Ni, [109]; ×, Steel, [168]; +, Cu, [169]; !, Au, [110]; #, SS, [111]; ", SS, [112] and (, Cu, [114]. Effective yields from low-current discharge data are shown by , Au, [90]. The dashed and solid curves show the results of application of the model of section 5. ◦ • The triangles are for evaporated Cu films [119]. Data (+) for untreated Cu and steel surfaces are available [120] for wavelengths from 150 to 180 nm. Recently pulsed lasers 0.0015(360/λ − 1)4 [1 + (110/λ)5.5 ] (4) where the wavelength λ is in nm. This expression has been chosen to decrease with decreasing wavelength as do the more recent experiments [116]. For untreated Cu the yield falls off much more rapidly with increasing wavelength than for treated Au, corresponding to an apparent higher-energy photoelectric threshold. On the other hand, the photoelectric yield is systematically larger for untreated surfaces than for treated surfaces at below about 130 nm. Our empirical expression for untreated Cu is γph = 0.13 . [1 + (λ/110)13 ][(35/λ)3.3 + 1] (5) The results for other metals, e.g., W (not shown), are generally similar in shape, although varying somewhat in magnitude [115, 117, 118]. The wavelengths of the Ar resonance lines are indicated by vertical lines, with the 106.7 line being the one used in our breakdown model in section 5. The spectral feature (continuum I) near 110 nm occurs at essentially the same wavelength. Also indicated by vertical lines is the wavelength of the molecular Ar2 band designated continuum II near 127 nm in the model of section 5. We note that the photoelectron yield for the resonance lines and for continuum I is about three times that for continuum II. We will find in section 5.3 that the γph values shown in figure 6 for R27 A V Phelps and Z Lj Petrović untreated surfaces are larger than the values required to fit all but the largest experimental γeff results. The applicability of the photoelectric yield measurements made in vacuum to gas-covered metal surfaces is critical to our analysis. The only absolute yield data we have found for metal surfaces exposed to Ar are those of Molnar [95] for flashed Mo and Ta and incident Ar resonance radiation. He found photoelectric yields about a factor of three lower than the solid curve of figure 6. On the other hand, several authors [100, 122] appear to find no evidence of change in photocurrent when Ar gas is removed from and readmitted to their apparatus. 5. Modelling the effective yield data In this section we present a model of the Ar+ , fast Ar, metastable and photon transport and reactions in lowcurrent, uniform-electric-field discharges in Ar. The model will be used with the electron yield data of figures 1, 2 and 6 and, with adjustments to that yield data, to calculate γeff for comparison with the wide range of experimental data presented in figures 3 and 5. The model includes the relevant features of the steady-state limit of the model of ion, metastable atom and resonancephoton transport and of excited-state reactions developed in [123]. Figures 4 and 5 and figures 7–12 in sections 5.2 and 5.3 show the results of application of the present model to the calculation of electrical breakdown and lowcurrent, discharge-maintenance voltages in Ar. These results (a) compare calculations including all processes with the experimental breakdown data, (b) show representative spatial distributions of particle fluxes and densities, (c) illustrate the behaviour with E/n of the individual secondary electron production process and (d) show the relative importance of the various secondary electron production processes. The effects of some of these processes on breakdown voltages in Ar have previously been calculated using a less-complete model and are shown figure 11 of [53]. 5.1. Details of model In this section we develop a mostly analytic, approximate model to describe the steady-state fluxes of electrons, Ar+ ions, fast Ar atoms, metastable atoms and resonance photons in a spatially uniform electric field. This model can be used to predict the role of these processes in steady-state-breakdown, optical-emission and current-growth experiments. The assumptions of the model are that: (a) The electron flux density leaving the cathode is determined by $e (0) = fes [γi $i (0) + γi2 $i2 (0) +γa $a (0) + γm $m (0) + γph $ph (0)]. (6) Here $e , $i , $i2 , $a and $ph are the flux densities of electrons, Ar+ , Ar+2 , fast Ar and photons evaluated at the cathode at z = 0; fes is the fractional escape of electrons emitted from the cathode into the gas and γi2 is the electron yield per ion for Ar+2 . The γph $ph term is actually a sum over the various lines and bands making up the photon flux reaching the cathode. R28 (b) The spatial growth of the electron flux is described by a fluid model in which a spatially constant spatial ionization (Townsend) coefficient is determined by the spatially independent E/n. At moderate and low E/n this is a good approximation for most of the gap. See appendix A. At very high E/n this constant ionization rate replaces our previous approximation to the nonequilibrium behaviour of the electrons [124] in which the electron energy and subsequent ionization rate are determined by the solution of an electron energy-balance equation, i.e. equation (3) of [53]. The present assumption is justified by emission experiments [125] and Monte Carlo solutions [125, 126] that yield an approximately exponential growth of electron flux and electron excitation with distance even at very high E/n. The differential equation for the electron flux is d$e = αei $e + αii $i + αai $a . dz (7) Here αei , αii , αai are the spatial (Townsend-like) coefficients for ionization of Ar by electrons, by Ar+ and by fast Ar; and z is the distance measured from the cathode. The values of the E/n-dependent αx coefficients are discussed in appendix B. At E/n < 3000 Td only electron impact ionization of Ar is important and equation (7) reduces to equation (12) of [123] in the limit of α1 = αei , β = 0 and λ = 0. Electron loss processes such as electron–ion recombination and electron production processes such as electron–excited atom collisions can be neglected for the low current densities of interest here. Because of the low currents we have neglected space-charge distortion of the electric field. (c) The Ar+ flux decreases as one moves away from the cathode, i.e. it increases with distance from the anode, as the result of ionization of Ar by electrons, Ar+ and fast Ar so that d$i = −αei $e − αii $i − αai $a . dz (8) The ion energy distribution is assumed to be a onedimensional Maxwellian with a ‘temperature’ T+ determined by the E/n and the charge transfer cross section [27, 127]. Inelastic Ar+ –Ar collisions are assumed negligible compared to charge transfer collisions. We neglect the deviations from the theoretical ion energy distributions found [128] at high ion energies and high E/n. The ionization coefficients αxi in equations (7) and (8) and the ion-induced excitation coefficients are averages of the corresponding cross section over the ion-energy distribution [53]. Formulae for these coefficients are given in appendix B. Very few ions are expected to be emitted from the anode as the result of electron bombardment [129]. As in [53], we can approximate the ionization produced near the anode by backscattered and reflected electrons [130] by using a finite Ar+ flux as the boundary condition at the anode, i.e., $i (d) = δ$e (d). (d) In this model we neglect the production of Ar+2 in three-body collisions and its dissociation in collisions with Ar. This simplification is possible because γi2 $i2 /γi $i is predicted to be significant only at E/n < 100 Td where photons are the dominant source of secondary electrons Cold-cathode discharges and breakdown in argon [131]. To take this low E/n process into account one would use relations such as equation (13) of [123]. (e) The production of fast Ar flux is by symmetric charge transfer collisions of Ar+ with Ar. We assume that these fast atoms have the same energy and direction as the ions. We assume that once the fast atom has an energy-loss collision, i.e. an elastic viscosity collision [27] or an inelastic collision, it is effectively thermalized. Use of these approximations has been found to give rather good agreement with observed emission distributions [53, 54] and with Monte Carlo models in most cases tested [126]. It would appear that the neglect of angular scattering leading to too few collisions in the model roughly compensates for the too-rapid loss of fast atoms at every elastic viscosity collision. The resultant fast-atom flux equation is d$a = −αct $i + (αaa + αai + αar + αam )$a . dz (9) Here αct , αaa , αar and αam are the spatial reaction coefficients for symmetric charge transfer, elastic viscosity collisions, resonance state excitation and metastable state excitation collisions. Analytic formulas for these coefficients are given in appendix B. (f) Our model for the excited states of Ar near 11.5 eV is a simplification that is only intended to predict the fluxes of metastables and uv radiation that reach the cathode in our steady-state breakdown and low-current discharge experiments. We replace the two resonance states by the lower resonance state (4s 3 P1 ) and replace the two metastable states by the lower metastable state (4s 3 P2 ) [132]. The total excitation rates for the two resonance levels [17] are assigned to the model resonance state. Similarly, for the metastable states. We assume that the molecular states associated with each of the lower two atomic states [133] are formed in three-body collisions of the excited atom with two Ar atoms with the measured rate coefficients [134]. The vibrationally excited molecules are assumed to relax immediately to their ground vibrational states. We neglect collisional coupling between the resonance and metastable states and between the molecular states. All of the excited Ar2 molecules relaxing to the lower vibrational states are assumed to radiate in the 127 nm band designated continuum II, [133], i.e. their quenching is neglected. In this model, the observed continuum I near 110 nm is absorbed into the wings of the resonance line. This very much simplified four-level model is intermediate in complexity between the afterglow models of Millet et al [134] and of Bretagne et al [14]. A critical advantage of this simple model is that, knowing the rates of production by electrons, ions and fast atoms, we can solve for the metastable atom and resonance-photon fluxes independently. We have compared the steady-state predictions of this four-level model with those of a model [135] with four atomic levels and four molecular levels when the electron excitation is nearly independent of position as at high Ar pressures [109]. For high-purity Ar ('0.999 99) the difference between the photoelectric flux leaving the cathode estimated using the four-level and eight-level models peaks at ≈20% for pressures near 10 Torr, where resonance radiation is the dominant photon flux. For low-purity Ar (0.999) the estimated error is (10% at all pressures. (g) Our treatment of resonance excitation is based directly on the model of [123], as verified for Ar in the current growth experiments by Menes [108]. The basic equation governing the steady-state density of excited atoms in the resonance state is [28] ! d nr (z. )K(|z − z. |) dz. (Ar + q)nr (z) − Ar 0 = αer $e + αar $a + αir $i = a1 + a2 exp(s2 z) + a3 exp(s3 z). (10) Here nr (z) is the resonance atom density, Ar is the radiative transition probability, q is the frequency of molecule formation collisions and K(|z − z. |) is the radiative transmission kernel. The right-hand side is the resonance state collisional production rate due to electron–Ar atom collisions αer , fast Ar–Ar atom collisions αar and ion–Ar atom collisions αir . Here s1 = 0, s2 and s3 are the roots of the third order determinant of the system of equations found by taking the Laplace transforms of equations (7)–(9) and a1 , a2 and a3 are the respective algebraic coefficients for the components of the resonance