Photon emission spectrum of ion pumps

Precision experiments require a detailed understanding of every aspect of the experimental setup. The advanced Laser Interferometer Gravitational-Wave Observatory (aLIGO) experiment is built to measure relative variations of length on the order of 10⁻¹⁸ as gravitational waves pass by.¹ In order to accomplish this, aLIGO uses a Michelson interferometer with two 4 km Fabry–Pèrot arms, all operated under ultra-high vacuum conditions. The mirrors are suspended and play the role of free-falling test masses that are as much isolated from their surroundings as possible. Every experimental apparatus operating near the mirrors must not interact with them in any way, lest it induces noise in the measurement of the arm-distance.

The ultra-high vacuum conditions of aLIGO are generated by a system of ion pumps located at several positions along the arms and around the mirrors. The mirrors act as freely falling test masses that aid in the detection of gravitational waves. These mirrors are made of insulating material with no conducting coating on them, meaning that any charge induced on them will diffuse very slowly around the mirror and potentially cause noise through its interaction with the electrostatic mirror actuators. Charge accumulation on the mirrors has been an issue for the operation of aLIGO, and understanding its origin is a challenge. In one scenario, a charge can be induced via the photoelectric effect when UV photons with energy higher than the work function of the mirror fall onto its surface. Another way to induce a charge on the mirror is if UV photons cause electron photoemission on the elements surrounding it. Given that an ion pump’s operation involves electrons and heavier ions impinging on its walls traveling across an electrostatic potential of a few kV,² it is plausible that ion pumps emit UV photons with high enough energy and flux to cause charge noise for the aLIGO experiment. While ion pumps do not have a direct line-of-sight to the mirrors, it is still possible that UV photons reach them after one or two bounces off of the walls. The LIGO detectors have in fact moved ion pumps further away from the test-mass mirrors, and a chevron baffle has been installed in front of the ion pump in an effort to reduce charging.³ 

In this paper, we present the results from a UV emission measurement of an Agilent VacIon Plus 2500 l/s ion pump, which, to the best of the authors’ knowledge, has not been reported before. The technique used is an inexpensive and fast way of measuring the spectrum. In Sec. II, we discuss the experimental details of the measurement, and in Sec. III, we present the results.

The ion pump tested is a 2500 l/s Agilent VacIon Plus with two conflat ports on two sides, one 16.5 in. and one 8 in. diameter. The measuring apparatus was attached to the 16.5 inch side, while the other side was used for initial pumping and as an entry point from which Argon was introduced in the pump for one of the tests. The main part of the experimental setup is shown in Fig. 1. The measuring apparatus consists of a photo detector and a filter wheel, on which filters of different frequency bands are mounted. Figure 2 shows a photograph of the filter wheel, as it is mounted on the flange. The silicon detector and seven filters, as well as three empty and two blocking positions, can be seen. The wheel was rotated from the air-side via a rotary feedthrough, allowing the placement of any one filter in front of the detector. In that way, we can select to measure the dark DC current of the detector or the current due to a particular frequency band defined by a filter.

The filters used were three premium hard-coated longpass filters with edges at 500, 750, and 1000 nm (2.48, 1.65, and 1.24 eV, respectively) from Thorlabs, two UV filters from eSource Optics, 130FBB and 172FNB, broadband and narrowband with centroids at 130 nm (9.5 eV) and 172 nm (7.2 eV), respectively, and two thin aluminum foils of 70 and 345 nm thickness. The transmission of all the filters except for filter 130FBB is shown in Fig. 3. Filter 130FBB was eventually not used in the analysis, as it was too broadband to give any information on the spectrum, All filters were 1 in. diameter and were mounted on a filter wheel with 0.8 in. diameter clearance holes. The distance between the filter and the detector was 0.4 in., resulting in a detection solid angle of a full 45° cone.

The main challenge of the experiment was to ensure that the signal we are seeing from the detector was not coming from the optical band since ion pumps emit a lot of photons above 300 nm. Both the aluminum filters and the 172FNB UV filter have a very low transmission in the optical band, but any deviation from factory specifications or any micro-holes on the foil in the case of the aluminum filters would overwhelm the measurement of UV emission. For this reason, the filters were tested with a Manson x-ray Fluorescence source.⁴ Electrons from a filament are accelerated to a few keV and strike on an anode made of some given material depending on the needed x-ray lines. A carbon anode was used, in this case, which gives one line at 280eV. Due to the high temperature of the electron-emitting filament, the source also emits a large flux of photons in the visible range. By placing the filter between the silicon detector and the Manson source, we could determine its transmission of visible photons and compare it with the expected from the manufacturer. Both the aluminum filters and the 172FNB filter had negligible transmission in the visible.

The thicknesses of the aluminum filters were determined using the same x-ray source described above, by measuring the transmission of the 280 eV photons from the carbon anode through the filters. The thicknesses were found to be 70 and 345 nm. Given these thicknesses, we determined the transmission of photons through the aluminum foils as a function of photon energy using the particle physics simulation software Geant4.⁵ The results for the transmission of the two foils are shown in Fig. 3.

Finally, the effect of variations in the transmission due to the angle of incidence was taken into account for the 172FNB and aluminum filters and was found to be 10% or less. In the case of the optical filters, no corrections were made, since Thorlabs could not provide any information.

