Arxiv:2010.03556V2 [Physics.Plasm-Ph] 1 Feb 2021 Uigo Oe Osmn.Figure Pro- Power Consuming

arXiv:2010.03556v2 [physics.plasm-ph] 1 Feb 2021 uigo oe osmn.Figure pro- power consuming. either power be or can from system ducing bias the applied work circuit, the the external and electri- the on electrodes two an Depending the drives of circuit. functions externally process the through are this current electrodes load, cal two a the across If connected thermionically the electrons. other of some emitted The absorbs the [7]. anode) (the function to electrode work density re- and current which temperature emitted equation, electrode Richardson thermionically current the the emitted by lates The given density. is current density electrons desired of a emission to at thermionic heated occur which is at cathode) inter- temperature (the the a electrode as One to gap). physically (referred form, distance electrode electrodes gap simplest two some its of by In consists separated diode [6]. thermionic gas ther- a [4], natural solar and include [5] sources for monuclear heat potential of its ap- widening Examples further of to impact. used, range agnostic heat of are wide source thermionics a the Furthermore, makes in technology appealing [3]. the converter plications of of size) type system this of efficiency independent conversion is (core scalability lack The and parts 2]. moving [1, of energy electrical into heat convert directly iue1 lcrnmtv iga oso eaino dif- diod thermionic of a relation in show interest to E of diagram parameters motive physical Electron ferent 1. Figure h nd okfnto and function device. work anode the f hrinceeg ovres(Es r eie that devices are (TECs) converters energy Thermionic niae em levels, Fermi indicates vacuum level cathode E cathode lsaldt ag oeta utain hc hne the Spec changed which fluctuations cases. plasm potential the two large as the to changes led between plasma gradually decay potential plasma electrostatic bo the the in in performed were observed plas Simulations were the surfaces. while electrode studied the is curren on electrodes, diode the time-averaged between The potential simulations. (PIC) in-cell oe ehiu fmauigtedffrnei okfunction work in vacuu difference anode converter. colle the and thermionic measuring cathode is of the current technique where time-averaged time- novel condition, maximum the flat-band the in the differences that to significant found creates was This it case. 2D the h yaiso nagnpam ntegpo hrincdiod thermionic a of gap the in plasma argon an of dynamics The iuain fAgnPam ea naTemoi Converter Thermionic a in Decay Plasma Argon of Simulations Introduction φ c .E Groenewald, E. R. stectoewr function, work cathode the is V out h uptvlaeo the of voltage output the 1 oenEeto n. ohl,W 81,USA 98011, WA Bothell, Inc., Electron Modern 1 vacuum level E hw typical a shows anode anode 1 Dtd eray2 2021) 2, February (Dated: .Clark, S. φ e. a 1 .Kannan, A. rmas hw hta ogas long as that shows also gram eie ovrey when Conversely, accelerating the called regime. anode, the towards accelerated are hnvrteaoeFrilvli talwrelectrical lower i.e. a cathode’s at the is than level Fermi potential anode the whenever V rnmtv iga lt h eaieo h electrical the of negative the ( potential elec- plots The diagram diode. motive thermionic tron a of diagram motive electron eadn eie h idepit when the point, termed middle The anode, the regime. towards retarding move they as decelerated scle h a-adcniin hscrepnst the to corresponds where This case condition. flat-band the called is φ icag,dfie steacdo,haiyrdcsthe reduces arc- heavily an arc-drop, maintain the to as required mitigating defined energy at discharge, effective the are charge, low plasmas is these space cesium While barrier of energy charge 16]. ionization [15, for space the employed heavily since the been purpose, have neutralize this plasmas to Cesium 14]. gap positive [13, use the to in is in ions programs, thermionic successfully development a more TEC of used previous operation approach, unable term Another been long converter. far stable, so produce have to bi- height. methods barrier and electrodes charge these [11] additional space Unfortunately, the (2013) used reduce [12] to al. positively (2016) ased et maintain Meir al. et to problem. Wanke spacers charge (10 used space gap [10] anode. the interelectrode the small (2014) sufficiently to the a it al. makes increase et that to current Belbachir 9]. reported emitted case [8, of been a law fraction such have Child-Langmuir the In the strategies of obeys Various most cathode. current reflects barrier the diode that charge to the space cathode back it A the electrons make of emitted anode. electrons front the emitted in to forms the gap wide, of microns the fraction few through a small than a more only is gap interelectrode the c nraiy o aumtemoi id nwhich in diode thermionic vacuum a for reality, In seetottcptnil.Tedvc rdcspower produces device The potential). electrostatic is − φ adniydpee u orecombination to due depletes density ma eas hl n1 utain nthe in fluctuations 1D in while decays, a a ,a ucino h eaieelectrical relative the of function a as t, h1 n Dadsgicn differences significant and 2D and 1D th . lsadcycaatrsisrltv to relative characteristics decay plasma ewe h ahd n nd na in anode and cathode the between s vrgddoecret Furthermore, current. diode averaged 1 φ issaeeul hssget a suggests This equal. are biases m tdwe h id otg sset is voltage diode the when cted V n .Scherpelz P. and = out fial,i Di a on that found was it 2D in ifically, − sivsiae sn particle- using investigated is e eV qastecnatptnildifference, potential contact the equals where , V e out steeetoi hreand charge electronic the is 1 + V out φ a V + out φ > φ a > c φ < V lcrn are electrons , µ out .Tedia- The 0. )t avoid to m) c + electrons , φ a = φ c , 2

