Impact of the photoelectric threshold sensitivity on the work function determination—Revealing ultra-low work functions of caesiated surfaces

A standard experimental procedure for the in situ determination of the absolute electronic work function is the measurement of photoelectric currents extracted by electromagnetic irradiation with varying photon energies. Due to a finite detection limit in the experiment and often limited available photon energies (e.g., restricted lower limit or energy resolution), the evaluation of the work function is then typically performed by fitting or extrapolating the measured photoelectric yield curve, for which the application of the semiclassical Fowler theory¹ for metal–vacuum interfaces is common practice.^2–10 Compared to the methods based on thermionic emission, field electron emission, or a contact potential electrode, the photoelectric method has the advantage that high temperatures, high electric fields, and contact with other materials are not required.^11,12

Although photoelectric yield curves can, in principle, easily be determined, it requires some—often underestimated—effort to deduce reliable work function values from it. First, the energy resolution of the irradiated light must be appropriate, i.e., a sufficient amount of photon energies with a narrow spectral bandwidth or a continuously tunable light source is required, and second, a sufficient signal-to-noise ratio (SNR) for the photocurrent detection must be provided. The SNR depends on the power irradiation onto the surface (for a fixed illumination area) and on the level of the dark current, i.e., the current that is measured without irradiation. The dark current represents the noise level for the photoelectric measurement and is determined by the quality of the electric circuit, external noise sources, the conditions in the vacuum system, etc. The SNR is especially decisive in the vicinity of the photoelectric threshold, where the photocurrents generally decrease rapidly when the photon energies hν approach the work function χ. In this region, i.e., hν ⪆ χ, the photocurrents must be tracked very carefully so that the Fowler method can be applied reliably.

The present paper is a continuation of the work described in Ref. 13, where the importance of the threshold determination is already discussed as decisive for the accuracy of determined work function values via the Fowler method and where the work function dynamics of a metal surface that is subject to Cs adsorption is investigated. In this work, the same adsorption system is analyzed, but the irradiation frequencies of the work function measurement setup are extended to photon energies $<3$ eV and the SNR is substantially enhanced. By these improvements, the impact of an insufficient experimental photoelectric sensitivity is now impressively and reproducibly shown in a much wider range of work function values. Moreover, it is now demonstrated that ultra-low surface work functions can be generated by the caesiation of metal substrates under vacuum conditions of some 10⁻⁶ mbar.

The photoelectric threshold is defined as the minimum photon energy required to extract an electron from the interior of a solid state into vacuum. For metals at absolute zero temperature, the photoelectric threshold is equal to the energy difference between the Fermi level and the vacuum level and, thus, to the electronic work function χ.¹⁴ For temperatures T > 0 K, a portion of the electrons at the Fermi edge is elevated to energies beyond the Fermi level due to thermal excitations, which is illustrated in Fig. 1 by plotting the product of the 3D density of states of a Fermi gas and the Fermi–Dirac distribution as a function of the electron energy ɛ_e for T = 300 and 1000 K, respectively. The chemical potential is set to a good approximation equal to the Fermi energy ɛ_F (here, 4 eV). The energy dispersion is of the order of k_BT, which is 0.026 eV at 300 K and 0.086 eV at 1000 K, with k_B denoting the Boltzmann constant. Since all the measurements in this work are performed at about room temperature, the energy smearing at the Fermi level is, hence, very weak and the photoelectric threshold of the metal can be set in a good approximation equal to the work function.

One of the most prominent theories for photoelectric electron emission at metal–vacuum interfaces is the Fowler theory. On the basis of Sommerfeld’s theory of free electron gas, Fowler assumed that the photoelectric current per unit incident light intensity, I_ph,u, is to a good approximation proportional to the density n_e,ph of electrons in the metal whose kinetic energy component normal to the surface exceeds the surface potential barrier after the augmentation by hν due to photon absorption.¹ Accordingly, the electron density determining the photoelectric current is calculated as

where u is the velocity component of the electrons normal to the surface and $u′=2(εF+χ−hν)/m$ is the minimum velocity component normal to the surface such that after the addition of hν, the electron can leave the solid state. m and T denote the free electron mass and the absolute temperature of the irradiated surface, respectively. By solving the ρ- and φ-integrals and by using the substitution variable y = 1/2mu²/(k_BT), Eq. (1) can be written as¹ 

with κ ≔ (hν − χ)/(k_BT) and the constant $C̃≔22π(kBm)3/2h−3≈1.36×1021$ m⁻³ K^−3/2.