state production. The expressions used for the αxr /n for resonance state excitation by electrons, ions and atoms are given in appendix B. Absorption measurements [136] show that the theoretical dipole-dipole resonance line profile used in this model should be valid for d ∼ 1 cm and p ( 200 Torr. At higher pressures the calculated excess absorption in the red wing will be partially offset by the collision-induced radiation effects [134]. Also, at 200 Torr collisional conversion to the radiating molecules dominates resonance radiation transport, i.e. q considerably exceeds the effective resonance state lifetime AI from [123]. The resonance photon flux reaching the cathode $rph is [28] ! ! ∞ d $rph = Ar nr (x) dx 0 K(|y|) dy. (11) x Rather than solve equations (10) and (11) numerically we have made use of the solutions obtained in [123]. When the exponential terms increase toward the anode, i.e. negative values of s2 and s3 , the resultant resonance atom density terms have a spatial dependence such as shown in figure 1 of that reference. The resonance flux transmitted to the cathode is that described in connection with figure 2 and equation (25). For an exponentially decreasing source term, the distance scale in figure 1 is reversed and equation (25) is modified by replacing the last factor by (2 − 1/[1 + 0.175(αd)2/3 ]). (h) The effects of Ar metastables in the 3 P2 and 3 P0 states of Ar are approximated by a single metastable state [14, 95, 123, 134] that is produced by electron, ion and fast-atom excitation of Ar. Metastable atoms are lost by diffusion to infinite parallel-plane electrodes, by collisioninduced emission and by three-body collisional conversion to radiating molecules. The steady-state metastable rate equation is − (nDm ) d2 nm (z) = αem $e + αam $a + αim $i n dz2 −[k2m n + k3m n2 ]nm (z). (12) R29 A V Phelps and Z Lj Petrović −[αa αii + αai αi ]/(s2 s3 ) −{exp(s3 z)[(αa + s3 )(αii + (δ − 1)s3 ) −1 +αai αi + (δ − 1)s3 ]}[s3 (s3 − s2 )] $i = {δ[exp(s2 z)(αai + αa + s2 ) $t − exp(s3 z)(αai + αa + s3 )]}(s2 − s3 )−1 " (αai + αa + s2 ) exp(s2 z) +αei s2 (s2 − s3 ) # (αai + αa αauv ) exp(s3 z)(αai + αp + s3 ) + + s2 s3 s3 (s3 − s2 ) $a = {αct [δ(exp(s2 z) − exp(s3 z))s2 s3 $t +αei (s2 − exp(s3 z)s2 + (exp(s2 z) − 1)s3 )]} ×(s2 (s2 − s3 )s3 )−1 (13) −4(αa αei − αa αii − αai αrc )]1/2 ). Normalized fluxes and densities 1 Γi 10-1 Γe nr´ /1000 -2 10 nm´ /1000 10-3 0 2 4 6 8 10 Distance from cathode (mm) Figure 7. Calculated spatial dependence of the normalized fluxes (solid curves) of electrons $e , Ar ions $i and fast Ar atoms $a and of the normalized densities (dashed curves) of metastables n.m and excited atoms in the resonance state n.r . The model does not distinguish between Ar+ and Ar+2 . For clean metal surfaces and calculated breakdown conditions at E/n = 50 Td, for which pd = 63 Torr cm. 10 Γa 1 Γi 10-1 Γe nm´ /10 10-2 nr´ /10 10-3 0 2 4 6 8 10 Distance from cathode (mm) (15) (16) The source terms for the production of metastables and resonance atoms resulting from equations (13)–(15) have the form of a constant plus two exponential terms. The exponent s2 obtained with the negative sign in equation (16) is negative for all E/n and our parameters results in growth toward the anode. The exponent s3 obtained with the positive sign in equation (16) may be positive at high E/n corresponding R30 Γa (14) where αa = αai + αap + αaq and αi = αii + αct . The roots of the determinant for equations (7)–(9) are 0 and s2 , s3 = 1/2(−αa − αei + αii ∓ [(αa + αei − αii )2 10 Normalized fluxes and densities Here αem , αam and αim are the spatial coefficients for metastable excitation by electrons, fast Ar atoms and Ar+ ions, nDm is the metastable diffusion coefficient at unit Ar density, k2m is the rate coefficient for collision-induced radiation, k3m is the three-body rate coefficient for excited molecule formation and n is the Ar density. The boundary conditions for this equation are that the metastable density is zero at the electrodes and the desired quantity is the metastable flux at the cathode, −(nDm /n) dm(z)/dz. Collision-induced radiation from the metastable state results in nonresonant emission near λ = 110 nm, i.e. continuum I. See equation (18) of [123]. The photoelectric yields at 110 nm from equations (4) and (5) essentially equals that for 106.6 nm. At the pressures for which this process is important, i.e. (10 Torr, the Ar is optically thin [136]. At higher pressures the excitation reaches the electrodes via by molecule formation, vibrational relaxation and radiation of continuum II. (j) Equations (7)–(9) can be solved independently of the equations governing resonance atoms and the products of resonance atom collisions with Ar. We have done this analytically using the Laplace transform technique [137]. The results for the electrons, ions and fast atom fluxes at the cathode are: $e = −{exp(s2 z)[(αa + s2 )(αii + (δ − 1)s2 ) $t +αai (αi + (δ − 1)s2 )]}[s2 (s2 − s3 )]−1 Figure 8. Calculated spatial dependence of the normalized fluxes (solid curves) of electrons $e , Ar+ ions $i and fast Ar atoms $a and of the normalized densities (dashed curves) of metastables n.m and excited atoms in the resonance state n.r . For clean metal surfaces and calculated breakdown conditions at E/n = 20 kTd, for which pd = 0.17 Torr cm. to growth toward the cathode resulting from heavy-particle excitation and ionization. (k) The resultant flux equations for metastables and for the resonance photons are much too complicated to publish, but are available on request [138]. Instead we show representative solutions for the metastable and resonance state densities in figures 7 and 8. These figures will be discussed in section 5.2. Cold-cathode discharges and breakdown in argon (l) The radial losses of all species are neglected in our one-dimensional model. The lateral losses of charge particles and of fast atoms will be small because of their high axial energies. For nonresonant photons and the experiments cited here the solid angle of the cathode from the centre of the anode is from 60 to 70% of that for infinite parallel planes. The effects of the decrease of the solid angle for photons emitted off axis is expected to be reduced because of the Bessel function distribution of current density in the fundamental diffusion mode [139]. We will use 0.3 for the fraction of the nonresonant photons reaching the cathode. The effects of a finite tube radius on the transport of resonance excitation appears to be small for typical discharge electrodes [140]. (m) The quasi-static breakdown and low-current discharge maintenance condition is that the sum of the fluxes of electrons produced at the cathode by the various incident species is equal to the flux of electrons leaving the cathode, i.e. that the electron flux given by equation (6) equals that given by equation (13) evaluated at the cathode. Formulation and the numerical solution of the very complex algebraic equations for this model were carried out on a personal computer [137]. The numerical solutions were obtained by assuming a value of E/n, using analytic forms for the collision coefficients and yields from appendix B, assuming an electrode spacing d and solving iteratively for the value of p, i.e. the eigenvalue, that satisfied the breakdown and low-current discharge condition. While Monte Carlo (MC) techniques for solution of the particle and photon fluxes [3, 5–7, 22, 126, 141] would avoid complex algebra and allow fewer assumptions, the number of MC calculations would be large and time consuming [142]. Tests for the contributions of the corrections for nonequilibrium effects at the cathode, i.e. electron backscattered to the cathode and delay in the onset of ionization, were made in the present analyses. See appendix A for the backscattering correction used. The effects of the delays in the onset of ionization on the derived γeff were small enough compared to the uncertainties in the electron yields so that they were not included in the final analyses. 5.2. Application of model to clean surfaces The results of applying the model just described to the calculation of conditions for breakdown and pre-breakdown for clean metal cathodes are presented in figures 7–10. Because of the very limited range of E/n for which experimental γeff data for clean surfaces is available, we have used the electron yield data estimated from the beam experiments and listed in appendix B without adjustment. Also, we have simplified the model by neglecting electron reflection at the anode. Figures 7 and 8 show the calculated spatial dependences of the normalized fluxes (solid curves) of electrons $e , ions $i and fast atoms $a . The calculated resonance and metastable state densities (dashed curves) are plotted as the dimensionless quantities n.r (z) = [(q + AI )d/ $e (d)]nr and n.m (z) = [νd/ $e (d)]nm , respectively. Here $e (d) is the electron flux density at the anode and ν = [k2m n+k3m n2 ]+νD , where νD = (3.14/d)2 (nDm )/n is the fundamental-mode, diffusion-loss frequency for metastables. For breakdown or a low-current discharge at E/n = 50 Td, the model gives pd = 63 Torr cm and the spatial distributions in figure 7. The electron flux grows exponentially with distance from the cathode, while the Ar+ and fast Ar atom fluxes increase toward the cathode from zero at the anode. The electron flux is normalized to unity at the anode. In this low-E/n case the ion and fast atom temperatures are less than 0.1 eV. As a result of production by electrons only and a loss dominated by collisions, the metastable atom density n.m (z) grows exponentially with distance except very close to the electrodes. Because of radiation transport, there are significant departures from an exponential growth for the resonance atom density n.r (z), just as shown in figure 1 of [123]. For the very high E/n of 20 kTd, the calculations for a clean cathode give pd = 0.17 Torr cm and the spatial distributions shown in figure 8. The calculated $e (z) curve shows significant departures from exponential growth because of ionization by Ar+ ions and fast Ar atoms produced by charge transfer from Ar+ . In this high-E/n case the ion and fast atom temperatures are about 50 eV. The heavyparticle collisions also result in a continued growth of ion and fast atom fluxes as one approaches the cathode rather than the saturation seen in figure 7 at low E/n. The excitation by heavy particles near the cathode and electrons near the anode results in a relatively flat resonance state density. The excitation by fast atoms and electrons and the large diffusion loss of metastables at low pressures combine to give an n.m (z) profile that is roughly a sinusoid that is shifted toward the cathode rather than the anode. We next consider what we call ‘limiting cases’ of γeff as a function of E/n resulting from models in which specific cathode directed particles or photons are assumed to produce all of the ‘secondary’ electrons required to cause breakdown or to maintain the discharge. The solid curve in figure 3 and repeated in figure 9, shows the calculated γeff when all processes are included in the model. The short-dashed curve labelled γi in figure 9 gives the values of γeff obtained when γi was set to a value of 0.07 and all of the other yields and the heavy-particle collision cross sections were set to zero [143]. As expected, the short-dashed curve agrees with the assumed γi value for E/n > 1000 Td, but drops below γi for E/n < 1000 Td because of the effects of electron backscattering to the cathode. See appendix A. The long-dashed curve shows the γeff values calculated when only photoelectric production of electrons at the cathode is included in the model. Here we use the photoelectric yield given by equation (4). Although no experimental γeff data are available for clean metals at E/n ( 50 Td, the model shows that photoelectric processes determine the γeff values. At high E/n the calculated γeff values and breakdown voltages are determined by secondary electron production by heavy particle ionization, rather than by secondary electron production at the cathode. The curves labelled γm and γa in figure 9 show that the calculated γeff values are small when only metastable atoms or fast atoms produce secondary electrons. Although the curves of γeff presented in figure 9 show which secondary electron production processes are dominant at various E/n, they do not show the contributions of R31 A V Phelps and Z Lj Petrović Effective electron yield per ion 10 heavy particle ionization only 1 all processes 10-1 γph 10-2 γi γm 10-3 10 102 γa 103 104 105 E/n (Td) Figure 9. Calculated effective electron yields versus E/n for clean surfaces. The curves are calculated for various limiting assumptions for the electron yields at the cathode and by heavy- particle ionization and excitation using electron yields from appendix B. The labels of the various curves are: γi , Ar ion-induced electrons at the cathode with no heavy-particle excitation or ionization; γa , Ar fast-atom-induced electrons at cathode with no heavy-particle excitation or ionization; heavy-particle ionization with no electron production at cathode; γph , vuv-photon-induced electrons at cathode with heavy-particle excitation of resonance atoms; γm , Ar metastable induced electrons at cathode with heavy-particle excitation of metastables, and all processes, all cathode emission and heavy-particle ionization and excitation processes. 