The photodetector was an XUV-100 detector from OSI Optoelectronics with an active area of 1 cm². It can be seen in Fig. 2 on the top left of the filter wheel. The responsivity of the detector up to 6.5 eV and from 50 eV and beyond was given by the manufacturer and is shown with a solid black line in Fig. 4. However, in the region between 6.5 and 50 eV, there were no reliable data, and it was calculated assuming a dead layer in front of the photodetector.⁶ The transmission through this dead layer determines the responsivity of the detector, assuming that for 100% transmission, the responsivity is 0.272 A/W, given by the manufacturer. Then, the thickness of this dead layer was chosen so that it matches the given responsivity of the detector at 6.5 eV. The result was a dead layer of 4.2 nm, and the corresponding responsivity is shown in the figure with a red-dashed line. The calculated responsivity follows reasonably well the measured one given by the manufacturer.

The DC current induced on the photodetector by the photons passed through a transimpedance amplifier and a low-pass filter before it was read as a voltage on the oscilloscope. The resulting sensitivity of the setup was ∼5 fA, which was sufficient for the measurement of the flux through our filters for most of our tests.

The measured signal from a given filter determines the integrated photon flux weighed by the transmission. Thus, the 1000 nm long-pass filter determines the photon flux approximately between 1.1 and 1.24 eV, where the lower bound comes from the responsivity threshold of the detector. Within that range, the flux is assumed constant. Then, the 750 nm long-pass filter determines the flux between 1.1 and 1.65 eV. By subtracting the contribution of the flux between 1.1 and 1.24 eV, we get the flux between 1.24 and 1.65 eV. This procedure is repeated for all the filters, in order to obtain spectrum energy-bins as fine as the setup allows.

While the three long-pass filters and the UV filter all have relatively narrow bandwidths, the two aluminum filters do not. They are opaque below 8–9 eV but have a complicated transmission function of up to 1 keV. However, using the ratio of the signals from these two filters, it was possible to infer more information on the photon spectrum. The ratio of the detector signal of the thin filter over the thick one for all the measurements was ∼2. Using the known transmission for the two aluminum filters (see Fig. 3), we can constrain the photon emission in the range between 10 and 80 eV. For this, we need to assume that the emission spectrum has a simple and smooth energy dependence.

The measured photon energy spectrum is shown as a histogram in Fig. 5, for different voltage settings of the ion pump, namely, 3, 5, and 7 kV. The figure shows that the photon flux from the ion pump is decreasing as the energy increases above 5 eV. While a full simulation of the emission is beyond the scope of this work, we were able to model the photon emission of our ion pump, with reasonably good agreement. The shaded region in the figure shows the modeled spectrum of the ion pump at 7 kV, assuming that the only photon production mechanism is bremsstrahlung radiation emitted by electrons impinging on the titanium-coated anodes and that each photon either reflects once off the walls of the ion pump (lower limit) or does not reflect at all (upper limit) before it comes out of the flange. For the bremsstrahlung calculation, we used the cross-sectional tables from Powell.⁷ The reasonable agreement with the measurements suggests that indeed the emission possibly comes from bremsstrahlung photons that might have reflected off the walls and that it falls off rapidly above a few tens of eV. The model used here only considers the bremsstrahlung production cross section, the single reflection probability of a photon from the walls of the pump, and the current of the ion pump. The model does not include possible effects of the geometry or other interactions. Figure 5 also shows the calculated spectra for singly reflected photons for the other two ion-pump voltages, 3 and 5 kV.

A quantity of particular interest for aLIGO and the charging of its test-mass mirrors is the number of photons emitted by the pump above the mirror’s work function. The mirror substrate is SiO₂, which has a work function close to 5 eV,⁸ and the mirror coating consists of multilayers of SiO₂ and Ta₂O₅, the latter having a 4.5 eV work function.⁹ These are the photons that will have sufficient energy to induce a charge on the test-mass mirror. We can obtain the number of charge-inducing photons by integrating the number of photons in Fig. 5 above 5 eV. The results are shown as solid markers in Fig. 6, where the emission integral is plotted as a function of ion pump current and for different voltages of operation. The companion faded symbols on the plot show the modeled integrated flux of singly reflected photons for comparison. The modeled flux was multiplied by 1.8 to match the measured flux. Given the simplicity of our model, we consider the agreement with the data adequate. Other than the overall normalization factor, the model matches very well the dependence of the integrated flux to the voltage setting of the ion pump, which supports the assumed photon production scenario.

The above measurements were all taken after pumping down from air. To test the dependence of the energy spectrum on the gas composition in the pump, the spectrum was measured also after introducing argon. No difference was observed in the two measurements, supporting the idea that the photons are mainly coming from the electrons impinging on the walls of the ion pump and are not related to the pumped species.

The photon emission spectrum of a 2500 l/s Agilent VacIon Plus ion pump was measured for the first time. The emission is decreasing as a function of energy and does not extend much above 80 eV. As expected, the emission is proportional to the ion-pump current and also depends on the voltage setting. The photon spectrum is consistent with bremsstrahlung radiation. Finally, the spectrum does not depend on the gas composition that is pumped. Experiments that might be sensitive to UV photons should account for the presence of ion pumps in the vacuum system. Since a significant fraction of the photons seems to be emitted below 7 eV, future studies can use commercially available spectrometers to obtain a more detailed emission spectrum at these lower energies.

This work was supported by the National Science Foundation under Grant No. 0757058.

The authors have no conflicts to disclose.

Antonios Kontos: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Visualization (equal); Writing – original draft (equal). Rainer Weiss: Methodology (equal); Project administration (equal); Supervision (equal); Writing – review & editing (equal).

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