In the inset of Fig. 1, the evolution of n_e,ph according to Eq. (2) is demonstrated as a function of hν − χ for T = 300 and 1000 K, respectively. As can be seen, the number of photoelectrically releasable electrons decreases strongly for hν → χ with hν > χ, and consequently, a high sensitivity is required in the experiment to detect photocurrents as close as possible to the work function. For hν ≤ χ, n_e,ph continues to drop steeply: while the decrease at 1000 K is two orders of magnitude within 0.4 eV, at 300 K, it is already within 0.1 eV.

Since Eq. (2) is not well suited for the experimental application due to the typically unknown Fermi energy (in particular in the context of adsorption processes), approximations have to be made. If only photon energies close to the work function are considered, i.e., hν ≈ χ, Eq. (2) can be simplified to¹ 

and further to¹ 

with the constant $C≔C̃kB1/2εF−1/2$ and f denoting the temperature dependent function

The predicted progression $Iph,u∝ne,ph2ndapprox.$ has proven excellent experimental agreement for irradiation frequencies near the photoelectric threshold,^1,15,16 and thus, it is routinely applied as the basis for photoelectric work function evaluations.^6,8,10,17 An even more succinct simplification is obtained for the temperature limit T → 0 K,¹ 

whose application is also common practice.^2–5,7,9 Akbi et al.¹⁰ pointed out, however, that the T = 0 K approximation may introduce errors of the order of k_BT, and thus, Eq. (4) should be preferred (as long as ɛ_F is not known).

In order to track the progression of the frequency dependent photoelectric yield experimentally, a tunable light source with a narrow spectral bandwidth is required. Since the application of a laser with a continuously variable frequency is expensive, a set of several discrete photon energies can be used instead, provided by lasers or light-emitting diodes, for instance. In this work, the discrete photon energies are provided conveniently by the broadband emission of a high pressure mercury lamp (Osram HBO 100W/2) in combination with a set of interference filters (10 nm nominal FWHM), as done in Ref. 13. Different from the setup used in Ref. 13, however, the interference filter set has been extended: 11 filters with central transmission wavelengths between 430 and 852 nm have been added to the former nine filters in the range of 239–405 nm in order to provide lower irradiation frequencies. The interference filters are mounted in a motorized filter wheel in front of the lamp, which can house six filters. The setup is installed at the laboratory experiment ACCesS,¹⁸ and as illustrated in Fig. 2, the transmitted light is collimated via two quartz lenses and is directed through a quartz viewport into a cylindrical stainless steel vacuum chamber. In the chamber, an electrically insulated sample holder is attached to the bottom plate, and the surface temperature of installed samples at the sample holder is measured with a K-type thermocouple. The illumination spot on the sample surface has a diameter of about 15 mm. The emission spectrum of the mercury lamp together with the respective transmitted spectra through the applied filters is recorded with a spectrometer and is depicted in Fig. 2(a). The corresponding transmitted mean photon energies ɛ_ph through the filters are determined after calibrating the transmission spectra and yield ɛ_ph = 5.04–3.05 eV for the formerly used restricted filter set and ɛ_ph = 5.04–1.45 eV for the extended filter set into the near-infrared (energy resolution mostly in the range of 0.1–0.3 eV). The irradiation power onto the sample surface is measured with a radiant powermeter (Newport model 818-UV/DB) and is shown in Fig. 2(b).

The photoemitted electrons are drawn to the grounded vessel walls by means of an electric field above space charge limitation, which is done by applying a bias of −30 V to the sample holder and surface. The execution of the work function measurement is realized by an automated data acquisition and evaluation procedure by using a Keithley 6487 picoammeter and a LabVIEW program: while the current is continuously recorded, a shutter opens in the optical path for each of the six filters in the filter wheel for a period of typically a few seconds, as demonstrated in Fig. 2(c). The respective absolute photocurrents I_ph(ɛ_ph) are subsequently calculated by averaging the current measured during the time the shutter is open and subtracting the dark current I_dark, which is, in turn, determined by averaging the currents measured when the shutter is closed before and after the respective irradiation phase. With the present setup, the SNR, i.e., the ratio of the photocurrent to the dark current, must be ≳0.03 so that the photocurrent can be distinguished from the dark current. The dark current is typically of the order of 10⁻¹¹–10⁻¹⁰ A, and thus, photocurrents down to the pA range can be detected. In case the dark current is highly dynamic during one measurement (factors influencing the dark current are discussed in Sec. III B), a polynomial interpolation and subsequent subtraction of the dark current is preferred. Finally, the absolute photocurrents are divided by the square of the measured surface temperature and the filter dependent relative intensities F(ɛ_ph) of the irradiated light [calculated from Fig. 2(b) after the conversion of the irradiation power from mW to photons/s], and the logarithm of I_ph/(FT²) is plotted as a function of the photon energy, which according to Eq. (4) is fitted via