10 vuv continuum Fractional electron emission various electron production processes at the cathode. These contributions are indicated by curves of figure 10 for the clean surfaces of figures 3 and 9. We see that Ar+ ions are the dominant source of electrons at the cathode for E/n ' 60 Td. Resonance photons produce most of the electrons needed for breakdown for E/n near 40 Td for a clean cathode. At lower E/n the nonresonant vuv photons emitted by excited Ar2 formed from resonance atoms and metastables, i.e. Continua I and II, are dominant. Because of the importance of heavyparticle ionization, the curves of figure 10 should not be interpreted as showing that the production of avalancheinitiating electrons at high E/n is dominated by ion impact on the cathode. 1 metastables 10-1 10-2 atoms resonance photons 5.3. Application of model to dirty surfaces The results of applying the models of section 5.1 to the calculation of conditions for breakdown and for low-current discharge maintenance for dirty surfaces are presented in figures 4, 5, 11 and 12. The spatial distributions of fluxes and excited state densities for E/n of 50 Td and 20 kTd are not shown, as they are very nearly the same as those presented for clean cathodes in figures 7 and 8. 5.3.1. Comparison with experiments. The results of the model are compared with experiment in figures 4 and 5. Here the solid and short-dashed curves are calculated using all of the surface and volume secondary electron production processes of our model. The short-dashed curves are calculated using the γi from the solid curve of figure 2 and γph from the dashed curve for untreated surfaces from figure 6. The solid curves are calculated using 10% of the photoelectric yield γph shown by the dashed curve of figure 6. R32 ions 10-3 10 102 103 104 105 E/N (Td) Figure 10. Calculated fractional contributions of various processes to electron production at the cathode for Ar at breakdown for clean surfaces. The model is that of section 5 using the parameters discussed in appendix B. The processes are indicated by the labels attached to the curves. The chain curves are calculated using γi = 0.07, as used in several recent low-current models [22, 139]. In this case, all other γ values and heavy-particle ionization rates are set to zero. At large pd the calculated voltages of figure 4 are nearly coincident for all three of these models. The small spread of voltages at fixed pd arises from the rapid decrease in αi /n Cold-cathode discharges and breakdown in argon Effective electron yield per ion 10 heavy particle ionization only all processes exp’t. γph 1 all processes adj. γph 10-1 γa 10-2 γph 10-3 10 γi 102 103 104 105 E/n (Td) Figure 11. Calculated effective electron yields versus E/n for dirty surfaces. The curves are calculated for various limiting assumptions for the electron yields at the cathode and by heavy- particle ionization and excitation using our best estimates of these electron yields from figure 1 and the model of section 5. The labelling of the curves is the same as in figure 9 except that the two curves labelled all processes are calculated for γi and γph values based on two different approximations to the data of figures 5 and 6. See text for details. Fractional electron emission vuv continuum resonance photons ions atoms 1 metastables 10-1 10-2 10 102 103 104 105 E/N (Td) Figure 12. Calculated fractional contributions of various processes to electron production at the cathode for Ar at breakdown for dirty surfaces. The model is that of section 5 using the parameters discussed in appendix B. The labels attached to the curves indicate the electron production process. with decreasing E/n. If these data are replotted as a function of E/n they cover a wide range of breakdown voltage or pd values. When converted to γeff , as shown in figure 5, the data at fixed E/n are spread over more than an order of magnitude. The comparison of the solid and dashed curves with the points of figure 5 leads to the conclusion that the γph values to be used in our model range from the values shown for untreated surfaces in figure 6 to more than an order of magnitude smaller. Unfortunately, there is a large uncertainty at very low E/n in the αei /n values used to calculate γeff from experimental breakdown and discharge maintenance data. This uncertainty is evidenced by the differences in αei /n reported by several groups [88, 109, 144] i.e. the diverging γeff values shown in figure 4 of Golden and Fisher [109]. The lower set of γeff values from this reference is the result of our analyses of their breakdown voltage data and the upper set is from their fitting of equation (1) to current growth data. An alternate approach at very low E/n is to use equation (3) to analyse data such as those of figure 3 of Golden and Fisher [109]. If the fraction of the metastable and resonance state excitation converted into continuum II photons is constant over the pressure range of the experiment, equation (3) shows that at constant E/n the quantity (1 − fes I0 /I ) is a linear function of pressure. The experimental data of figure 3 of [109] fit this expression very well and yield the product fes αph γph /p = 0.0013 Torr−1 . Using our values of αph and fes and equation (B28) for the fraction of photons reaching the cathode, we find γph = 0.006. This photoelectric yield is about a factor of three smaller than we expect from figure 6. In this experiment, the effects of known impurities [109] are estimated to reduce the vuv emission by less than a factor of two. How much more of the discrepancy can be attributed to impurities is unknown. In principle, equation (3) should apply to steady-state experiments under many scintillation detector conditions, e.g., the normalized current data of figure 4 of [99]. At pd ≈ 0.07 Torr cm and E/n ≈ 1000 Td, the assumption of γi = 0.07 (chain curve) leads to much too small a breakdown voltage and to much too large γeff values. This comparison also shows that although our adjusted γi values are roughly constant for ion energies between 10 and 100 eV in figure 2, the decrease in γi at lower energies is crucial to the fit to experiment. At pd near 0.03 Torr cm, the calculated V versus pd curve (chain curve) for γi = 0.07 becomes nearly vertical. This behaviour is the result of a maximum in αei /n versus E/n. We see from this example that the assumption of constant γi or R33 A V Phelps and Z Lj Petrović constant γeff , common in many current Ar discharge models, [5–8, 12, 14, 15, 22, 141], is a poor approximation. At the high E/n where heavy-particle collisions are important, the model results shown in figures 4 and 5 are in satisfactory agreement with the badly scattered data. The inclusion of ionization by electrons backscattered from the anode would lower the breakdown voltage and raise the γeff values. It remains to be seen whether a better model of ion and fast atom motion, i.e. a Monte Carlo model incorporating angular scattering and elastic energy loss by heavy particles [5, 126, 141] and electron backscattering, will significantly change the comparisons with experiment. As indicated in equation (3) and in the equations in [123], the photoelectric feedback term appears as a product of the photoelectric yield and the effective vuv excitation coefficient. Therefore an increase in photoelectric yield to the expected values could be compensated for by a decrease in effective excitation coefficient caused by errors in our kinetics model. We have not found useful measurements of the vuv emission efficiency for Ar that test the model. Finally, including the delay in the onset of ionization near the cathode, as discussed in appendix A, changes the calculated γeff values by amounts much less than the scatter in the experimental data. 5.3.2. Importance of various processes. The discussions of the relative importance of the various process in this section for dirty surfaces are similar to those for clean surfaces in section 5.2. We will therefore emphasize the differences. The values of the coefficients used are from the appendices or from section 4. In figure 11 we show γeff values for dirty surfaces when we limit the kinds of cathode directed particle or photon that produce secondary electrons. The dotted curve labelled γi gives the values of γeff obtained when the γi was set equal to the values given by the solid curve labelled Ar+ in figure 2 and then averaged over the one-dimensional energy distribution of Ar+ energies. Other secondary processes were omitted. For E/n > 1000 Td this curve agrees with the assumed average γi , but is lower for E/n < 500 Td because of electron backscattering to the cathode. Comparison of the dotted curve of figure 11 and the short-dashed curve of figure 9 shows that our adjustment of γi to fit the breakdown data for dirty surfaces results in much lower values than for clean surfaces at E/n < 10 kTd or Ar+ energies below 5 eV. The dot–dashed curve labelled γa of figure 11 shows calculated γeff values when the yield for fast atoms equals the values shown by the solid curve labelled Ar in figure 2. Again, all other secondary electron production is zero. Most of the increase in γeff for fast atoms at high E/n compared to that for ions is caused by the large number of fast atoms produced in charge transfer collisions between Ar+ ions and Ar atoms. As expected from the data of figures 1 and 2, fast Ar atoms are much more effective for dirty surfaces than for clean surfaces. As pointed out previously, [53, 58, 145], the ionization of Ar by fast Ar and by Ar+ is an important electron production processes at very high E/n. This importance is shown in figure 11 by the double-dot–dashed curve labelled heavyparticle ionization. This curve gives the values of γeff R34 when ionization by Ar+ and fast Ar were included, but other secondary electron production process were omitted. The resultant γeff values for E/n near 10 kTd agree with the experimental values (see below) to within the large scatter in the data. The γeff curve for ionization by ions only (not shown) is less than 20% of that shown for ionization by atoms and corresponds to that originally proposed by Townsend to explain breakdown [30]. In these heavy-particle ionization cases, we are particularly conscious that the γeff has become a bookkeeping factor that is a measure of the voltage across the discharge. Because of the large contribution of heavyparticle ionization, it is difficult to determine electron yields for ions and fast atoms striking the cathode from breakdown data at very high E/n in Ar. The contribution to γeff for dirty surfaces of photons produced by excitation of the resonance state and the molecular continua are similar to those for clean surfaces. The main difference is that for