The decay of an argon plasma in the gap of a thermionic converter as a function of time was studied with different biases applied to the anode relative to the cathode. Specifics of the simulation setup are discussed in the Methods section. It was found that the plasma lifetime is strongly dependent on the anode bias. Thereby a characteristic of the IV-curve was identified that allows extraction of the difference in electrode work functions. Furthermore, simulations were performed in both 1D and Figure 2. Schematic of a plasma-based thermionic diode. The 2D, and it was found that the decay characteristics of the diode consists of two parallel plates, the cathode (left) and plasma are strongly dependent on the dimensionality of anode (right). Arrows indicate the direction of motion of the system. The Results section describes the differences thermionically emitted electrons as they travel from the cath- in the plasma decay under different anode biases as well ode to the anode. An argon plasma is present in the inter- as the differences that arise from dimensionality. In the electrode gap to neutralize the space charge formed by the emitted electrons, and thereby facilitates the flow of current. Discussion section the plasma sheaths are studied and the origin of the dimensional dependence is explored. efficiency of the converter [2]. This is due to the large Theoretical considerations neutral scattering cross section of cesium atoms for low energy electrons. As a solution to this problem, using inert gas plasmas have been suggested where the Ram- A well documented characteristic of pulsed inert-gas sauer minimum makes these gases mostly transparent to plasma thermionic converters is that once the plasma low energy electrons. Using an inert gas plasma requires producing pulse is turned off, the plasma sheath in front the plasma ignition be engineered to be very energy ef- of the cathode rapidly inverts from an ion-accelerating to ficient. This can be achieved by including highly biased an ion-retaining (electron-rich) sheath [21, 22]. This is a auxiliary electrodes that are optimally placed to produce desirable situation, since if instead the plasma producing an inert gas plasma in the gap [17, 18]. Another way is pulse created a dense enough plasma to completely miti- to apply short high voltage pulses across a diode [19, 20]. gate the thermionic space charge barrier for an extended As electrons accelerate towards the anode they collide amount of time, it means for some number of produced with neutral atoms in the gap, causing ionization and ions their decay doesn’t affect the diode current. There- eventually striking a plasma in the gap. A schematic fore, the energy used to create those ions was wasted. It of this is shown in Fig. 2. When the plasma producing consequently would be more energy efficient to pulse for pulse is turned off, the plasma starts to slowly decay to a shorter amount of time and instead re-pulse more fre- the electrodes (the dynamics of this decay are discussed quently so that each produced ion has a maximal impact in detail later). During this time, a large portion of the on the power producing diode current. For this reason, space charge cloud from thermionically emitted electrons we can assume that during the power producing phase, is neutralized and the diode current remains high. Once the cathode sheath will always be ion-retaining. Accord- the plasma density has decayed to such a level where it ing to McVey (1990) [23] the electron and ion currents can no longer support high diode current, the plasma ig- can be described by the following equations: nition pulse is repeated to start the process over again. 2 eV By keeping a very low duty cycle of on-to-off phases of the J = J exp c J (1a) e 3 R kT − re ignition pulse, high time-averaged diode current can be c sustained with relatively little energy spent to repeatedly eVc J = (J /2 J )exp (1b) strike the plasma [21]. i i − ri kT c While pulsed argon plasma thermionics have been studied concerning the energy required to repeatedly where JR is the Richardson current emitted from the strike the plasma, little is known about the impact of cathode, Vc is the height of the sheath in front of the output voltage on the decay dynamics of the plasma, or cathode, Tc is the cathode temperature, and Jr i,e are { } phrased differently, how the relative bias between cath- the random ion/electron currents given by, ode and anode affects the average diode current. Un- derstanding the IV-curve of a pulsed plasma converter is en 8kBT i,e Jr i,e = { } , (2) important since often this is the only diagnostic avail- { } 4 s πM i,e able which researchers have to deduce several system { } parameters such as cathode and anode work functions, where e is the electron charge, n the plasma density, T plasma density, gap, etc. In the following, this ques- the species temperature in the bulk plasma, kB Boltz- tion is explored using particle-in-cell (PIC) simulations. mann’s constant, and M the particle mass. Notice that 3 the same barrier, Vc, that hinders thermionically emit- Using typical values of the diode parameters d =0.5 mm, ted electrons from entering the plasma also retains ions Tc = 1100 ◦C, Ti = 700 ◦C, d/λ 2 and Js/J0 2 the in the bulk plasma. This clearly indicates that achieving pulse repetition period is found to≈ be 4τ 50 µ≈s. For higher diode currents also leads to faster ion decay. this reason