with the work function χ and the constant A being fit parameters. The representation of the photocurrents according to Eq. (8) corresponds to a (classical) Fowler plot,¹ and the respective work function evaluation is known as the Fowler method of isothermal curves. In accordance with the considerations given in Ref. 13, the six interference filters that are inserted into the filter wheel are chosen after identifying the filter with the lowest photon energy with which a photoelectric response is still obtained in order to meet the condition hν ⪆ χ. Since one measurement takes typically about 2 min [see Fig. 2(c)], the identification of the appropriate filter set can last several minutes. This is relevant for dynamic surface work functions as in the case of the conducted caesiation processes in this work, where changes on the minute scale can occur (see discussion in Sec. IV).

The vacuum vessel with a diameter of 15 cm and a height of 10 cm is evacuated by using a turbomolecular and roughing pump and provides vacuum conditions of 10⁻⁶–10⁻⁵ mbar (unbaked, limited by Viton O-ring seals). The residual gas pressure is measured with a cold cathode gauge, and its composition is monitored with a differentially pumped residual gas analyzer (RGA, not yet calibrated for absolute partial pressures). The alkali metal Cs can be evaporated into the experiment by means of a Cs oven, which is attached to the bottom plate of the vacuum vessel and which was designed and manufactured at IPP Garching (Germany).¹⁹ The Cs oven contains a liquid Cs reservoir (1 g Cs ampoule), which can be heated in a controlled manner to finely adjust the Cs evaporation rate. The nozzle of the oven is directed toward the sample holder so that installed metal samples can be coated via Cs adsorption, which is well known to strongly reduce the surface work function due to the high electropositivity of Cs.^20,21 To monitor the Cs evaporation rate, a surface ionization detector (SID) is located a few millimeters behind the oven nozzle. The SID consists of two tungsten filaments (300 $µ$m diameter) that are biased against each other with $≈60$ V and are ohmically heated by passing a current of 2–3 A through each of them. Cs atoms are ionized when they approach the first hot tungsten filament and are subsequently accelerated and collected onto the second filament. The resulting current is proportional to the impinging flux of Cs atoms onto the first filament and is thus a measure of the Cs outflow from the oven nozzle.²² Moreover, the neutral Cs density n_Cs in the chamber is measured along the 15 cm line-of-sight parallel to and close to the sample surface by means of a tunable diode laser absorption spectroscopy (TDLAS) system (detection limit $≈2×1013$ m⁻³).²³ 

The ACCesS experiment is mainly dedicated to study the work function performance and H⁻ yield of caesiated metallic surfaces in a low pressure low temperature hydrogen plasma environment.^17,24,25 This is done in view of negative hydrogen ion sources that rely on the production of negative ions on caesiated converter surfaces,²⁶ which are used for fusion and accelerator applications for instance.²⁷ The discharges at ACCesS are generated via inductive radio frequency (RF) coupling by using a planar solenoid on top of the vessel and an RF generator operating at 27.12 MHz and with 600 W maximum output power. Gas feeding is done by using calibrated mass flow controllers.

It is well known that the work function of a surface strongly depends on its physical and chemical conditions, such as surface roughness, crystallographic orientation, the presence of contaminants, and surface adsorbates. The analysis of the photoelectric work function evaluation in this work is performed with a polycrystalline Mo substrate (3 $µ$m Mo coating on a 30 × 30 × 5 mm³ Cu sample) in a vacuum environment of about 5 × 10⁻⁶ mbar at room temperature. The Mo substrate, whose work function is $>4$ eV (experimentally measured after hydrogen plasma cleaning, see Sec. IV, and in accordance with literature values^28–30) is exposed to Cs vapor in order to provide a strongly varying work function upon surface adsorption. Since the nozzle of the Cs oven in the vacuum chamber is set to a temperature of $≈550$ K to avoid Cs sticking, the temperature of the Mo substrate increases slightly by a few degrees over a period of several hours due to thermal radiation (∼1 K/h). It has to be noted that the Mo substrate used in this section is pre-caesiated from an earlier campaign, leading to an already reduced work function $<4$ eV.

