February - April 2025
Nowadays, ultra-high vacuum (UHV) components such as CF pipes, CF windows, valves, copper gaskets, etc. are quite easy to purchase at reasonable prices from a number of mainland Chinese and Taiwanese companies. However, evacuating the system from atmospheric to UHV pressures without spending a lot of money remains a challenge. Not many kinds of pumps are capable of reaching UHV, and the sputter ion pump is the main technology employed there. Such pumps are not able to start from atmospheric pressure, and they are commonly roughed with a turbomolecular pump which is itself backed by a mechanical pump, often of an oil-free dry type to avoid contamination of the vacuum chamber. Scroll pumps can be employed, or even diaphragm pumps when the turbomolecular pump is equipped with a Holweck drag stage (most modern ones are). Diaphragm pumps are less expensive, but in any case a complete small turbomolecular pumping cart that can start from atmosphere usually costs just south of USD 10k. And it is something that can be instantly ruined should it be opened to atmosphere at the wrong time.
But it is interesting to note that ion pumps are capable of starting at pressures as high as 10-2 mbar for triode-type pumps, which is within the range attainable with a two-stage rotary vane pump alone. So it may sound tempting to save the high cost, hassle, and safety issues associated with turbomolecular pumps, and rough the ion pump with a low-cost mechanical pump only. However this is not advisable for two reasons. Obviously, starting at a high pressure increases the wear on the ion pump. Secondly, reaching 10-2 mbar and below with a rotary vane pump alone can take a long time. This is more than a mere incovenience: below 10-1 mbar, a small fraction of the mechanical pump oil starts to boil off and contaminates the vacuum chamber, and the longer the vacuum chamber is exposed to a rotary vane pump under such conditions, the more contamination can build up.
An intermediate solution is to rough using a mechanical pump followed by a small diffusion pump. Such a system can reach pressures well below 10-3 mbar in a matter of minutes, after which the UHV section of the system is immediately sealed off and then the roughing section vented in order to mitigate the oil contamination problem.
To explore this idea, I have used a low-cost 2XZ-0.5 mechanical pump for the first stage, which is manufactured by a number of Chinese vendors. That pump looks like something that may have come right out of a roadside truck repair shop, but despite its appearance and its price it actually works very well and it is easy to repair when it does not. After intermittently using one for about 10 years, the only serious issue it had developed was a leaky TC oil seal. A replacement seal was easily purchased for a few dollars and while its replacement required a near-total disassembly of the pump (including removal of one set of rotary vanes), this was done in a matter of hours with just a screwdriver and a few Allen keys. A better mechanic than me would probably get it done in 30 minutes or less.
To mitigate the oil contamination issue further, the inlet of the pump was fitted with a 20cm long KF25 tube filled with copper wool (available from U.S. gun shops and easily mailed internationally) which was kept in place by KF25 gaskets containing a metallic mesh inside their centering rings. I have not investigated how much of a difference this actually makes, this is simply based on advice from several electron microscopist publications.
The pump was also equipped with an anti-suckback valve, which prevents the oil from being drawn into the entire evacuated chamber when the pump is turned off and making a horrible mess. This, on the other hand, is unfortunately something that I have direct experience with.
The pump was first filled with Chinese "VALUE" oil from AC service shops. The result was pretty disastrous with a pitiful ultimate pressure above 10-1 mbar, which actually increased with time as the pump warmed up and the oil's vapor pressure increased. A low-cost oil that turned out to work well was "Diffusion Pump Oil No. 3", from the Dalian Petrochemical Company. I cannot recommend its use as an actual diffusion pump oil - it does pull a vacuum, but you will end up with tarry, sticky deposits inside and all around your diffusion pump which will take hours of very boring work should you wish to clean them up. Additionally, it could potentially ignite or explode if the chamber is abruptly vented to atmosphere with the diffusion pump hot. But as a mechanical pump oil, it works just fine, as indicated on the side of its container.