the diode current density was averaged≈ over a The direction of the anode sheath depends on the ap- 50 µs time interval of the plasma decay in the simulations plied bias. Warner and Hansen (1967) [24] noted that if discussed in this article. the anode vacuum potential is sufficiently low (compared to the cathode’s), the anode sheath will be electron- retaining (ion-accelerating). In this configuration the ion Methods current at the anode is simply given by Ji =2Jri. Seeing as the lack of an anode barrier doesn’t increase the diode The simulations discussed in this article were per- current, this configuration only serves to deplete the ion formed with an adapted version of the PIC code Warp density, leading to shorter plasma lifetimes and conse- [25–27]. Specialized electrostatic field solvers were writ- quently lower time-averaged currents. Clearly, a more ten to directly solve Poisson’s equation for the goeme- favorable configuration is to also have an ion-retaining try of problems studied here. The 1D solver uses sheath on the anode side, which similarly as before gives, Gaussian elimination to efficiently solve the 1D Poisson equation[28] while the 2D solver uses superLU[29] to de- compose the finite difference matrix and quickly solve the J =2J (3a) e re linear system. This was found to be much faster than the eV J = (J /2+ J )exp a . (3b) multigrid solver implemented in Warp. MCC handling of i i ri kT a particle collisions[30] was added and extensively bench- If we assume a perfectly neutral and uniform plasma marked against the results of Turner et al. (2013)[31] to without sheaths, an electric field will exist in the gap with ensure accuracy (see Supplemental Material). The argon cross-sections parameterized by Phelps (1994)[32] were E = (Vc Va)/d, where d is the interelectrode gap distance. This− field vanishes when V = V or equivalently used through the LXCat Phelps database, including elas- c a tic scattering processes as well as ion-neutral charge ex- (see Fig. 9 of Ref. [13]) when Vout = φc φa. Although this argument serves to form intuition, in− a real system change. All simulations were done with spatial resolution plasma sheaths will be present and their heights will be of 0.7 µm. This value was chosen as it is at least 30% less affected by the energy distribution of the plasma parti- than the Debye length of the densest plasma simulated, cles. Seeing as the sheaths determine the plasma decay which was 1.03 µm. The PIC timestep was chosen as the rate, correctly calculating their properties are vital to ob- maximum value such that the CFL condition is still satis- taining an accurate picture of the plasma decay. For this fied, assuming a maximum electron energy of 5 eV (much higher than the average energy simulated). This resulted reason PIC simulations were used since the first principle 13 nature of these calculations provide the required accuracy in a timestep of 6.15 10− s. The 1D simulations injected 10 macroparticles× per timestep and assumed an in describing the plasma sheaths. The PIC simulations 2 discussed in the following sections show that the criti- emission area of 1 m . The macro-particle weighting cal point that leads to the highest time-averaged current was then determined by the Richardson saturation cur- corresponds to the flat-band condition. rent for the temperature conditions simulated. The 2D The approximate plasma decay constant at flat-band simulations used 24 cells in the x-direction from which was derived by Rasor (1991) [13], as 1 macro-particle was emitted per cell every timestep. The y-length was taken as 1 m for the calculation of Tc/Ti the emission area and consequently the macro-particle T 2J /J0 τ =2τ i s , (4) weight. In 1D (2D) the initial plasma density was sim- i T 3 d c "1+ 8 λ # ulated with 1000 (100) macro-particles per cell. Particle splitting was implemented to continuously ensure suffi- where τi is the ion crossing time, d is the gap distance, ciently high particle count per cell during the plasma λ the electron mean free path, Js the Richardson satu- decay. The particle splitting algorithm ran every 500 ration current density emitted from the cathode, J0 the simulation steps. It identified cells in which the particle initial diode current density, Ti the average ion temper- count was less than 100 (10) for 1D (2D) and doubled ature in the gap and Tc the cathode temperature. The the particle count in those cells by cloning all its parti- ion crossing time, τi = d/v¯ can be estimated by noting cles while halving the weight. A convergence study was that the ion current is given by Ji = nev¯, where n is the plasma density andv ¯ is the drift velocity of the ions done to ensure simulation parameters were appropriately (Eq. 2), giving chosen as shown in the Supplemental Material. The simulations were performed by introducing a quasi-neutral d πM plasma of a specified density between two parallel plates, τ = =4d i . (5) i v¯ 8k T as in the schematic in Fig. 2. The parallel plates are r B i 4 modelled as perfect conductors, thereby creating Dirich- let boundary conditions for the ends of thez ˆ domain and 0 charges are absorbed when entering the conductor do- mains. During 2D simulations, periodic boundary condi- 0 dd tions are used for thex ˆ domain. The conductor plate on vacuum the left of the computational domain will be referred to 0 00 as the ‘cathode’, and is modelled as a thermionic emitter