In Fig. 3(a), measured surface work functions during a re-caesiation process of the Mo substrate are shown, where the neutral Cs density in the gas phase is continuously monitored via laser absorption. Photocurrents obtained with some of the irradiated photon energies together with the dark current are plotted in Fig. 3(b). The initial work function of the degraded Cs adlayer is 3.5 eV. By ramping up the Cs evaporation rate, the photocurrents significantly increase and the surface work function is reduced. The evaluation of the measured photocurrents via a Fowler plot [Eq. (8)] using the extended interference filter set (applied filters in the filter wheel are appropriately adjusted) reveals a drastic reduction of the surface work function to slightly below 2 eV at a reached Cs density of 2.6 × 10¹⁵ m⁻³ (t ≈ 2.1 h). Under the assumption of an isotropic thermal distribution of Cs within the vessel, the corresponding Cs flux onto the surface can be calculated via Γ_Cs = n_Cs⟨v_Cs⟩/4 = 1.6 × 10¹⁷ m⁻² s⁻¹, with ⟨v_Cs⟩ being the mean thermal velocity of the Cs atoms (temperature ∼400 K, obtained by the TDLAS line profile). Ripples in the measured Cs density are a characteristic of the Cs oven used and result from a slightly oscillating temperature regulation (≈±5 K) of some of the oven’s inner surfaces.

By keeping the Cs density above 10¹⁵ m⁻³ for $≈1$ h, the surface work function of slightly below 2 eV is maintained. The generation of such a low work function (here in the range of bulk Cs^29,30) is remarkable due to the moderate vacuum level in the experimental chamber and is further discussed in Sec. IV. As soon as the Cs evaporation is reduced and finally shut down (t > 3 h), the photoelectric yields decrease rapidly and the photoelectric threshold and, thus, the work function increase strongly.

As can be seen in Fig. 3(a), the work functions χ_restr that are evaluated in case the restricted filter set (minimum photon energy = 3.05 eV; see Fig. 2) is used are higher compared to the values χ_ext ≲ 2.8 eV that are evaluated with the extended filter set. The deviation becomes larger as χ_ext becomes smaller, and when χ_ext ∼ 2 eV, a 0.7 eV larger value is evaluated with the restricted filter set. The reason behind this is demonstrated in the following: an exemplary Fowler plot created from the measured photocurrents at t ≈ 3.0 h for photon energies in the range of ɛ_ph = 2.04–5.04 eV is shown in Fig. 4(a), with 2.04 eV being the lowest photon energy with which photoelectric response is detected (next lower photon energy without photoelectric signal is 1.88 eV). The x-error bars result from the transmitted wavelength ranges of the interference filters with a cutoff at 5% of the normalized transmitted spectrum. A fit of the five photocurrents obtained for the lowest photon energies, i.e., ɛ_ph = 2.04–2.57 eV (gray shaded area), yields χ_fit = 1.95 ± 0.01 eV, with ±0.01 eV being the standard error of the performed fit. The fitted function is extrapolated to ɛ_ph ∼ 5 eV, which shows that the photocurrents for photon energies far above the evaluated threshold of 1.95 eV are underestimated by the Fowler theory, with the deviation becoming larger the further the photon energies are above the threshold. Consequently, a shift of the fitting range consisting of five consecutive photocurrents (ɛ_ph,max − ɛ_ph,min ≲ 1 eV) to higher ɛ_ph,min leads to a gradually increasing steepness of the Fowler fit and, thus, to the evaluation of increasing work function values. The dependence of the evaluated work function as a function of the fitting range is shown in Fig. 4(c), where the distances of ɛ_ph,min and ɛ_ph,max of each fitting range to the threshold of χ = 1.95 eV (which is expected to be close to reality) are also depicted. In the case that the minimum available photon energy is limited to 3.05 eV, the corresponding Fowler fit yields χ_fit = 2.63 ± 0.02 eV, as depicted in Fig. 4(a) with the red shaded area indicating the fitting range 7. This fit also describes the progression of the photocurrents in the respective range very well and is thus easily misleading in the absence of lower irradiation frequencies. As already mentioned, the Fowler T = 0 K approximation [Eq. (7)] is also routinely applied by many authors for the work function evaluation. Therefore, the respective evaluations are demonstrated in Fig. 4(b) for the sake of completeness, where the x-intercept of a linear least-squares regression of the square root of the photocurrents yields the work function (also called linearized Fowler plots). The results of the two methods are compared in Fig. 4(c) and show a good agreement in the order of k_BT, in agreement with the statements of Ref. 10.