dirty surfaces the resonancephoton contribution is large to higher E/n because of the lower values of γi . The calculated values of γeff obtained assuming that only metastable Ar atoms produce electrons at the cathode with a constant yield of 0.02 electrons per metastable are too small to show in figure 11. The relative contributions of the various processes at the cathode found using the breakdown and discharge maintenance parameters from the fitted curves of figures 4, 5 and 11 are indicated in figure 12. These results show that ions and fast atoms are the dominant source of electrons at the cathode at E/n above about 400 Td. Resonance photons produce most of the electrons needed for breakdown for E/n between 40 Td and 1000 Td. At lower E/n the nonresonant photons emitted by excited Ar2 formed from resonance atoms, i.e. continuum II, are dominant. It should be noted that because of the importance of secondary electron production by resonance photons at low E/n, with their effective lifetime [28] varying as d 1/2 , the conventional scaling [32] of the breakdown or low-current discharge voltage (or E/n) with pd no longer applies. The pd (Paschen) scaling is also lost because of the n2 dependent terms in the excited state destruction. In order to test the scaling of γeff with electrode separation d for a fixed E/n, we repeated the calculations for the conditions of the dashed curves of figures 5 and 11, except that the distance was 4 cm instead of 1 cm. We find that the new γeff are barely distinguishable from those shown. Effects of the change are more noticeable for the fractional contributions to the secondary electron production, where there is a shift toward higher E/n of the transition from dominance by resonance photons to dominance by continuum radiation. Presumably this is caused by a shift in the competition between resonance radiation decay and Ar2 formation, with larger distances reducing the escape rate for resonance photons. While the departures from pd scaling are predicted to be small compared with the overall scatter in the experimental results shown in this paper, the effects of such departures in the scaling on γeff may have been observed in the experiments of Golden and Fisher [109]. Cold-cathode discharges and breakdown in argon 6. Summary and discussion In this paper we have compiled and analysed over 60 years of data concerned with cathode processes of importance in quasi-static, uniform-electric-field breakdown and in lowcurrent, steady-state discharges for Ar. We have successfully applied an updated model that describes the important surface and gas phase collision processes for the whole range of experimentally available data, i.e. E/n and pd values from 15 Td and 2000 Torr cm to 100 kTd and 0.06 Torr cm. While there are many details that require further work, the analysis delineates quantitatively the regions of importance of several long-debated processes responsible for secondary electron production by cathodedirected species, e.g., electron emission at the cathode induced by ions, fast atoms, metastables and photons versus electron impact ionization by ions and fast atoms. Our review of electron yield data and our model of the role of various secondary electron production processes in Ar has shown that: (1) The yields of electrons per ion, fast atom or photon at a metallic cathode of a discharge in Ar are highly dependent on the condition of the metal surface. Unless there are compelling reasons to the contrary, this means that any estimates of electron yield made by a modeller on the basis of the cathode preparation technique should be regarded as tentative and subject to adjustment. The data presented in this paper should allow the modellers to make informed initial estimates as to the electron yields at the cathode of Ar discharges. In particular, these data should dispel the commonly held assumption that the electron yield per Ar ion is independent of E/n and equal to the value found for flashed metals, e.g., that found by Hagstrum [68] for molybdenum or by Oechsner [73] for copper. (2) For most metal surfaces cleaned by repeated sputtering and flashing and exposed to very pure Ar one should expect a yield per ion γi of about 0.1 that increases slowly with E/n only at extremely high E/n. At E/n below about 100 Td our model indicates that photon-induced electron production at the cathode becomes dominant. Because the high electron yields per ion for clean surfaces also apply to Ar metastable atoms, transient current growth and decay experiments at intermediate E/n and long times show time constants characteristic of metastable diffusion. (3) For dirty surfaces the measured yields per ion γi at ion energies below 200 eV corresponding to E/n at the cathode below about 70 kTd can vary by orders of magnitude depending on the surface condition. The beam studies of γi shown in figure 2 need to be extended to much lower ion energies, e.g., energies well below 2 eV, to be applicable to breakdown at moderate E/n. At ion energies above 300 eV, corresponding to E/n > 100 kTd, the yields are remarkably independent of the substrate and the nature of the gaseous contaminant, but are much larger than for clean surfaces. (4) At E/n between 10 and 100 kTd our present model confirms the previous analysis [53] of breakdown in Ar that showed that ionization of the Ar gas by fast Ar atoms and by Ar+ ions becomes the dominant source of electrons for both clean and dirty cathodes. Here the fast Ar atoms are formed by charge transfer in Ar+ collisions with Ar. As yet there is no direct experimental evidence of the importance of secondary electron production by fast atoms from swarm or discharge experiments in Ar, although the closely related phenomena of Ar excitation by fast atoms and ions has been demonstrated experimentally and successfully modelled in recent years [5, 53, 63, 126, 141]. An interesting rule-of-thumb brought out by the calculations of figure 11 is that heavy-particle ionization becomes the dominant source of secondary electrons for breakdown and low-current, steady-state discharges at E/n above about 7 kTd or, from figure 4, at voltages greater than 1000 V. Because of the rapid increase in spatial ionization coefficient for ions and fast atoms with E/n, we expect this transition voltage to decrease when the discharge current is increased enough to form the high-field region of the cathode fall. See appendix B. (5) A critical need is for better independent experimental means of determining the electron yield from surfaces under conditions approximating those in the discharges to be modelled. (6) The cross sections for collisions of electrons and ions with the more common gases are relatively well known and techniques for calculating the electron and ion behaviour in the uniform electric fields of swarm experiments are well established. In the case of electrons in Ar there is good agreement between well tested experiments and models. Therefore modellers should be extremely cautious about changing these electron-gas parameters on the basis of model results for complicated discharges, e.g., discharges involving space charge electric fields and externally applied dc or radiofrequency fields. (7) It is obvious from the comparison of γeff from the model and from swarm experiments at E/n below 1000 Td in figure 5 that one must use surprisingly low photon yields in order to fit some of the data for dirty surfaces. Such low values appear to be inconsistent with measurements made in vacuum. It should be kept in mind that our model essentially assumes that at very low E/n and Ar high pressures all of the excitation of Ar by electrons appears as nonresonance photons. We have not found measurements of vuv emission efficiency for high-pressure Ar by other techniques that test this assumption. (8) The very limited data available [88] for low energy Ar+ ions incident on surfaces cleaned by sputtering in the presence of Ar indicates that high-dosage sputtering of Cu and Au without high temperature flashing results in significant decreases the γi values. We suggest that sputtering may cause similar reductions in the photoelectric yield. (9) The present modelling work should be extended to include greater detail as to the role of excited states of Ar, to more accurate models [5, 126, 141] of the fast ion and fast atom fluxes in Ar, to the differential voltage–current behaviour of low current discharges in Ar, to the time dependent growth of current in Ar discharges and to the analyses of discharges in other gases. An analysis of ionand photon-induced electron emission in low-current H2 discharges has evaluated the contributions of H+ , H+2 , H+3 and vuv photons [146]. R35 A V Phelps and Z Lj Petrović 1 Fraction escaping cathode (10) Although the model applied to the analysis of uniform-field data in this paper is applicable only for low enough discharge current densities so that one can neglect space distortion of the electric field, the same secondary electron production processes will occur at the higher current densities of cold-cathode discharges characterized as the normal and abnormal cathode fall [32, 61]. For example, values of γeff for Ar discharges at moderate current densities have been inferred by comparison of models with voltage–current measurements in the abnormal glow mode [10]. Most models of glow discharges [1–3, 5, 7, 8, 15, 23] use estimates of the γ values that are significantly larger than found in this paper. Nagorny and Drallos 10-1 Thomson - Loeb empirical fit Felsch and Pech Acknowledgments One of the authors (AVP) would particularly like to acknowledge numerous important suggestions and critical reviews of the drafts by A Gallagher. He also would like to thank B M Jelenković, L C Pitchford, J Broad and B D Paul for helpful contributions. The other author (ZLP) would like to thank the JILA Information Center and J. Broad for their hospitality during the preparation of a bibliography and notes concerned with electron collision cross sections, excited state quenching and discharge–surface interactions relevant to discharges in Ar. We would like to thank C A N Conde and T H V T Dias for recent references on scintillation-detector discharges. The present work was supported in part by the National Institute of Standards and Technology. 10-2 1 R36 102 103 104 E/n (Td) Figure A1. Fraction of electrons escaping the cathode versus E/n. The symbols, electron mean energies and references are: #, 0.2 eV, [122]; , 0.6 eV, [122]; !, 0.9 eV, [100]; ", 1.4 eV, [147]; , 3? eV, [95]; and ×, 1 eV, [148]. The solid, long-dashed and dot–dashed curves give the theoretical predictions of the modes of Thomson and Loeb [31], Felsch and Pech [100] and Nagorny and Drallos [150], respectively. Each of these models is evaluated for an average electron injection energy of 0.6 eV. The short-dashed curve shows our empirical fit to experiment for an average electron injection energy of 0.6 eV. • The Thomson model for electron escape as modified by Loeb [31] can be written as Appendix A. Electron backscattering and onset of ionization In this appendix we review the treatment of electron nonequilibrium in Ar near the cathode through the use of boundary conditions in local-field fluid models [2, 7, 8, 32, 90]. Based on experimental observations of drift-tube currents versus E/n and pd, the corrections are conventionally divided into two parts. One part is the backscattering to the cathode of electrons emitted from the surface. A second part is the delay in reaching the steady-state rates of electron excitation and ionization as determined from either the spatial dependence of emission or the pd dependence of the current. While these processes are accounted for without special effort in Monte Carlo treatments of electrons leaving absorbing cathodes, they must be taken into account separately in ‘local field’ or ‘equilibrium’ fluid models utilizing steady-state electron ionization and transport coefficients. In the case of electron backscattering to the cathode, we summarize available experimental data and models for Ar. Figure A1 shows the results of measurements [95, 100, 122, 147, 148] and models [31, 100, 149, 150] for the fraction of electrons escaping the cathode versus E/n for Ar. The measured escape fractions fes shown decrease monotonically with increasing electron injection energy. Some theories [149, 151, 152] find a minimum in the escape probability as the initial electron energy is increased. 