i.e. electrons are emitted from the face of the conductor. 0 -cathode 0

The emitted electrons have velocities sampled from the potential difference (V) difference potential thermionic emission distribution derived in the Supple- 0 mental Material. The conductor plate on the right of the 0

computational domain will be referred to as the ‘anode’. 0 0 0 0 0 0 In all simulations the cathode vacuum potential is used as the zero potential reference while the anode’s vacuum potential is varied in order to study the impact of changing Figure 3. Simulation results for a system with a 500 µm interelectrode gap, 10 Torr background argon, initial plasma the output voltage of the thermionic diode. The simula- 12 −3 2 density of 10.2 × 10 cm and 2.2 A/cm thermionic cur- tions are seeded with a neutral argon plasma of specified ◦ rent emitted from the cathode (Tc = 1100 C and φc = 2.1 peak density. The seeded plasma density follows a sine- eV). The diode current as a function of time is shown in the distribution that peaks in the middle of the gap. Simula- top panel for different anode potentials. The average plasma tions of plasma ignition indicate that this is close to the density in the gap is shown in the bottom panel. expected plasma density profile (see Supplemental Mate- rial for further details), confirming the same result from Ref. [33]. The seed ions are assumed to be at the neu- suppressed. As the diode output voltage is decreased, tral gas temperature (for simplicity taken as the average moving into the accelerating regime, the diode current of the cathode and anode temperatures) while the seed at the start of the simulation is increased, but the electrons are injected with a temperature equal to the plasma lifetime is significantly decreased. This leads to cathode temperature, an assumption commonly made in a decrease in the diode current as the plasma density modelling the electrons in inert-gas plasma thermionic diminishes. The highest time-averaged diode current is converters, see Ref. [13] for example. This approach has seen when the diode is in the flat-band configuration, as two main shortcomings, namely, in a real pulsed mode shown in Fig. 4. The same simulations were done with TEC the plasma particles would have non-zero drift ve- different initial plasma densities for which the time- locity due to the ignition pulse and the particle density averaged diode current is also shown in Fig. 4. It was at the electrode surfaces are not as low as predicted by found in all simulated cases that the time-averaged cur- the sine function. Nonetheless this approach is used to avoid the computationally intractable problem of simulating multiple pulse periods. The PIC simulations are evolved up to 50 µs during which the current through the diode is continuously − − tracked, along with several other quantities such as the − spatially resolved plasma density and electrostatic potential. Simulations were performed varying several as- pects of the system including the spacing between the electrodes, the density of the initial plasma, the current density emitted from the cathode, and the density of the background neutral gas.

Results − − 2D Simulations. The first set of results, shown in Fig. 3, is for a system where the interelectrode gap Figure 4. Time-averaged diode current for the output voltage was set to 500 µm, the initial plasma density was set 12 3 cases shown in Fig. 3 as well as cases with different initial to 10.2 10 cm− , and the cathode emission current × plasma densities. In all three cases the time-averaged cur- density was set to 2.2 A/cm2. The simulation results rent peaks at the flat-band condition. Results for both 1D show that as the diode output voltage is increased, and 2D simulations are shown, highlighting the differences in moving into the retarding regime, the diode current is simulation results. 5 rent peaks at flat-band. This same result was also seen Electron motive diagram at different simulation times with simulations of different gap values (250 µm, 1 mm), 1D simulation 2D simulation different background pressures (15 Torr, 25 Torr) and 2.5 other emission current densities (8.5 A/cm2, 3.9 A/cm2). 2.0 1D Simulations. The computational benefit of be- ing able to do 1D simulations that accurately describe 1.5 40 systems with translational invariance is clear. Unfor- tunately, it was found that the specific simulations dis- 1.0 s) Electron motive (eV) motive Electron μ cussed in this paper do not have the same results when 30 performed in 1D as in 2D. 0.0 0.2 0.4 0.0 0.2 0.4 As shown in Fig. 5, it was found that the plasma de- z (mm) z (mm) cay simulated in 2D did not match results from perform- 20 ing the same simulation in 1D. Specifically, 1D simu- 2.4 lations consistently showed faster decay of the plasma time ( Simulation density at the beginning of the simulation compared to 2.2 the 2D simulations, which then slowed down significantly 10 2.0 as the simulation progressed. This leads to lower average current densities in the accelerating regime where 1.8 the plasma density in the early times of the simulations (eV) motive Electron matters most and higher average current densities in the 1.6 0.00 0.02 0.04 0.00 0.02 0.04 retarding regime where the slower decay late in the sim- z (mm) z (mm) ulation matters more, when compared to the 2D results (see Fig. 4). The difference in plasma decay characteristics between Figure 6. Time evolution of the electron motive for both 1D the 1D and 2D simulations are manifested in the electron and 2D simulations, (top) over the full domain and (middle) motive evolution, as shown in Fig. 6. In 2D, the plasma zoomed in near the cathode. The simulations are started with × 12 −3 screening is well captured leading to relatively smooth the same plasma density (10.2 10 cm ), electron temper- electrostatic potential profiles throughout the plasma de- ature, and ion temperature. The case shown is where the anode vacuum level is biased 0.1 V relative to the cathode’s. cay. The time-averaged RMS deviation between the motive at x = 0 and the motive averaged over the x- domain is only 61 meV, showing that small charge inho- mogeneities are well screened by the surrounding plasma. In 1D, however, the reduced dimensionality of the simulations is unable to capture the plasma screening sufficiently leading to abrupt changes in the electrostatic potential as small regions of charge separation form. At 0 and above flat-band, these potential variations lead to