By the extension of irradiation frequencies, it is now shown that the photoelectric yield progression can differ significantly from the Fowler prediction for photon energies ≫χ. Thus, an insufficient lower limit of the available photon energies results in the evaluation of an overestimated work function, which is not necessarily apparent from the performed Fowler fit. In the case of using the restricted filter set, an overestimation of 0.7 eV with ɛ_ph,min − χ ≈ 1.1 eV is given, and hence, a severe underestimation of the given work function dynamics would be the result. One reason behind the failure of the Fowler theory for hν ≫ χ is the fact that the Fowler fit function is derived with near threshold approximations (see Sec. II A) and is thus only accurate in a limited range above the work function (typically 0.5–1.0 eV). The near threshold approximations lead to an underestimation of the photocurrents for photon energies far above the work function, which becomes more pronounced the lower the Fermi energy is.¹³ Furthermore, photon frequency dependent coefficients such as the reflectance of the surface or the transition probability of electrons through the surface potential barrier and factors such as electronic band structures and limitations of the free electron model are generally not considered in the Fowler theory. Such factors can, however, strongly influence the frequency dependent photoelectric quantum efficiency of the surface.^2,31 Since the photocurrent from the lowest photon energy mainly determines the threshold of the Fowler fit, a sufficient amount of available photon energies with a narrow spectral bandwidth is indispensable.

An indication of overestimated work functions might be given by considering the near threshold approximations that are performed in Sec. II A in order to derive the expression for the Fowler fit [Eq. (2) $→approx.$ Eq. (3) $→approx.$ Eq. (4)]. The higher the photon energies are above the threshold, these approximations induce an error that gets more pronounced. As described in detail in Ref. 13, it is for a certain work function and constant Fermi energy

with the differences being larger for smaller ɛ_F. Here, I_ph(ɛ_ph) denotes the actually measured photocurrent for a particular photon energy and $IphEq.(x)$ represents the theoretical photocurrent based on the respective equation from the Fowler theory. We now consider the scenario of the decreasing surface work function during the caesiation process presented in Fig. 3 and the case that 3.05 eV is the lowest available irradiation photon energy for the work function determination, i.e., the application of the restricted filter set. Then, state “0” is defined when the work function χ₀ ≈ 3 eV is reached, i.e., when a photocurrent for this photon energy is detected for the first time (at t = 0.75 h in Fig. 3), and hence, the five applied photon energies for the Fowler evaluation are as close as possible to the expected real work function. The theoretical photocurrents obtained for ɛ_ph = 3.05 eV and χ ≲ 3 eV can be subsequently calculated via¹³ 

and

In Fig. 5, $IphEq. (4)$ and $IphEq. (3)$ are plotted according to Eqs. (10) and (11), respectively (T = T₀ = 306 K). Since in the present case the evolution of the Fermi energy is unknown, physically reasonable limits are used for the calculation of $IphEq. (3)$⁠: ɛ_F,0 = ɛ_F,Mo = 6.77 eV³² and ɛ_F = ɛ_F,Cs = 1.59 eV.³⁰ The region in which according to Eq. (9) the actually measured photocurrent I_ph(3.05 eV) for χ ≲ 3 eV should lie is then given by the gray shaded area. This is confirmed by the measured values of I_ph(3.05 eV) that are plotted against the work function χ_ext evaluated with the extended filter set. The plot of I_ph(3.05 eV) against the work function χ_restr evaluated with the restricted filter set shows, however, an increasing deviation from the gray shaded area for χ_restr < 2.86 eV. At the same time, the fitting constant A of the Fowler fit increases, which indicates an erroneous work function evaluation¹³ and which is not the case when using the extended filter set. The correction of the work functions (χ_restr → χ_corr) is now illustrated by the green arrows: χ_restr is shifted to lower values until the condition $IphEq. (3)(χ)=Iph(3.05eV)(χcorr)$ is met since $IphEq. (3)$ constitutes the upper limit for the evolution of I_ph(3.05 eV) according to the Fowler theory. The corrected work functions are plotted in Fig. 3, and for the case of χ_restr = 2.63 eV for instance (evaluation see Fig. 4), the correction method yields χ_corr = 2.07 eV, which is only 0.12 eV larger than the value of χ_ext that is expected to be close to reality since all conditions of the Fowler theory were met. With this, the evolution of the fitting constant A as a tool for a plausibility check could be confirmed and the correction procedure was useful to come closer to the real work function. Finally, the control parameter A together with the correction procedure is a useful tool to decide on the application of lower photon energies in order to obtain absolute work function values with highest accuracy.