10 fes = 1/(1 + 1v2/4We ) (A1) where 1v2 is the mean velocity of the injected electrons and We is the electron drift velocity [19] at the E/n near the cathode. The results of such a calculation, shown by the solid curve of figure A1, are well below the experimental data. Recent theoretical results by Nagorny and Drallos [150] for E/n from 10 to 1100 Td (dot–dash curve) are barely distinguishable from the empirical fit given next. It should be noted that the Thomson–Loeb model works well for many molecular gases [31, 137]. The short-dash curve of figure A1 shows a fit to experiment of an empirical expression for the escape fraction given by $ [((es /0.6)2 + (E/n)/30] 100 fes = 1 + [1 + (E/n)/30] (E/n) %−1/2 (A2) where the E/n value is in Td and (es is the energy of the electrons ejected from the surface in eV. The exponent 2 in equation (A2) should be reduced to about 1 for (es = 1 eV. In our models we used (es = 0.6 eV. The long-dash curve shows that the empirical fit by Felsch and Pech [100] varies too rapidly with E/n at the lower E/n. Experiments [153, 154] and Monte Carlo calculations [149, 151, 152, 154–156] show that results such as plotted in figure A1 and represented by equation (A2) are independent of electrode separation [122] only for a limited range of pd. At low pd, fes decreases as Cold-cathode discharges and breakdown in argon backscattering sets in [95]. Secondary electron production causes a growth of current at high pd [95, 98, 13]. The escape fraction data of figure A1 are used in the analysis of the γi data of figure 3 and of the γeff data of figure 5. At E/n from 100 to 1000 Td, the escape fraction is close to unity and the energy of the electrons emitted from the cathode is not important. At the low E/n, where photoelectric emission dominates secondary electron production, uncertainties in the energy of the injected electrons and the escape fraction caused by unknown surface conditions may reach 30%, but cannot account for the low and variable photoelectric yields derived from breakdown experiments. Measurements of the delay in the onset of the steadystate ionization have been reported for Ar by Druyvesteyn and Penning [32] and by Kruithof [88]. An empirical fit to the former data expressed as the effective value of the electrode potential difference V0 required before the current begins to grow exponentially with distance is given by V0 = 16{1 + [(E/n)/1000]2 }0.5 (A3) where V0 is in V and E/n is in Td. Monte Carlo techniques have been used to model the initial nonequilibrium for electrons in Ar [157]. The application of the singlebeam, energy-balance model of the nonequilibrium motion of electrons [53, 158] gives similar values of V0 at low E/n, but much larger values at high E/n. The addition of this correction to the equations of the metastable and resonance atom models of this paper at very high E/n makes the algebra extremely complicated. Because the effects of this correction on breakdown pd values at intermediate and low E/n are smaller than the spread in the experimental data, the correction is omitted throughout the present calculations. Appendix B. Coefficients for modelling Ar discharges In this appendix we assemble what we believe are the best available analytical expressions for cross sections, rate coefficients and electron yields for modelling electron, ion, fast atom and photon behaviour in Ar under conditions appropriate to electrical breakdown and to low- and moderate-current density discharges. The data are given in some detail because of their use in the models of this and a number of other papers [27, 53, 54, 90, 126] and their potential for use in future research. Analytical formulations of model parameters often result in a considerable reduction of computer time. In this section the αx /n values are in m2 , Qx values are in 10−20 m2 , the particle energies (x and kTx are in eV and the E/n are in Td unless otherwise noted. B.1. Electron collisions with Ar The empirical electron ionization coefficient expression used in this paper has been modified from that of [90] so as to be in better agreement with the experimental values [88, 107, 144] at very low E/n. At high E/n the expression is adjusted to be consistent with the results of Jelenak et al [159], Božin et al [160], unpublished calculations of Nanbu and Konoko [23] and unpublished experimental results from our group, rather than the higher experimental values of Kruithof [88]. An empirical fit is αei /n = 1.1 × 10−22 exp[−72/(E/n)] +5.5 × 10−21 exp[−187/(E/n)] +3.2 × 10−20 exp[−700/(E/n)] −1.5 × 10−20 exp[−10 000/(E/n)]. (B1) This and other fits in this paper are not linear least-squares fits, but are visual fits to data that are generally good to better than 10% at all E/n for which data exist. We have adjusted equation (B1) for an even closer fit to experiment [88] at E/n from 100 to 1000 Td, where exp(αei d) is largest and has the greatest effect on the computed γi . Note that our expression is consistent with that proposed many years ago by Ward [161] for 100 < E/n < 1000 Td, but differs significantly at higher E/n. On the basis of the calculations of Puech and Torchin [17] we conclude that the contribution of associative ionization to αei is less than about 10% at E/n ' 20 Td and neglect its role in our model. The coefficients for electron excitation of Ar to the metastable state are the sum of the calculated excitation coefficients to the two metastable levels from Puech and Torchin [17]. These values are about twice those measured by Tachibana [16]. We have adopted the theoretical values because of the overall consistency of the calculated data with swarm experiments [17, 19]. A fit to the spatial metastableexcitation coefficient data is αem /n = 2 × 10−21 exp[−16/(E/n)] +2.3 × 10−21 exp[−80/(E/n)] −1.5 × 10−21 exp[−2000/(E/n)]. (B2) The coefficients for electron excitation of Ar to the resonance state are also based on calculations of the total rate of excitation of the two lowest radiating states of Ar by Puech and Torchin [17]. The empirical fit to the spatial resonance-state excitation coefficient data is αer /n = 2 × 10−21 exp[−20/(E/n)] +4 × 10−21 exp[−100/(E/n)] −2 × 10−21 exp[−2000/(E/n)]. (B3) An empirical fit to the mean energy of electrons in Ar from our solutions of the Boltzmann equation [25] is 1(2 = 2.5(E/n)0.5 {1 + [(E/n)/5]2 }−0.2 ×{1 + [(E/n)/1000]2 }0.7 . (B4) Note that for 20 < E/n < 100 Td the sum of the excitation coefficients for the resonance and metastable states times their respective excitation thresholds in eV is numerically within 20% of the value of E/n, as expected if all of the electron energy is used for excitation of the Ar. This agreement extends to higher E/n =10 kTd when the ionization coefficient times the sum of the ionization potential and the electron mean energy is added to the energy input to excitation. The ‘single-beam’ models of electron motion in Ar of [53] and [54] utilize the elastic momentum transfer cross section Qmel , the total excitation cross section Qex and the ionization cross section Qion . Also, one uses the effective R37 A V Phelps and Z Lj Petrović momentum transfer cross section Qmeff , where Qmeff = Qmel + Qex + Qion in two-term solutions to the electron Boltzmann code for Ar [3, 18, 19, 21, 22, 53]. The cross section for elastic momentum transfer collisions of electrons with Ar Qmel has been presented graphically in several recent papers [3, 17–19, 21–25]. Here we express the results tabulated in [25] as 2 3.3 Qmel (() = ABS{6/[1 + ((/0.1) + ((/0.6) ] −1.1 × ( 1.4 /[1 + ((/15)1.2 ]/[1 + ((/5.5)2.5 +((/60)4.1 ]0.5 } + 0.05/(1 + (/10)2 +0.01 × ( 3 /[1 + ((/12)6 ]. (B5) This cross section is intended to be treated as isotropic scattering [25] when used to determine a differential cross section. The sum of the cross section for electronic excitation in collisions of electrons with Ar Qex obtained by fitting the results tabulated in [25] is approximated by 1.1 2.8 5.5 Qex (() = 0.034(( − 11.5) [1 + ((/15) ]/[1 + ((/23) ] +0.023(( − 11.5)/[1 + ((/80)]1.9 . (B6) In the single-beam model [53, 54] this cross section is multiplied by a representative excitation energy of 11.5 eV and used as the excitation ‘energy-loss’ function. This cross section is intended to be treated as forward scattering [25] when used to determine a differential cross section. The cross section for electron impact ionization in collisions of electrons with Ar, Qion , obtained by fitting the results tabulated in [25] is approximated by Qion (() = 970(( − 15.8)/(70 + ()2 +0.06(( − 15.8)2 exp(−(/9). (B7) This cross section is intended to be treated as forward scattering [25] when used to determine differential scattering cross sections. B.2. Ion energy distribution and drift velocity The ion energy distribution is a one-dimensional Maxwellian with a temperature T+ . An empirical fit to the kT+ versus E/n values of figure 8 of [26] is KT+ = +1.9[(E/n)/1000]1.1 + 0.026. (B8) Note that here kT+ is twice the ion mean energy [26]. The Ar+ drift velocity is approximated closely by [26] W+ = 4(E/n)/{1 + [0.007(E/n)]1.5 }0.33 (B9) electron yield per fast atom. When numerically averaged over the one-dimensional Maxwellian energy distribution of the Ar+ the electron yield per ion is approximated by 1γic 2 = 0.07 + 1 × 10−5 (kT+ )1.2 × exp(−500/kT+ )/[1 + (kT+ /100 000)0.7 ]. (B11) Note that this expression can be obtained from equation (B10) by replacing the threshold factor ((i − 500)1.2 by (kT+ )1.2 exp(−500/kT+ ), replacing (i by kT+ elsewhere and by modifying slightly the high-energy behaviour. In all cases examined, we find that this procedure provides a useful first guess as to the function fitting the average over the Maxwellian. The approximation to the yield of electrons per fast atom γac for clean, annealed electrodes is shown by the dashed curve in figure 1. For atom energies above the apparent threshold at 500 eV this curve is described by γac = 1 × 10−5 ((a − 500)1.2 /[1 + ((i /70 000)0.7 ] (B12) where the yield is zero for (a below 500. As discussed in section 5, we assume that the energy distribution of the fast Ar atoms is the same as that for the Ar+ ions. When this yield is averaged over the Maxwellian energy distribution the result is 1γac 2 = 1 × 10−5 (kT+ )1.2 × exp(−500/kT+ )/[1 + (kT+ /100 000)0.7 ]. (B13) For clean Cu surfaces that have been heavily sputtered, as in the experiments of Kruithof, [88], we will approximate the electron yield per ion γic at low energies by a typical value from the open points of figure 3 and at high energies by a term that approaches Oechsner’s value at 1 keV [73]. Thus, we suggest γics = 0.02 + 5 × 10−5 ((i − 300) (B14) where the contribution to the second term is zero for (i below 300. An even better approximation would have the constant term decrease to as low as 0.01 with an increase in sputtering. Unfortunately, the dependence of this decrease on ion energy and the initial surface condition is unknown. These results have been obtained when the ion and electron mean free paths are large compared to the expected surface roughness. The effects of surface roughness on the ‘effective secondary emission coefficient’ have been examined when the ion mean free path is small compared to the scale of the roughness [162]. For fast Ar atoms we suggest only the second term of equation (B14). where W+ is in m s−1 . B.4. Ion- and atom-induced electron yields for dirty metals B.3. Ion- and atom-induced electron yields for clean metals For dirty surfaces our approximation to the electron yield per fast Ar+ ion γid shown by the upper solid curve in figure 2 is given by For clean surfaces we will approximate the electron yield per ion γic by the solid curve in figure 1. This curve is