a higher average barrier for beam electrons (see Fig. 6) 0 resulting in an under-prediction of the diode current compared to the 2D case. The bottom panel in Fig. 6 shows 0 00 the evolution of the cathode sheath. Initially the poten-

tial from the cathode is much steeper in 1D than in 2D

0 vacuum -cathode which explains why the initial plasma decay is faster in 1D than in 2D (ions are accelerated into the cathode with potential difference (V) difference potential

0 0 greater force in 1D). In both 1D and 2D simulations the sheath starts out at the same potential as the cathode, 0 0 0 0 0 0 but quickly decreases in energy as the plasma decays and μ an ion-retaining sheath forms. The so-called ”inverse”- mode described by Campanell (2018)[34] is seen in both Figure 5. Average plasma density as a function of time for cases towards the end of the simulation. In Ref. [22] different output voltages showing results from 1D (dashed) and 2D (solid) simulations. Note the differences in the plasma Campanell and Umansky argue that this state is formed density between the two cases, which is the cause for the due to charge-exchange collisions between ions and neu- differences in time-averaged collected current. The simulation trals in the cathode sheath trapping ions and leading to a parameters were as follows: 500 µm gap, 10 Torr background broadening of that sheath towards the anode. Evidence 12 −3 argon, initial plasma density of 2.2 × 10 cm , Tc = 1100 ◦ of this mechanishm is seen in Fig. 7 where it is clear C and φc = 2.1 eV. that ion density shifts from the center of the gap to the 6

Relative ion density in the gap 1D 2D 30 1.0 0 0.9 25 0.8 0 20 vacuum 0.6 00 s)

( 0 -cathode

10 0.3 potential difference (V) difference potential 0

5 00 0 0 0 0 0 0.0 0 0.0 0.1 0.2 0.0 0.1 0.2 z (mm) z (mm) Figure 9. The electron motive averaged over time for diﬀerent anode bias cases. The ﬂat-band condition is shown with a Figure 7. Time evolution of the ion density distribution is dashed line to highlight it. The simulation cases shown here shown for a region in front of the cathode for the same simu- are the same as is shown in Fig. 3. lation case as shown in Fig. 6.

1.5 Tcathode in 1D. Greiner et al. (1995)[35] reported a cathode sheath. It is also notable that the 1D simulation similar× formation of vortices in the electron phase space shows a much earlier accumulation of ions in that region diagram of a thermionic converter. They used 1D PIC which explains why the average diode current is lower for simulations to study this instability (a specific electro- the 1D simulation in the accelerating regime. static instability known as a Pierce-Buneman instabil- A cut of the electron phase space (z vs vz) is shown in ity) and observed the same undamped behavior in their Fig. 8 at different times. The phase space plots show 1D simulations. However, they specifically studied the a beam instability in both 1D and 2D. However, in 2D plasma dynamics at high anode bias, not close to flat- the wavelength of the beam instability increases as time band. Levko (2015)[36] also showed similar beam insta- progresses, indicating a damping of its growth (not seen bilities in 1D PIC simulations of a nitrogen plasma in a in 1D). This damping is due to transverse scattering thermionic converter and studied how they are affected of the electrons off electrostatic waves, something that by electron reflection from the anode. cannot happen in 1D since there E~ = Ezˆ. Evidence of this is that at t = 12 µs, the transverse tempera- me 2 Discussion ture, T = 2k v , is 2.1 Tcathode in 2D but only ⊥ B ⊥ ×

The impact of diode output voltage was studied for ar- v gon plasma-based thermionic diodes. It was found that the maximum time-averaged current is collected when the diode is in the ﬂat-band conﬁguration. This result can be understood by studying the average electron

v motive for diﬀerently biased cases, as shown in Fig. 9. − These results conﬁrm the intuition discussed earlier. At

( highly negative biases the anode sheath becomes ion- accelerating which understandably leads to low electron

current. At highly positive biases the bulk plasma po- v ( tential is higher than the cathode’s vacuum potential.