In addition to the necessity of an appropriate photon energy range and resolution for the sample irradiation, the SNR of the photoelectric measurement is an equally decisive parameter for the accuracy of the work function determination. Apart from the sensitivity of the ammeter, the SNR is mainly determined by the irradiated power density onto the surface and the level of the dark current in the electric circuit. The irradiated power density at the present setup is moderate [∼0.1–1 mW; see Fig. 2(b)] in order to ensure that the surface is not modified by strong photon irradiation. To minimize the dark current, external noise sources (e.g., induction of electrostatic or electromagnetic interference), ground loops, tribo- and piezoelectric effects, etc., should be avoided. At the setup, unnecessary devices are switched off or are disconnected from the system during the work function measurement (e.g., optional heating of the sample holder via pulse width modulation) and the measurement itself is performed remotely, which is possible due to the automated measurement system. The active Cs evaporation into the vacuum chamber can lead to an increase of the dark current [see Fig. 3(b)], which is probably due to ionized Cs atoms and/or a higher sample leakage current caused by the work function reduction. The largest impact at the setup is given, however, by the SID diagnostic. During a similar caesiation process as shown in Fig. 3, the oven SID is activated at a point where χ ≈ 1.9 eV and n_Cs ≈ 1.5 × 10¹⁵ m⁻³. With this, the dark current instantly increases by about three orders of magnitude. As a consequence, photocurrents can now only be detected for photon energies down to 2.86 eV (detection limit $≈2×10−9$ A), as is shown in Fig. 6 by the measurement labeled 1. Comparable to Fig. 4(a), the respective Fowler evaluation yields an overestimated work function of 2.59 eV. After some time, the dark current increases further and the photocurrents for ɛ_ph = 3.69–2.86 eV are then also below the detection limit (measurement labeled 2). The high dark current resulting from the oven SID can be attributed to thermionic electron emission and/or ionized Cs atoms due to the hot tungsten filaments.²² When the oven SID is switched off again, the dark current decreases considerably and the photocurrent detection limit is $≈10−11$ A. With this, the SNR is high enough to measure photocurrents for photon energies below 2.86 eV (measurement labeled 3) and a work function of 1.9 eV is evaluated again (measurement with an adjusted interference filter set not shown in Fig. 6). This impressively demonstrates that the noise level on which the photoelectric measurement is based must be critically taken into account when the accuracy of the determined work function value is evaluated. Measurements performed with the SID operated during the caesiation process^{13,17,24,25,33,34} therefore represent only an upper limit of the actual work function.

The above-described enhancement of the photoelectric threshold sensitivity revealed work functions during caesiation in vacuum that are substantially lower than previously measured at the ACCesS experiment.^13,17,24 In the following, the enhanced work function measurement setup is applied to monitor the caesiation process of a freshly installed polycrystalline Mo surface (SID diagnostic not in operation). One day before the conducted caesiation, a well characterized hydrogen plasma with 10 Pa gas pressure and 250 W RF power is ignited for a couple of hours in the vacuum chamber (electron temperature ∼2 eV and electron density ∼10¹⁶ m⁻³)¹⁸ in order to remove adsorbed impurities from the Mo surface via the plasma–surface interaction and the plasma induced temperature increase close to 500 K. Since the vacuum chamber had been extensively cleaned from previous caesiations, no Cs redistribution could be detected during the plasma phase via laser absorption or optical emission spectroscopy (monitoring of 852.1 nm resonance line).

In Fig. 7(a), the temporal behavior of the surface work function during the conducted caesiation process at about room temperature is depicted. As soon as Cs vapor enters the vacuum chamber, the initial surface work function of about 4.3 eV [see the photocurrent measurement in Fig. 2(c)] is dramatically decreased. After about 45 min (t ≈ 2.2 h), a value of slightly below 2 eV is obtained, with a neutral Cs density of ∼2 × 10¹⁴ m⁻³. The surface work function of slightly below 2 eV is kept constant as long as the Cs density in the gas phase is maintained. By further increasing the Cs evaporation rate until a neutral Cs density in the range of 10¹⁵ m⁻³ is reached, the photoelectric threshold decreases even further until, after some minutes, a photoelectric response is obtained in the near-infrared. This is demonstrated in Fig. 7(b), where an exemplary photocurrent monitoring with photon energies in the range of 2.04–1.45 eV is plotted, with the dark current of 4.8 × 10⁻¹⁰ A already subtracted. Since the absolute photocurrents for ɛ_ph = 1.85–1.45 eV are in the range of 10⁻¹¹–10⁻¹⁰ A, a dark current of about one order of magnitude higher would already prevent photocurrent detection, which again demonstrates the importance of a sufficient SNR. In Fig. 7(b), the respective work function evaluation via a Fowler fit is shown in the inset and yields 1.26 ± 0.01 eV. Since the control parameter A stays reasonably constant during the work function reduction down to the determined value of 1.26 eV, the trustworthiness of the Fowler evaluation can be rated as very good. Considering the finite photon energy resolution (here in the range of 0.03–0.24 eV) and the fact that photon energies below 1.45 eV are not available at the present setup, however, an absolute error of ±0.1 eV should be taken into account.