given by γic = 0.07+1×10−5 ((i −500)1.2 /[1+((i /70 000)0.7 ] (B10) where the contribution to the second term is zero for (i below 500. The second term in this relation is also equal to the R38 γid = 0.002(i /[1 + ((i /30)1.5 ] +1.05 × 10−4 ((i − 80)1.2 /[1 + ((i /8000)]1.5 (B15) where the second term is zero for ion energies (i below 80 eV. As pointed out above, the continued decrease in γi with decreasing ion energy at energies below those shown in Cold-cathode discharges and breakdown in argon figure 2 is important for our fit to the γeff data of figure 5. In the energy range from 10 to 500 eV, these γi values are similar to those adopted by Neu [61]. When averaged over the one-dimensional Maxwellian velocity distribution of the fast Ar atoms, the yield is approximated by 1γid 2 = 0.06[kT+ − 30 exp(30/kT+ )$(0, 30/kT+ )]/kT+ +1 × 10−4 (kT+ )1.2 exp(−80/kT+ ) (B16) ×[1 + (kT+ /6000)1.5 ] where $[n, x] is the incomplete gamma function [163]. Note that for the first term of equation (B16) we have used the exact form of the integral obtained by averaging over a Maxwellian, rather than the procedure discussed in connection with equation (B11). Also, note that this first term is set to zero for calculating the short-dashed curves of figures 4, 5 and 11. For dirty surfaces the approximation to the electron yield per fast Ar atom γad shown by the lower solid curve in figure 2 is given by γad = 1 × 10−4 ((a − 90)1.2 /[1 + ((a /8000)1.5 ] +7.0 × 10−5 ((a − 32)1.2 /[1 + ((a /2800)1.5 ] (B17) where the first and second terms are zero for (a below 90 and 32 eV, respectively. The second term in equation (B17) is basically a shift of the first term to lower energies and lower magnitude so as to represent the structure found by Amme [50]. In the energy range from 10 to 500 eV, these γa values are similar to those adopted by Neu [61]. When averaged over the one-dimensional Maxwellian velocity distribution of the fast Ar atoms the yield is 1γad 2 = 1 × 10−4 (kT+ )1.2 × exp(−90/kT+ )/[1 + (kT+ /6000)1.5 ] +7.0 × 10−5 (kT+ )1.2 × exp(−32/kT+ )[1 + (kT+ /2100)1.5 ]. (B18) B.5. Fast ion–atom reaction coefficients The empirical approximation for symmetric charge transfer cross section as obtained by averaging the cross section from table 7 and figure 7 of [26] over the one-dimensional ion energy distribution and transforming from a kT+ energy scale to an E/n scale is αct /n = 1.3 × 10 −18 −0.117 (E/n) . (B19) The following spatial ionization and excitation coefficients for Ar + + Ar collisions are obtained by averaging the cross sections from [20] and [26] over the one-dimensional ion energy distribution and transforming from a kT+ energy scale to an E/n scale. Thus, the ionization coefficient for the Ar + + Ar → 2Ar + + e reaction is approximated by αii /n = 8.2 × 10−22 exp[−19 900/(E/n)] (B20) +7.1 × 10−21 exp[−46 000/(E/n)]. The empirical approximation for the spatial excitation coefficient for Ar + + Ar → Ar(1 P1 ) + Ar + is αir /n = 1.07 × 10−22 exp[−41 700/(E/n)]. (B21) The empirical spatial excitation coefficient for Ar + + Ar → Ar(3 P2 ) + Ar + is αim /n = 3.5 × 10−21 exp[−18 000/(E/n)] (B22) +5.0 × 10−21 exp[−50 000/(E/n)]. B.6. Fast atom–atom reaction coefficients Our empirical expression for the isotropic scattering approximation to the elastic viscosity collision cross section from [20] is −0.23 −0.28 −2 2 Qaa = 16(rel /[1 + (rel ] + 11.3(rel /[1 + (rel ]. (B23) This cross section results from a more detailed analysis of viscosity and beam experiments than that of [27]. It is in good agreement with that used in a recent model of sputtering experiments [6]. The corresponding approximation for the spatial loss coefficient for fast Ar atoms as the result of elastic viscosity collisions is obtained by replacing (rel by kT+ /2, i.e. assuming that the atoms have the same energy as the ions from which they were formed. We then transform from a kT+ scale to an E/n scale using the high-energy limit of equation (B8) to obtain αaa /n = 1.62 × 10−19 (E/n)−0.25 /[1 + 1.1(E/n)2.2 ] +1.15 × 10−19 (E/n)−0.31 /[1 + 0.9(E/n)−2.2 ]. (B24) In this equation only, E/n are in kTd. Because of the simplified energy transformation, this approximation is intended for use at Ar atom energies well above thermal. The following spatial ionization and excitation coefficient for collisions of fast Ar with thermal Ar are obtained by averaging the cross section from [26] and [20] over the onedimensional ion energy distribution and transforming from a kT+ energy scale to an E/n scale. The ionization coefficient for the Ar + Ar → Ar + + Ar + e reaction is αai /n = 2.1 × 10−21 exp[−25 000/(E/n)] +4.8 × 10−21 exp[−40 800/(E/n)]. (B25) The empirical approximation for the spatial resonance state excitation coefficient for the Ar + Ar → Ar(1 P1 ) + Ar reaction is αar /n = 2.7 × 10−21 exp[−32 800/(E/n)]. (B26) The empirical approximation for the spatial metastable excitation coefficient for the Ar+Ar → Ar(3 P2 )+Ar reaction is αam /n = 2.1 × 10−21 exp[−25 000/(E/n)] +4.8 × 10−21 exp[−40 800/(E/n)]. (B27) B.7. Thermal reaction and transport coefficients The collisional rate coefficients for Ar at 300 K defined in [123] and used here are: nDa = 1.7 × 1020 m−1 s−1 , nDm = 1.7 × 1020 m−1 s−1 , d = 0.01 m, Ba = 1.2 × 10−21 m3 s−1 , Ca = 1.3 × 10−44 m6 s−1 , Bm = not needed, Cm = not needed, AI = 8 × 104 s−1 , G = 1.5 × 10−45 n2 + 1.5 × 10−20 n s−1 . Here the gas density n is in m−3 . B.8. Photon collection at the cathode We have calculated numerically the fraction of isotropically emitted photons that reach the cathode assuming that the photons are produced with a distribution that is radially uniform but varies as exp(αei z) in the cylindrical volume of height d and radius r between the electrodes. We assume R39 A V Phelps and Z Lj Petrović that the photons are not absorbed or scattered by the gas. The results can be approximated by fgeom = 0.5/[1 + (d/r)(1 + αei d)0.55 /0.9]. (B28) For d/r ( 3 and αei d = 0 this expression is good to better than 5%. For d/r ( 3 and αei d ( 3 the fit is good to better than 10%. If the photons are produced by an electron flux that varies radially as a Bessel function, the calculated fgeom values increase by 10 to 15% for d/r ( 3 and αei d ( 3. The fgeom factor has not been explicitly shown in the formulas given in this paper, but approximations to it have been applied to the calculations of the contributions of nonresonant photons to the breakdown condition and to the current growth equations. References [1] Belenguer Ph and Boeuf J P 1990 Phys. Rev. A 41 4447 [2] Lieberman M A and Lichtenberg A J 1994 Principles of Plasma Discharges and Materials Processing (New York: Wiley) ch 1 [3] Surendra M, Graves D B and Jellum G M 1990 Phys. Rev. A 41 1112. These authors do not document tests of this set against experimental ionization and transport swarm data. One of us (AVP) has constructed a cross section set from the published graphs and finds that the calculated ionization coefficients are a factor of two too low and a factor of three too high compared to experiment at E/n of 30 and 1000 Td, respectively. One of the authors of this paper found records of a few calculations that are about 60% of our values. [4] Lister G G 1992 J. Phys. D: Appl. Phys. 25 1649 [5] Bogaerts A, van Straaten M and Gijbels R 1995 Spectrochim. Acta B 5 179. This is one of a series of models of moderate current density discharges in Ar. [6] Serikov V V and Nanbu K 1996 J. Vac. Sci. Technol. A 14 3108 Serikov V V and Nanbu K 1997 J. Appl. Phys. 82 5948. We thank Professor Nanbu for providing details of their viscosity cross sections that resulted in corrections to our original cross sections for Ar–Ar elastic collisions at low energies. See [27]. [7] Nagorny V P, Drallos P J and Williamson W Jr 1995 J. Appl. Phys. 77 3645 [8] Meunier J, Belenguer Ph and Boeuf J P 1995 J. Appl. Phys. 78 731 [9] Waymouth J F 1971 Electric Discharge Lamps (Cambridge, MA: MIT Press) ch 4 [10] Lagushenko R and Maya J 1983 J. Appl. Phys. 54 2255 [11] Pak H and Kushner M J 1990 Appl. Phys. Lett. 57 1619 Sommerer T J, Pak H and Kushner M J 1992 J. Appl. Phys. 72 3374 [12] Boeuf J P and Pitchford L C 1991 IEEE Trans. Plasma Sci. 19 286 Alberta M P, Derouard J, Pitchford L C, Quadoudi N and Boeuf J P 1994 Phys. Rev. E 50 2239 [13] Conde C A N, Santos M C M, Fátima M, Ferreira A and Sousa C A 1975 IEEE Trans. Nucl. Sci. 22 104 Santos F P, Dias T H V T, Rachinhas P J B M, Stauffer A D and Conde C A N 1998 IEEE Trans. Nucl. Sci. 45 176 [14] Bretagne J, Godart J and Puech V 1983 Beitr. Plasmaphys. 23 295 [15] Belasri A, Boeuf J P and Pitchford L C 1993 J. Appl. Phys. 74 1553 [16] Tachibana K 1986 Phys. Rev. A 34 1007. This paper documents tests of this electron–Ar cross section set against experimental metastable and ion production coefficients and transport data. This rather detailed cross section set (eight levels) is apparently not available on the internet. R40 [17] Puech V and Torchin L 1986 J. Phys. D: Appl. Phys. 19 2309. This paper documents tests of this very complete set of electron–Ar cross sections (≈30 levels) against experimental ionization, excitation and transport swarm data. Unfortunately, this highly recommended cross section set is not available on the internet. [18] Hayashi M 1990 unpublished. This set of 25 cross sections for electrons in Ar has been found to be consistent with swarm experiments. Contact Z Lj Petrović for details on these tests and regarding the associated compilation of excited state and transport data. [19] Pack J L, Voshall R E, Phelps A V and Kline L E 1992 J. Appl. Phys. 71 5363. A tabulation of this very much simplified cross section set for Ar atoms is available. See [20]. [20] Phelps A V 1999 unpublished. These data are available via the JILA web site at http://jilawww.colorado.edu/ on the ‘Atomic physics’ page under the subject ‘Collision data’ or at ftp://jila.colorado.edu/collision data. The file electron.txt also documents tests of calculated coefficients against experimental ionization and transport data from swarm experiments. Other files list recommended heavy-particle collision data. [21] Morgan W L 1992 Plasma Chem. Plasma Proc. 12 449 Morgan W L 1992 Plasma Chem. Plasma Proc. 12 477. This simplified cross section set is available for Ar and other gases from The Siglo Data Base, CPAT and Kinema Software at http://www.sni.net/siglo. [22] Fiala A, Pitchford L C and Boeuf J P 1994 Phys. Rev. E 49 5607. According to the web site in [21], the cross section set for electrons in Ar used in this paper has been revised (1998) to be consistent with electron swarm experiments. [23] Nanbu K and Kageyama J 1996 Vacuum 47 1031. Electron drift velocities and ionization coefficients calculated using this cross section set agree with experiment. See Nanbu K and Kondo S 1997 Japan. J. Appl. Phys. 36 4808 and Nanbu K 1998 private communication. The assumed γi values are significantly larger than the values expected for sputtered Cu from the data shown in figure 3. [24] Vasenkov A V 1998 Phys. Rev. E 57 2212. This cross section set is designed for application to models of high energy electron beams, rather than electric discharges. Tests of calculated quantities such as the mean energy per ion pair and electron range are presented. No tests of electron transport and ionization coefficients are given. [25] Phelps A V 1997 Bull. Am. Phys. Soc. 42 1721. Here it is strongly advocated that each publication and internet site that makes electron cross section sets available for modelling gas discharges provide documentation of the tests that have been made of the consistency of the cross section set with simple swarm experiments, such a spatial (Townsend) ionization coefficients, drift velocity and the ratio of the diffusion to mobility coefficients. Such a procedure would reduce the number of inconsistent sets in circulation and reduce the tendency of some modellers to attribute discrepancies between complex models and experiment to errors in the electron cross sections. [26] Phelps A V 1991 J. Phys. Chem. Ref. Data 20 557. It should be noted that in the absence of more detailed data the inelastic cross sections in this and the associated series of papers should be used with a differential scattering cross section sharply peaked in the forward direction, as is typical of high energy collisions. An isotropic inelastic scattering model for the inelastic collisions leads to excessive angular scattering and much larger contribution to the momentum transfer than was intended. Cold-cathode discharges and breakdown in argon [27] Phelps A V 1994 J. Appl. Phys. 76 747. In addition to citations [3]–[10] of this reference, numerous recent papers have misused and/or misstated the concepts and published data concerned with elastic scattering and its symmetric-charge-transfer component in collisions of Ar+ with Ar. For example, very recently Zhong X X, Wu J D, Wu C Z and Li F M 1998 J. Appl. Phys. 83 5069 have overestimated the isotropic component of the ion scattering at 100 eV by more than an order of magnitude. [28] Holstein T 1947 Phys. Rev. 72 1212 Holstein T 1951 Phys. Rev. 83 1159 [29] Payne M G, Talmage J E, Hurst G S and Wagner E B 1974 Phys. Rev. A 9 1050 [30] Townsend J S 1915 Electricity in Gases (Oxford: Clarendon) p 313 [31] Loeb L B 1939 Fundamental Processes of Electrical Discharge in Gases (New York: Wiley) p 313 [32] Druyvesteyn M J and Penning F M 1940 Rev. Mod. Phys. 12 87. This paper also discusses surface preparation procedures used in gas discharge experiments. [33] Little P F 1956 Handbuch der Physik vol 21, ed S Flügge (Berlin: Springer) p 574 [34] Raether H 1964 Electron Avalanches and Breakdown in Gases (London: Butterworth) [35] Dutton J 1978 Electrical Breakdown of Gases ed J M Meek and J D Graggs (Chinchester: Wiley) ch 3 [36] Raizer Yu P 1986 Tepolfiz. Vys. Temp. 24 984 (Engl. Transl. 