In this condition ions from the bulk plasma easily have ( enough energy to overcome the ion-retaining sheaths in

front of the cathode, causing fast plasma decay. Towards the flat-band condition, the sheaths in front of both elec- v ( trodes become ion-retaining and single valued, which is the desired condition for slow plasma decay (as discussed earlier). The barrier index (height of cathode sheath relative to the cathode potential) is lowest exactly at flat- Figure 8. Phase space plots of z versus vz for different times band which explains why the maximum time averaged of the same simulation shown in Fig. 6. diode current is seen at flat-band. 7

This result provides a novel technique to measure the SUPPLEMENTAL MATERIAL TO: difference in work functions between the cathode and SIMULATIONS OF ARGON PLASMA DECAY IN anode in a thermionic converter. In practice it is dif- A THERMIONIC CONVERTER ficult to determine electrode work functions in operating thermionic diodes. Typically the work functions will be VELOCITY DISTRIBUTION OF THERMIONICALLY EMITTED ELECTRONS sensitive to temperature, as is the case with dispenser cathodes [37] and refractory metal electrodes that rely on cesium-oxide coverage to achieve useful work func- We start by assuming the velocities of electrons that tions [38]. Furthermore, evaporation and deposition of could participate in thermionic emission (electrons close electrode material on the opposite electrode greatly alters to the top of the conduction band) can be well described the electrode work functions. For these reasons measure- by a Maxwellian distribution function, ments have to be done in operating conditions to be reli- 3 2 2 2 able. The results discussed here indicate that an in situ 1 vx + vy + vz f~v(vx, vy, vz)= exp , (6) measurement of work function differences can be done by √2πσ − 2σ2 " # sweeping the output voltage of an operating pulsed argon plasma diode while recording the time-averaged current. where σ = kB T is the thermal velocity of an electron. A peak in the time-averaged IV-curve indicates the differ- m ence in electrode work functions. This in turn allows one Let the cathodeq work function be φ and assume an ide- to study more carefully the impact of changing operating alized step function in the potential between the interior conditions by tracking changes to work functions rather of the cathode and the vacuum. By assumption there is than solely the output power of the converter, which is a no barrier to escape in thex ˆ ory ˆ directions so these velocity components are not perturbed. In thez ˆ direction, convoluted measurement of many factors. Experimental 1 2 work is currently ongoing to test this proposed method. however, only particles with vz > 0 and 2 mvz > φ can It was also observed that simulating the plasma decay, escape from the cathode. Hence, the electrons that can using a PIC approach, resulted in different behavior in be thermionically emitted are characterized by 1D than in 2D. Specifically, higher (lower) average currents were seen in the retarding (accelerating) regime in 2φ vz∗ > . (7) 1D compared to the 2D simulations. Therefore, in its r m current implementation, this type of simulation cannot The cumulative distribution function for the longitudinal be done with high fidelity in 1D. It is believed that the velocity of electrons that will be thermionically emitted dominant electron thermalization mechanism simulated just prior to emission, is thus given by: is scattering off electrostatic waves, which excludes transverse scattering in the 1D case. Adding an anomalous 0, v < 2φ scattering cross section in this case to enable transverse 2φ z∗ m P vz vz∗ vz > = (8) thermalization could recover fidelity in such simulations. ≤ m !  A , v q 2φ r  B z∗ m This is left to future work. It was also noted that other ≥ q authors have studied such 1D beam instabilities in simi- where  lar systems using PIC codes (Ref. [35, 36]). Studying the vz∗ dampening of these instabilities in 2D in further detail is 1 vz∗ 2φ/m A = f(vz)dvz = erf erf also left to future work. √ 2φ 2 √2σ − √2σ Z m " p !# (9a)

ACKNOWLEDGEMENT ∞ 1 2φ/m B = f(vz)dvz = 1 erf . √ 2φ 2 − √2σ Z m " p !# The authors are grateful to Phil Miller for his assis- (9b) tance in improving the parallel scaling of the internal version of Warp used for the simulations discussed in this We can now get the distribution function by diﬀerentiat- article. ing the CDF above, giving

2φ 0, vz∗ < m AUTHOR CONTRIBUTIONS 2 1 vz∗ f (v∗)=  exp 2 q (10) inside z  √2πσ − 2σ 2φ  , vz∗  1 √2φ/m m All authors contributed to the code development 2 1 erf ≥ − √2σ q needed for the discussed simulations. RG ran the sim-  ulations and wrote the manuscript with input from the Finally, we compute the velocity distribution function other authors. immediately outside the cathode. Conservation of energy 8 requires that the electron kinetic energy be reduced by φ 0V upon escaping the cathode, so anode 00

0 2 2φ vz′ = vz∗ . (11) r − m 0 We now use the change-of-variables rule for probabil- 00 ( ity distributions to calculate the distribution of vz′ from 00

Eq. 10, giving, t 0 0, vz′ < 0 2 0 1 vz′ +2φ/m 0 exp 2 foutside(vz′ )=  v′ √2πσ 2σ z 2 , v′ 0  2 φ 1 √2φ/m z √vz′ + m 1 erf ≥ 0  2 √2σ − 0 (12) 000 0 00 0 00 000 0 00 0 00  z z Seeing as there is no barrier in thex ˆ ory ˆ directions we have Figure 10. Argon ion density as a function of time and space during plasma ignition as simulated in pypd1 [39] compared foutside(v′ )= finside(vx) (13a) x to the current authors’ implementation of the same physics 2 foutside(vy′ )= finside(vy). (13b) in Warp. Simulated saturation current was 0.14 A/cm . In the typical case of thermionic converters, cathode work functions are on the order of a few eV while tem- Warp 1D versus 2D test showing anode current during plasma ignition 2 peratures are on the order of 0.1 eV, thus kBT <<φ. In gap = 0.5 mm, Jsat = 0.95 A/cm , Vanode = 25 V, PAr = 1 Torr this regime we can take 1.0 Warp 1d Warp 2d 2 2φ 2φ 0.8 vz′ + )