The generation of Cs coatings with an average work function of 1.25 ± 0.10 eV on a polycrystalline Mo substrate can be reliably reproduced under comparable conditions. Furthermore, the same result is obtained when the Mo substrate is replaced with a stainless steel substrate. In Fig. 8, different caesiation processes comparable to Fig. 7 are compiled for Mo and stainless steel by plotting the measured work function as a function of the applied neutral Cs flux. Since during the initial ramp-up phase of the Cs evaporation a rapid decrease of the work function on the minute scale is given, the error in the determined work function is very likely to be larger than ±0.1 eV due to the necessity of a continuous adjustment of the interference filter set and the 2 min needed for one work function measurement. Therefore, a more pronounced scattering is given in the temporal correlation of the Cs flux and the work function for χ ≳ 2.1 eV than for χ ≲ 2.1 eV. For Cs fluxes in the range of ∼10¹⁷ m⁻² s⁻¹ (n_Cs ∼ 1.5 × 10¹⁵ m⁻³), the ultra-low surface work function is achieved each time, with a background pressure of some 10⁻⁶ mbar. It is observed that such a threshold Cs flux is needed and cannot be compensated by increasing the Cs fluence with a lower Cs density instead (not shown here). When the Cs flux is increased close to 10¹⁸ m⁻² s⁻¹, the determined work function by the Fowler method stays constant within the error bars. However, the photoelectric yield in the near-infrared increases further, and thus, the threshold of 1.25 eV might be an upper limit here. Photon energies below 1.45 eV would be needed in this case to assess the photoelectric threshold more accurately.

The generated work functions are far below the work function of bulk Cs, which is about 1.95–2.14 eV.^29,30 It is well known that work functions below the Cs bulk value can be produced via Cs coatings on metallic substrates in the (sub-)monolayer regime, which has been extensively investigated under ultra-high vacuum (UHV) conditions.^20,28,35,36 Minimum work functions for various metals, such as Cu, Ni, or W, are in the range of 1.4–1.8 eV for optimal Cs coverages, and for Mo and stainless steel, minimum work function values of 1.54–1.61 and 1.52 eV are reported, respectively.^28,35,36 Apart from the fact that no UHV conditions are given at ACCesS, a (sub-)monolayer regime as an explanation for the measured work function of 1.25 eV can, however, be excluded due to the following reasons: first, the measured work function is even lower than the reported work function minima for low Cs coverages. Second, the work function does not increase by increasing the Cs fluence. In the present case, the ultra-low work function can be maintained for several hours if the respective Cs flux is sustained. Third, a color change of the surface is visible by the eye after some time during the caesiation process, which must be due to the growth of a thick adlayer. Hence, the presence of a hydrogenated substrate, which is likely to be the case due to the hydrogen plasma treatment one day before the Cs deposition and which was found to decrease the work function minimum further in the Cs (sub-) monolayer regime,³⁷ is also not likely to be relevant here. Instead, it is suggested that the caesiation process under the given vacuum conditions leads to the growth of an oxidized Cs multilayer on the metal substrates and that the ultra-low work function is thus a bulk effect. It is well known that the coadsorption of Cs with electronegative species can lead to work functions that are lower than what is possible with pure Cs adsorption^36–40 and that thick overlayers with work functions down to 1.0 eV can be attained by the formation of Cs–O compounds.^41–44 In particular, the Cs oxide Cs₂O is usually proposed to provide an ultra-low work function,^41,45,46 but the mixture of Cs with other compounds such as Cs suboxides or peroxides might play a role.^42,47–50 At the given unbaked vacuum system of ACCesS, the residual gas flux is mainly composed of water vapor (indicated by RGA measurements) and is estimated to be of the order of $ΓH2O∼1019$ m⁻² s⁻¹. Thus, considerable codeposition of Cs and H₂O inevitably takes place. Interactions of the evaporated Cs with the residual gases during the caesiation process are obvious due to the decreasing background pressure by a factor of 1.5–1.8 (see Figs. 7 and 8) and are known as the Cs getter effect, i.e., the removal of residual gases due to physisorption and chemisorption. In the RGA monitoring, this effect is visible by a reduction of the H₂O signal (strong hygroscopic characteristic of Cs). Based on the measurements shown in Fig. 8, the surface work function of 1.25 eV is reached with an estimated flux ratio of the order of $ΓCs/ΓH2O∼0.01$⁠. It is suggested that the coadsorption of the alkali metal with residual H₂O strongly promotes the partial and/or full dissociation of H₂O molecules on the metal surface^51–55 and consequently leads to the formation of a thick film of Cs–O(–H) compounds that provides the ultra-low surface work function. Interestingly, a good agreement is given with the reported work function of 1.22 eV in Ref. 41, which was obtained by exposing a Ag substrate alternately to Cs and H₂O in an UHV environment. To the best of the authors’ knowledge, however, no measurements of the formation of a stable ultra-low Cs work function in an unbaked vacuum chamber with a base pressure of 10⁻⁶–10⁻⁵ mbar are reported so far.