1986 High Temp. (USSR) 24 744) [37] Kaminsky M 1965 Atomic and Ionic Impact Phenomena on Metal Surfaces (New York: Academic) ch 12 and 14 [38] Abroyan I A, Eremeev M A and Petrov N N 1967 Usp. Fiz. Nauk 92 105 (Engl. Transl. 1967 Sov. Phys. Usp. 10 332) [39] Carter G and Colligon J S 1968 Ion Bombardment of Solids (New York: Elsevier) ch 3 [40] Krebs K H 1968 Fortschr. Phys. 16 419 Krebs K H 1983 Vacuum 33 555 [41] Weston G F 1968 Cold Cathode Discharge Tubes (London: ILIFFE) ch 3 [42] Hasselkamp H D 1992 Particle Induced Electron Emission II ed G Höhler (Berlin: Springer) p 1 [43] Varga P and Winter H 1992 Particle Induced Electron Emission II ed G Höhler (Berlin: Springer) p 149 [44] Baragiola R A 1994 Low Energy Ion-Surface Interactions ed J W Rabalais (New York: Wiley) ch 4 [45] Rostagni A 1934 Z. Phys. 88 55. Surface treatment of Cu not stated. Background pressure 10−5 Torr. [46] Arifov U A, Rakhimov R R and Dzhurakulov Kh 1962 Dokl. Akad. Nauk SSSR 143 309 (Engl. Transl. 1962 Sov. Phys.–Dokl. 7 209). Prolonged heating of Mo at 2100–2200 K. Measurements at 1100 K following brief heating to 2000 K at 10−7 Torr. [47] Medved D B, Mahadevan P and Layton J K 1963 Phys. Rev. 129 2086 Mahadevan P, Layton J K and Medved D B 1963 Phys. Rev. 129 79. Yield extrapolated to immediately after flashing Mo to 2000 K at 10−8 Torr. [48] Hayden H C and Utterbach N G 1964 Phys. Rev. 135 A1575. Surface (Au) washed with detergent. Background pressure 10−6 to 10−5 Torr. [49] Haugsjaa P O, McIlwain J F and Amme R C 1968 J. Chem. Phys. 48 527. Surface (Au) washed with detergent. Background pressure ∼ 10−6 Torr. [50] Amme R C 1969 J. Chem. Phys. 50 1891. Unspecified surface and treatment. Background pressure 10−6 to 10−5 Torr. [51] Kadota K and Kaneko Y 1974 Japan. J. Appl. Phys. 13 1554. Polished, washed and outgassed at low temperatures. The surface was the first dynode of an electron multiplier. [52] See, for example, Armour D G, Valisadeh H, Soliman F A H and Carter G 1984 Vacuum 34 295 Rickards J 1984 Vacuum 34 559 [53] Phelps A V and Jelenković B M 1988 Phys. Rev. A 38 2975. Cathode (Cu) electropolished and chemically cleaned. Mild bakeout of system gave rate of pressure rise of 10−4 Torr min−1 . [54] Scott D A and Phelps A V 1991 Phys. Rev. A 43 3043 Scott D A and Phelps A V 1992 Phys. Rev. A 45 4198 [55] Mason R S and Alott R M 1994 J. Phys. D: Appl. Phys. 27 2372 [56] Townsend J S and Llewellyn Jones F 1933 Phil. Mag. 15 282. See Townsend J S and Yarnold G D 1934 Phil. Mag. 17 594 for measurements in He. [57] McClure G W 1959 J. Electron. Control 7 439 [58] Ul’yanov K N 1970 Zh. Tech. Fiz. 40 2138 (Engl. Transl. 1971 Sov. Phys.–Tech. Phys. 15 1667) Ul’yanov K N and Tskhai A B 1981 Tepolfiz. Vys. Temp. 19 41 (Engl. Transl. 1981 High Temp. (USSR) 19 32) [59] Blasberg H A M and de Hoog F J 1971 Physica 54 468 [60] Lauer E J, Yu S S and Cox D M 1981 Phys. Rev. A 23 2250 [61] Neu H 1959 Z. Phys. 154 423 Neu H 1959 Z. Phys. 155 77 [62] Warren R 1955 Phys. Rev. 98 1650 Warren R 1955 Phys. Rev. 98 1658 [63] Rózsa K, Gallagher A and Donkó Z 1995 Phys. Rev. E 52 913 [64] Phelps A V, Petrović Z Lj and Jelenković B M 1993 Phys. Rev. E 47 2825 [65] Cleaning techniques for various elements have been summarized by Musket R G, McLean W, Colmenares C A, Makowiecki D M and Siekhaus W J 1982 Appl. Surf. Sci. 10 143; Grunze M, Ruppender H and Elshazly O 1988 J. Vac. Sci. Technol. A 6 1266. We thank a referee for pointing out these references. [66] Ferrón J, Alonso E V, Baragiola R A and Oliva-Florio A 1981 J. Phys. D: Appl. Phys. 14 1707. Cleaned Mo target by ion bombardment until yield was independent of time. Apparently no annealing was used. Measurements at 10−10 Torr. [67] Güntherschulze A 1930 Z. Phys. 62 600. Probably these data should be characterized as obtained using a discharge technique, rather than a beam technique. However, we have not been able to determine the appropriate E/n values and so have adopted the author’s assignment of the incident ion energy. Copper targets cleaned by electron and ion bombardment. [68] Hagstrum H D 1954 Phys. Rev. 96 325 Hagstrum H D 1954 Phys. Rev. 96 336. The tungsten was flashed to 2200 K at 10−11 Torr. [69] Parker J H Jr 1954 Phys. Rev. 93 1148. The system was baked at 670 K and the targets were flashed to 1670 K for Pt and 1670–2270 K for Ta. The measurements were made at 2 × 10−5 Torr. [70] Vance D W 1967 Phys. Rev. 164 372. The Mo target was flashed to 2000 K for 15 s and measurements were made at a pressure of 4 × 10−9 Torr. The higher results shown in figure 1 are for Mo treated with O2 to remove carbon. [71] Winter H, Aumayr F and Lakits G 1991 Nucl. Instrum. Methods B 58 301 Töglhofer K, Aumayr F and Winter H P 1993 Surf. Sci. 281 143. Gold target cleaned by ion sputtering in vacuum. Measurements made at 10−3 Torr. Lakits G, Aumayr F and Winter H 1989 Rev. Sci. Instrum. 60 3151 cite evidence that these Au targets are ‘atomically clean’. [72] Alonso E V, Alurralde M A and Baragiola R A 1986 Surf. Sci. 166 L155. For Xe+ incident on sputtered Au these authors find an exponential decrease in electron yield with decreasing energy at 0.5 to 3 keV. These yields are much smaller than those of [76]. [73] Oechsner H 1978 Phys. Rev. B 17 1052. Copper target cleaned by sputtering at 5 mA cm−2 and measurements made at 10−5 Torr. R41 A V Phelps and Z Lj Petrović [74] Magnuson G D and Carlston C E 1963 Phys. Rev. 129 2409. Targets cleaned by ion bombardment and measurements made quickly at a background pressure of 5 × 10−8 . [75] Szapiro B and Rocca J J 1989 J. Appl. Phys. 65 3713. Commercial purity samples were polished and then operated as the cathode of a glow discharge. The system was then pumped to 10−6 Torr for yield measurements. [76] Hagstrum H 1953 Phys. Rev. 89 244. The Mo target was flashed periodically to 1750 K for minutes and measurements were made at a He pressure of 10−7 Torr. [77] Kirchhoff J F, Gay T J and Hale E B 1993 Ionization of Solids by Heavy Particles ed R A Baragiola (New York: Plenum) p 283 [78] Böhm C and Perrin J 1993 Rev. Sci. Instrum. 64 31. Targets cleaned by operating discharge in H2 –Ar mixture. [79] Hagstrum H D 1960 J. Appl. Phys. 31 897. The treatment of the contaminated tungsten was not given. The background pressure was at 10−9 Torr. [80] Berry H W 1948 Phys. Rev. 74 848. Surface (Ta) preparation and pressures not given. [81] Ghosh S N and Sheridan W F 1957 J. Chem. Phys. 26 480 [82] Hofer W 1983 unpublished. Cited in figure 3.3 of [43]. Cleaning procedure not given, but presumably similar to [71]. [83] Rundel R D and Stebbings R F 1972 Case Studies in Atomic Collision Physics II ed E W McDaniel and M R C McDowell (Amsterdam: North-Holland) ch 8. This paper reviews surface preparation procedures. [84] Schohl S, Klar D, Kraft T, Meijer H A J, Ruf M-W, Schmitz U, Smith S J and Hotop H 1991 Z. Phys. D 21 25. Au and Mo surfaces prepared by evaporation on to polished stainless steel and then chemically cleaned. Measurements were made at 300 to 360 K at a background pressure of 10−7 Torr. [85] Schall H, Beckert Th, Alvariño J M, Vecchiocattivi F and Kempter V 1981 Nuovo. Cimento B 63 378 [86] Alvariño J M, Hepp C, Kreiensen M, Staudenmayer B, Vecchiocattivi F and Kempter V 1984 J. Chem. Phys. 80 765 [87] Neynaber R H and Magnuson G D 1977 J. Chem. Phys. 67 430 [88] Kruithof A A 1940 Physica 7 519. Tubes baked with H2 at 670 K and then under a vacuum at 740 K. These thermally grounded, large area Cu cathodes were heavily sputtered (∼1019 ions per cycle) with an Ne discharge at 2 mA cm−2 , but at unspecified pressure and voltage. Earlier results with Ar in which the Cu cathode of the same experimental tube was cleaned by sputtering and was not heated to high temperatures also showed low γi values (0.009 to 0.04). The breakdown data were obtained with d = 1 cm. See Kruithof A A and Penning F M 1936 Physica 3 515. [89] Llewellyn-Jones F 1953 Rep. Prog. Phys. 16 216 [90] Petrović Z Lj and Phelps A V 1997 Phys. Rev. E 56 5920. Cathode (Cu) chemically cleaned, electropolished and gold-plated. Mild bakeout of system gave rate of pressure rise of better than 10−4 Torr h−1 . The voltages cited in the present paper are the discharge maintenance voltages extrapolated to zero current. In our present notation the γi values of this reference should be designated as γeff values. [91] Phelps A V 1998 unpublished. Solution of the appropriate continuity equations shows that when a change in E/n causes αei − αii to increase through zero the curvature of a plot of i(nd)/i(0) versus electrode separation changes from positive to negative. On the other hand the curvature of equation (1) is always positive. The change in curvature is calculated to occur for low-current discharges in Ar and He at E/n of the order of 50 kTd. Here we have neglected the effects of changes in γi , ionization by fast atoms and electrons backscattered from the anode. R42 [92] Hornbeck J A 1951 Phys. Rev. 83 374. Here we are only interested in the technique, because the cathodes were coated with metal oxides. [93] Varney R N 1954 Phys. Rev. 93 1156. The tube baked at 670 K and the Mo electrodes repeatedly ‘glowed by induction heating’ and sputtered at high currents. The residual pressure was 4 × 10−9 Torr. [94] Engstrom R W and Huxford W S 1940 Phys. Rev. 58 67 [95] Molnar J P 1951 Phys. Rev. 83 940. Repeated baking and outgassing of the cathode with a glow discharge. The tubes were filled and sealed off after flashing a barium getter. [96] Brunker S A and Haydon S C 1983 J. Physique Coll. C7 55 Ernest A D, Haydon S C and Fewell M P 1994 J. Phys. D: Appl. Phys. 27 2531 [97] Policarpo A J P L 1981 Phys. Scr. 23 539 Policarpo A J P L 1982 Nucl. Instrum. Methods A 196 53 [98] Breskin A, Buzulutskov A, Chechik R, Di Mauro A, Nappi E, Paic G and Piuz F 1995 Nucl. Instrum. Methods A 367 342 [99] Di Mauro A, Nappi E, Posa F, Breskin A, Buzulutskov A, Chechik R, Biagi S F, Paic G and Piuz F 1996 Nucl. Instrum. Methods A 371 137. These authors say their Monte Carlo calculations do not include photoelectron feedback. We therefore note that the MC results shown in their figure 4 are inconsistent with the experimental results of [122] for E/p between 2 and 5 V cm−1 Torr, where photoelectron feedback is expected to be small. [100] Felsch D and Pech P 1973 Beitr. Plasmaphys. 13 197 Felsch D and Pech P 1973 Beitr. Plasmaphys. 