m ≈ m 2 0.6

and Eq. 12 simpliﬁes to (A/cm 0.4 anode J

0, vz′ < 0 0.2 f (v′ )= 2 (14) outside z vz′ vz′ ( σ2 exp 2σ2 , vz′ 0 − ≥ 0.0 0 20 40 60 80 100 This distribution is used to sample thez ˆ direction veloci- Time (ns) ties of electrons injected in the PIC simulations discussed in the main paper. Figure 11. Diode current as a function of time during plasma ignition as simulated in 1D and 2D in Warp. CODE BENCHMARK that the MCC implementation in the modified Warp code The MCC implementation, we added to Warp, was behaves the same in 1D as in 2D. This indicates that the benchmarked against the results of Turner et. al. differences seen in 1D and 2D as highlighted in the article (2013)[31]. The results of this benchmark is given in are due to the dimensional dependence of the problem, Fig. 12 not the PIC implementation. In order to also test the Warp implementation in a more appropriate situation to the simulations performed in this study, results were compared to those from the PLASMA DENSITY PROFILE AFTER IGNITION PIC code oopd1, through the python wrapper pypd1 [39]. PULSE The plot in Fig. 10 shows the ion density evolution during plasma ignition in a 1 mm diode over 200 ns. The simulations discussed in this article started with Finally a test of the 1D versus 2D implementations of seeded plasma densities distributed according to a sine the code was done. The plasma ignition in a 500 µm gap function. This distribution was chosen after simulations diode with 1 Torr of argon and the anode biased to 25 of plasma ignition indicated that it is a good approxima- V was simulated. The current collected on the anode as tion for the plasma density formed, as shown in Fig. 13. a function of time as calculated in 1D and 2D in Warp. The seed density specified in the labels of figures refers to Results for this test are shown in Fig. 11. The test shows the plasma density at the maximum point (in the middle 9

-MCC implementation

Warp Turner et. al. (2013)

0.14

0.12

0.10 0.08

density density (+ (+ + 0.06

0.04

0.0 0.5 1.0 x/L

Figure 12. Helium ion density in a capacitively coupled discharge for three cases as presented by Turner et. al. (2013)[31], in their work to benchmark PIC-MCC implementations. Results from the current authors’ implementation of MCC in Warp[25] is overlayed with the literature results to demonstrate accuracy of the implementation.

of the gap). where Vsim is the volume of the computational domain, The initial plasma was seeded as follows: Let ninit the desired peak initial density and we the macro- X1, ..., XN be an i.i.d. sample from U(0, 1) (uniformly particle weight calculated from the specified thermionic distributed between 0 and 1), where N is the desired emission current density and cathode area combined with number of seed macro-particles. The z position of each the specified NPPC injection rate. In 2D the x position of particle was then calculated according to the seed particles were simply set by another i.i.d. sample D from U(0, xmax). The electron and ion velocities were zi = arccos(1 2Xi), sampled from Maxwellian distributions at the specified π − temperatures for each species. where D is the interelectrode gap. The statistical weight of each particle was given by 2 V n W = sim init , π Nwe CONVERGENCE STUDY

A study to check whether the initial number of particles used and the number of particles injected per timestep was suﬃcient to achieve converged results was performed by running a case with a 500 µm gap, cathode temperature of 1100 ◦C, cathode work function of 2.1 12 3 eV, initial peak plasma density of 10.2 10 cm− and background argon pressure of 10 Torr.× The simulation was run up to 10 µs of simulation time while comparing time traces for the diode current and plasma density. The results for 1D and 2D runs are shown in Fig. 14. In the simulations discussed in the main paper 1000 (100) seed particles per cell were used in 1D (2D) while injecting Figure 13. Plasma density after ignition as simulated in Warp, 10 (1) particles per cell per timestep from the cathode. by applying a 25 V bias across a diode of 500 µm gap for 100 During the convergence study all particle numbers were ns. Also shown is a sine function scaled to have a wavelength doubled without any signiﬁcant changes in the simulation of d/2. results indicating convergence.

1 [9] I. Langmuir, Physical Review 2, 450 (1913). [10] R. Y. Belbachir, Z. An, and T. Ono, 1 Journal of Micromechanics and Microengineering 24, 085009 (2014). 1 [11] S. Meir, C. Stephanos, T. H. Geballe, and J. Mannhart, 5 dd Journal of Renewable and Sustainable Energy , 043127 (2013).

1 μd [12] R. Wanke, G. W. J. Hassink, C. Stephanos, 1 μ I. Rastegar, W. Braun, and J. Mannhart, 1μd μ Journal of Applied Physics 119, 244507 (2016).

1μd [13] N. Rasor, IEEE Transactions on Plasma Science 19, 1191 (Dec./1991).