At the present setup, work functions of 2 eV and below can be kept constant only if the respective Cs flux is maintained. When a work function of 1.25 eV is reached and the Cs evaporation is subsequently shut down, a rapid degradation occurs due to the residual gas flux onto the surface: $≈2$ eV is measured after about 15 min and 2.75 eV the next day. Furthermore, it should be pointed out that depending on the history of the substrate (e.g., impurity gas adsorptions or plasma exposures), re-caesiations typically lead to lower quantum efficiencies and higher surface work functions even if similar Cs densities and fluences compared to the first caesiation are applied, as can be seen for instance when comparing Fig. 7 with Fig. 3. The interaction of Cs with impurities is thus very complex, and in situ/operando surface analysis techniques are needed for a better understanding of the evolved surface composition and compounds.

The application of the Fowler theory is a well-known method for the photoelectric determination of surface work functions in vacuum. In this work, an easy-to-use setup equipped with an automated data acquisition and analysis system was used to analyze the work function evaluation of metal surfaces exposed to Cs vapor. The irradiation with photon energies from the UV to the near-infrared region is provided by a broadband emitting light source in combination with interference filters. Dedicated measurements demonstrated that a poor threshold sensitivity, which can be the result of a high dark current and/or insufficiently low available photon energies, can lead to severe errors of the evaluated work function due to the restriction of photoelectric yield data to photon energies too far away from the threshold. In the presented case of caesiated surfaces, an overestimation is the result, which increases the further the lowest irradiated photon energy is away from the threshold. The optimization of the photoelectric threshold sensitivity is thus of utmost importance to accurately access the photoelectric yield in the region close to the work function (≲1 eV), in which the Fowler theory can be reliably applied.

The optimized work function measurement setup used in this work provides a photon energy resolution of mostly 0.1–0.3 eV in the range of 5.04–1.45 eV, a dark current of the order of 10⁻¹¹–10⁻¹⁰ A, and a photocurrent detection down to the pA range, which enables access to the measurement of work functions in the ultra-low region. With this, it is now unveiled that surface work functions of 1.25 ± 0.10 eV can be reproducibly generated via caesiation of metallic surfaces (polycrystalline Mo and stainless steel) in an unbaked vacuum environment of 10⁻⁶–10⁻⁵ mbar. Considering the moderate vacuum conditions, such a low work function is remarkable. As long as the Cs evaporation with a neutral Cs density of the order of 10¹⁵ m⁻³ is sustained, the work function is temporally stable. It is assumed that this is the result of Cs coadsorption with the dominant residual water molecules, resulting in the growth of an oxidized Cs multi-adlayer on the metallic substrate. Re-caesiations of degraded Cs coatings lead to lower quantum efficiencies and higher surface work functions even for comparable Cs densities and fluences. In view of negative hydrogen ion sources, which often operate under limited vacuum conditions and in which Cs is injected (e.g., large-scale sources that are developed for upcoming fusion devices, such as ITER^56,57), the generation of work functions in the ultra-low range on the converter surface would be highly beneficial for the enhancement of the surface production of negative ions. In a next step, the influence of plasma irradiation on such surfaces will, thus, be studied with the sensitivity enhanced work function setup at ACCesS.

This work has been carried out within the framework of the EUROfusion Consortium and has received funding from the Euratomresearch and Training Programme 2014-2018 and 2019-2020 under grant agreement No. 633053. The views and opinions expressed herein do not necessarily reflect those of the European Commission. This work was carried out within the framework of the EUROfusion Consortium, funded by the European Union via the Euratom Research and Training Programme (Grant Agreement Nos. 101052200 and 633053—EUROfusion). Views and opinions expressed are, however, those of the authors only and do not necessarily reflect those of the European Union or the European Commission. Neither the European Union nor the European Commission can be held responsible for them.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.