13 253. Electrodes were inductively heated to 1670 K for Mo and 1270 K for Ni and Fe. The cathode was then bombarded with ions from a glow discharge. [101] Jurriaanse T, Penning F M and Moubis J H A 1946 Philips Res. Rep. 1 225. These small, thermally isolated Mo cathodes were sputtered at high enough currents to raise the cathode temperature to above 1500 K. [102] Hornbeck J A and Molnar J P 1951 Phys. Rev. 84 621 [103] Kebarle P, Haynes R M and Searles S K 1967 J. Chem. Phys. 47 1684 [104] Gallagher A C 1996 private communication [105] Changes in surfaces produced by sputtering are reviewed by Navinšek B 1976 Prog. Surf. Sci. 7 49 [106] Schade R 1938 Z. Phys. 108 353. Tube baked at 770 K and Ni electrodes heated to bright red. Electrode spacing from 0.3 to 5 cm. [107] Kachickas G A and Fisher L H 1953 Phys. Rev. 91 775. Electrode spacings were 0.3 to 3 cm. [108] Menes M 1959 Phys. Rev. 116 481. System baked at 620 K and residual pressure of 3 × 10−9 . Probable purity of gas supply is 0.9999. Electrode spacing 0.3m to 1 cm. [109] Golden D E and Fisher L H 1961 Phys. Rev. 123 1079. Electrodes (Ni) exposed to (a) a 10−6 A discharge for several days and (b) ‘allowing an intermittent spark discharge to pass between the electrodes for about 50 hours’. Measured purity of gas supply was 0.999. The electrode separation was 0.5 to 5 cm and their diameter was 20 cm. The observed large increase in breakdown voltage as the gas aged may have been caused by the build-up of impurities that quenched the long-lived lower molecular state of Ar2 [134]. [110] Heylen A E D 1968 J. Phys. D: Appl. Phys. 1 179. Baking system at 470 K produced pressure of 10−7 Torr. Typical electrode spacings at breakdown were 0.5 cm. Electrodes exposed to ethane from mixture experiments. [111] Pace J D and Parker A B 1973 J. Phys. D: Appl. Phys. 6 1525. Electropolished and chemically cleaned stainless steel electrodes separated by 0.3 to 2 cm. System baked to give 10−8 Torr. Cold-cathode discharges and breakdown in argon [112] Bhasavanich D and Parker A B 1977 Proc. R. Soc. A 358 385. Electropolished and chemically cleaned stainless steel electrodes separated by 0.2 to 3 cm. System baked to give 10−9 Torr. Electrodes conditioned with glow discharge to give reproducible results. [113] Stefanović I and Petrović Z Lj 1997 Japan. J. Appl. Phys. 36 4728 [114] Auday G, Guillot Ph, Galy J and Brunet H 1998 J. Appl. Phys. 83 5917. Tube heated to 240 ◦ C for 24 h. Electrode diameter 5 cm, d = 0.2 to 1 cm. [115] Weissler G L 1956 Handbuch der Physik vol 21, ed S Flügge (Berlin: Springer) p 304 [116] Cairns R B and Samson J A R 1966 J. Opt. Soc. Am. 56 1568 [117] Kenty C 1933 Phys. Rev. 44 891 [118] Hinterreger H E and Watanabe K 1953 J. Opt. Soc. Am. 43 604 [119] Krolikowski W F and Spicer W E 1969 Phys. Rev. 185 882. Very similar photoelectric yields have been obtained for Ni by Blodgett A J Jr and Spicer W E 1966 Phys. Rev. 146 390. [120] Séguinot J, Charpak G, Giomataris Y, Peskov V, Tischhauser J and Ypsilantis T 1990 Nucl. Instrum. Methods A 297 133 [121] Travier C 1994 Nucl. Instrum. Methods A 340 26 [122] Theobald J K 1953 J. Appl. Phys. 24 123 [123] Phelps A V 1960 Phys. Rev. 117 619. In equation (20) of this reference the symbol E should be Cam . In the line following equation (22), xi = (2i − 1)d/2q and in the line following equation (23), pi = exp[α(2i − 1)d/2q]. The lower limit to the integral for θm should be (2m − 1)d/2q. [124] Nonequilibrium behaviour of the electrons is expected at breakdown and for low-current discharges for E/n > 3000 Td or pd < 0.3 Torr cm. The condition for equilibrium and for the applicability of the local field model is examined theoretically in many recent papers. See, for example, the reviews by Kolobov V I and Godyak V A 1995 IEEE Trans. Plasma Sci. 23 503 Kortshagen U, Busch C and Tsendin L D 1996 Plasma Sources Sci. Technol. 5 1. The experimental evidence for nonequilibrium is summarized in appendix A. [125] Jelenković B M and Phelps A V 1996 Phys. Rev. E 53 1852 and unpublished measurements in Ar and Monte Carlo calculations in H2 . Deviations from the exponential growth predicted by a constant spatial ionization coefficient have been calculated by Hayashi in [152]. [126] Petrović Z Lj and Stojanović V D 1998 J. Vac. Sci. Technol. A 16 329. The excitation and ionization coefficients for Ar+ –Ar collisions calculated by these authors using the MC technique at their lower E/n are significantly larger than those calculated using the one-dimensional distribution of the present paper. This difference may be important for breakdown models at E/n near 3000 Td. [127] Lawler J E 1985 Phys. Rev. A 32 2977 [128] Rao M V V S, Van Brunt R J and Olthoff J K 1996 Phys. Rev. E 54 5641 [129] Madey T E and Yates J T Jr 1971 J. Vacuum Sci. Tech. 8 525 Langley R A, Bohdansky J, Eckstein W, Mioduszewski P, Roth J, Taglauer E, Thomas E W, Verbeek H and Wilson K L 1984 Nucl. Fusion Special Issue ch 4 [130] Klyarfel’d B N, Guseva L G and Vlasov V V 1968 Zh. Tekh. Fiz. 38 1288 (Engl. Transl. 1969 Sov. Phys.–Tech. Phys. 13 1056) [131] We have not found data for the yield of electrons from Ar+2 collisions with either clean or dirty metals. Because about 14.3 eV of energy is available [164] in a vertical transition from the lowest vibrational state of the ion to the repulsive ground state of Ar2 we assume that the electron yield for clean metals is similar [74] to that for Kr+ (ionization potential = 14 eV) on clean Mo, i.e. ∼0.05 or about half that for Ar+ . [132] Weise W L, Smith M W and Miles B M 1969 Atomic Transition Probabilities vol II (Washington, DC: US Government) p 192 [133] Castex M C, Morlais M, Spieglemann F and Malrieu J P 1981 J. Chem. Phys. 75 5006 [134] Millet P, Briot A, Brunet H, Dijols H, Galy J and Salamero Y 1982 J. Phys. B: At. Mol. Phys. 15 2935 [135] The model with eight excited states is subject to considerable uncertainty because of poorly known rate coefficients and/or reaction products. This is particularly true for the 4s 1 P1 resonance state and the 4s 3 P0 metastable state where the potential curves [133] suggest no collisional coupling to lower levels at 300 K and experiments show two and three-body collision loss [165, 166]. Fortunately, only 25% of the electron excitation is to these levels [17]. Most, but not all, of our rate coefficients for the lower six levels agree with those given by Millet et al [134]. We have not examined the transient solutions for our eight-level or our four-level model. [136] Wilkinson P G 1968 Can. J. Phys. 46 315. These results can be used to show that the vuv absorption by Ar near 106.6 nm scales as p 2 d and approximately as 1/*λ at wavelength shifts *λ from +0.6 to +30 nm, as expected for dipole–dipole collisional broadening. Over most of this range the normalized absorption is about twice that predicted by dipole–dipole broadening theory used in our resonance-radiation transport model [28]. At wavelengths closer to line centre the limited data approach the theory. [137] The calculations presented in this paper were carried out using Mathematica 3.0 © on a 300 MHz personal computer and required about 25 s per E/n value. [138] Details are available on request. Send email to [email protected]. [139] Kolobov V I and Fiala A 1994 Phys. Rev. E 50 3018 [140] Avery L W, House L L and Skumanich A 1969 J. Quant. Spectrosc. Radiat. Transfer 9 519. From these calculations for a Doppler broadened profile, spatially uniform excitation, a finite cylinder with an axial optical depth at line centre of 100 and ratios of radius to length of 1:1 and 4:1, roughly 45% and 15% of the radiation is lost to the side wall. A collision-broadened line profile appropriate to breakdown in Ar, with its weaker dependence of photon transmission on distance, will presumably lead to somewhat larger losses to the side wall. For comparison, the corresponding losses by diffusion in the fundamental spatial mode are 37% and 13%. [141] Revel I, Pitchford L C and Boeuf J P 1997 Proc. 23rd Int. Conf. on Phenomena in Ionized Gases vol II, ed M C Bordage and A Gliezes (Toulouse: University Paul Sabatier) p 56 [142] For example, the determination of the steady-state breakdown and discharge maintenance pd at a given E/n would require interpolation between pd values for MC solutions yielding temporally growing and decaying discharge currents. Unfortunately, a recent application of MC techniques for a finite growth rate during pulsed breakdown appears to neglect the multiple avalanches required to observe breakdown [167]. The use of Monte Carlo techniques for the modelling of high current cathode fall discharges [5, 6] usually does not require multiple solutions because the nonlinear dependence of space-charge electric fields on current density means that one can often reach a steady-state current with an arbitrary choice of discharge voltage and pd. [143] In some cases it was not possible to set parameters equal to zero because of indeterminant results. Therefore the numerical results reported were calculated with arbitrary values from 10−4 to 10−10 instead of zero. R43 A V Phelps and Z Lj Petrović [144] Specht L T, Lawton S A and DeTemple T A 1980 J. Appl. Phys. 51 166. These authors suggest their large αei /n values at low E/n are caused by photon feedback. If we use equation (B28) and our αph fes values, we find satisfactory agreement, ±50%, with their apparent αi values for 5 < E/n < 15 Td. For higher E/n we believe their αi values are too small because of the neglect of the initial period of low ionization discussed in Appendix A. [145] Pustynskii L N and Shumilin V P 1987 Zh. Tekh. Fiz. 57 1699 (Engl. Transl. 1987 Sov. Phys.–Tech. Phys. 32 1016) [146] Šimko T, Bretagne J and Gousset G 1997 Proc. 23rd Int. Conf. on Phenomena in Ionized Gases vol IV, ed M C Bordage and A Gliezes (Toulouse: University Paul Sabatier) p 184. See also Šimko T, Donkó Z and Rózsa K 1997 Proc. 23rd Int. Conf. on Phenomena in Ionized Gases vol II, ed M C Bordage and A Gliezes (Toulouse: University Paul Sabatier) p 64 [147] Dahlquist J A 1962 Phys. Rev. 128 1988 [148] Smejtek P, Silver M, Dy K S and Onn D G 1973 J. Chem. Phys. 59 1374. These experiments extend the measurements to very low E/n (0.01 Td) and up to liquid Ar densities. [149] Pareathumby S and Segur P 1979 Lett. Nuovo Cimento 26 243 [150] Nagorny V P and Drallos P J 1997 Plasma Sources Sci. Technol. 6 212 [151] Thomas R W L and Thomas W R L 1969 J. Phys. B: At. Mol. Phys. 2 562 [152] Hayashi M 1976 Proc. 4th IEE Int. Conf. on Gas Discharges (London: Institution of Electrical Engineering) p 195 Hayashi M 1982 J. Phys. D: Appl. Phys. 15 1411 R44 [153] Orient O J 1965 Can. J. Phys. 43 422 [154] Békiarian A 1968 J. Physique 29 434 [155] Burch D S and Whealton J H 1977 J. Appl. Phys. 48 2213 Whealton J H, Burch D S and Phelps A V 1977 Phys. Rev. A 15 1685 [156] Kuntz P J and Schmidt W F 1982 J. Chem. Phys. 76 1136 [157] Sakai Y, Tagashira H and Sakamoto S 1972 J. Phys. B: At. Mol. Phys. 5 1010 [158] Phelps A V, Jelenković B M and Pitchford L C 1987 Phys. Rev. A 36 5327 [159] Jelenak Z M, Velikić Z B, Božin J V, Petrović Z Lj and Jelenković B M 1993 Phys. Rev. E 47 3566 [160] Božin J V, Jelenak Z M, Velikić Z V, Belča I D, Petrović Z Lj and Jelenković B M 1996 Phys. Rev. E 53 4007 [161] Ward A L 1962 J. Appl. Phys. 33 2789 [162] Drallos P J, Nagorny V P, Ryutov D D and Williamson W Jr 1997 Comments Plasma Phys. Control. Fusion 18 37 [163] Davis P J 1964 Handbook of Mathematical Functions ed M Abramowitz and I A Stegun (Washington, DC: US Government) ch 6, equation (6.5.3). [164] Kuo C-Y and Keto J W 1983 J. Chem. Phys. 78 1851 [165] Klots J H and Setser D W 1978 J. Chem. Phys. 68 4848 [166] Firestone R F, Oka T and Takao S 1979 J. Chem. Phys. 70 123 [167] Agache M, Fitaire M and Ludac E 1996 J. Phys. D: Appl. Phys. 29 1217 [168] Guseva L G 1964 Investigations into Electrical Discharges in Gases ed B N Klyarfel’d (New York: Pergamon) p 1 [169] Klyarfel’d B N, Guseva L G and Pokrovskaya-Soboleva A S 1966 Zh. Tekh. Fiz. 36 704 (Engl. Transl. 1966 Sov. Phys.–Tech. Phys 11 520)
© Copyright 2024 Paperzz