1μ [14] F. G. Baksht, G. A. Dyvzhev, A. M. Martsinovskiy, B. Y. Moyzhes, G. Y. Dikus, E. B. Sonin, and V. G. Yuryev,

1 “Thermionic converters and low-temperature plasma,” 1 NASA STI/Recon Technical Report N (1978). 51 1 [15] K. Hernqvist, Proceedings of the IEEE , 748 (1963). [16] D. R. Wilkins and E. P. Gyftopoulos, Journal of Applied Physics 37, 2892 (1966). [17] P. E. Oettinger and F. N. Hussman, IEEE Transactions on Plasma Science 6, 83 (1978). 1 [18] F. Huﬀman, A. Sommer, C. Balestra, D. Briere, and P. Oettinger, (1976). dd [19] V. Zherebtsov and V. Talanova, Pisma v Zhurnal 1 μd μ Tekhnischeskoi Fiziki 2, 124 (1976).

1μd [20] J. McVey, in Proceedings of the 25th Intersociety Energy Conversion Engineering Conference, μ Vol. 2 (IEEE, Reno, Nevada, 1990) pp. 357–361.

1μd 1μ [21] Rasor Associates, Inc., Sunnyvale, Calif. (USA), Advanced Thermionic Energy Conversion. Progress Report, September 1, 1974–August 31, 1975 , Tech. Rep. COO-2263-4, 7199927 (1975).

[22] M. D. Campanell and M. V. Umansky, 11 Plasma Sources Science and Technology 26, 124002 (2017). 1 [23] L. Wolff, W. Veltkamp, J. Schoonen, and H. Hendriksen, eds., Thermionic energy conversion : specialist conference Eindhoven, The Netherlands October 11-12, 1989 : Figure 14. Test of simulation convergence with respect to proceedings (Eindhoven University of Technology, 1990) macro-particle count in 1D (top) and 2D (bottom). The pp. 71–95. number of macro-particles per cell (Nseed) used to simulated [24] C. Warner and L. K. Hansen, the seed plasma was varied as well as the number of macro- Journal of Applied Physics 38, 491 (1967). particles injected per cathode cell per timestep (Ninject) and [25] See http://warp.lbl.gov/ for code details. the results compared. [26] A. Friedman, D. P. Grote, and I. Haber, Physics of Fluids B: Plasma Physics 4, 2203 (1992). [27] D. P. Grote, in AIP Conference Proceedings, Vol. 749 (AIP, Berkeley, California (USA), 2005) pp. 55–58. [28] C. Birdsall, Plasma physics via computer simulation (McGraw-Hill, New York, 1985). [1] G. N. Hatsopoulos and E. P. Gyftopoulos, Thermionic [29] X. S. Li, ACM Trans. Math. Software 31, 302 (2005). Energy Conversion (MIT Press, Cambridge, 1973). [30] C. Birdsall, IEEE Transactions on Plasma Science 19, 65 (1991). [2] D. B. Go, J. R. Haase, J. George, J. Mannhart, [31] M. M. Turner, A. Derzsi, Z. Donkó, D. Eremin, R. Wanke, A. Nojeh, and R. Nemanich, S. J. Kelly, T. Lafleur, and T. Mussenbrock, Frontiers in Mechanical Engineering 3, 13 (2017). Physics of Plasmas 20, 013507 (2013). [3] K. A. Abdul Khalid, T. J. Leong, and K. Mohamed, [32] A. V. Phelps, Journal of Applied Physics 76, 747 (1994). IEEE Transactions on Electron Devices 63, 2231 (2016). [33] J. L. Lawless and S. H. Lam, [4] P. N. Clark, in AIP Conference Proceedings, Vol. 813 Journal of Applied Physics 59, 1875 (1986). (AIP, Albuquerque, New Mexico (USA), 2006) pp. 598– [34] M. D. Campanell, Phys. Rev. E 97, 043207 (2018). 606. [35] F. Greiner, T. Klinger, and A. Piel, [5] E. P. Gyftopoulos and G. N. Hatsopoulos, Physics of Plasmas 2, 1810 (1995). Electrical Engineering 82, 108 (1963). [36] D. Levko, Physics of Plasmas 22, 073501 (2015). [6] J. B. Ashton, S. E. Clark, W. Kokonaski, D. Kraemer, [37] B. Levush, Y. Lau, and K. Jensen, J. J. Lorr, M. N. Mankin, D. J. Menacher, P. D. IEEE Transactions on Plasma Science 28, 772 (2000). Noble, T. S. Pan, A. De Pijper, and L. L. Wood, [38] J.-L. Desplat, in Functionally Graded Materials 1996 “Combined heating and power modules and devices,” (Elsevier, 1997) pp. 639–646. (2020). [39] J. Verboncoeur, A. Langdon, and N. Gladd, [7] C. Crowell, Solid-State Electronics 8, 395 (1965). Computer Physics Communications 87, 199 (1995). [8] C. D. Child, Physical Review (Series I) 32, 492 (1911).