The second stage is a Chinese clone of the Edwards EO50 air-cooled diffusion pump, filled with cleaner, safer, and better-performing DC705 oil. This is where things get a bit more difficult. First of all, that diffusion pump is janky, and expensive for what it is ($400). The inlet had an intermittent leak due to a welder as competent as a McKinsey consultant, which, after plenty of tedious troubleshooting, was successfully patched with copious amounts of Apiezon Wax W. Then due to insufficient ventilation and/or poor construction of the heatsink, this pump overheats and stops working after about half an hour of use (probably filling much of the vacuum system with oil vapors as well). The fan that it came with was recycled e-waste "made in West Germany", with mains wires that had been cut next to the motor windings, and haphazardly refurbished by the pump's manufacturer - eventually those short-circuited against the (earthed) metal casing. Replacing it with a more powerful fan might help, but as indicated above I would only use this pump for a few minutes at a time anyway.
There are other EO50-style pumps made in China that look better on the pictures, but at higher prices ($800+), which may be worth it if they are better built by a more serious manufacturer who doesn't recklessly cut corners and/or doesn't know what they're doing. Or you might be better off anyway with a larger water-cooled industrial-style diffusion pump, made of cast iron, which unlike pompous "research-grade" garbage such as the EO50 and its clones, actually has to perform at a competitive price - the smallest models in that category are cheaper than $400 (but they still take 1kW of heating power).
Obtaining silicone diffusion pump oil is also a little adventure in 2025. Most U.S. suppliers for this kind of product won't ship it internationally anymore - isn't it obvious that if you can't pull a good vacuum, then America is great again? If you are not in the U.S., you probably shouldn't even bother asking them and directly look for local stock instead. There might also be the option of using the "Efele" Russian copy of DC704/DC705, but it seems pretty overpriced, and additionally, in many places the mere act of receiving a package from Russia may earn you an entry in one of the defamatory lists popular with certain businesses, such as KYC-obsessed bank scum who are looking for any opportunity to waste even more of your time with their vapid and vexatious rubbish.
The pressure after these two stages of rouging pumps is monitored using a ASAIR AGP3000 gauge, which are convenient to use, high-quality, and reasonably priced. After the diffusion pump has started, the gauge will max out and indicate "1.00e-03" mbar, which is sufficient indication that the ion pump can be ignited. For venting this section of the system, a Swagelok needle valve is used - abrupt venting such as by disconnecting a KF flange should be avoided as it produces a mist of diffusion pump oil, which also occurs when the diffusion pump is at room temperature. Gentle venting over a dozen seconds through the needle valve prevents this issue.
The roughing pumps are then connected to the UHV chamber with the ion pump, through a relatively inexpensive Taiwan-made all-metal valve from Highlight Tech, using a KF40 hydroformed bellows hose which was the largest diameter that could be conveniently used (as the inner pipe diameter is similar to the CF35 of the all-metal valve, and CF35/KF40 adapters are inexpensive and widely available, which is not the case for e.g. CF35/KF50). Using KF flanges on the roughing section of the system does not make a difference in terms of achievable vacuum pressures since they are not the limiting factor there, and they are cheaper and a lot more easy to work with than CF.
The ion pump is by far the priciest component in the entire setup. There aren't many manufacturers of ion pumps, they are systematically expensive, and most second-hand units on offer are damaged, worn-out, and/or contaminated (often with the seller not being clear about the real condition of the item). I have taken the magnet from an Agilent VacIon 25 911-5030 (manual) in typical eBay condition, and mounted it on a new replacement pump element from a reputable supplier, at a total cost of about $1,400. Apparently, ion pump elements in eBay condition can sometimes be refurbished using nitric acid and extended bakeouts at high temperatures under vacuum and without magnet, but I have yet to try that.
To evacuate the system, open the all-metal valve, close the Swagelok venting valve, and turn on the mechanical pump, the diffusion pump heater, and the diffusion pump fan. The pressure reading on the gauge will drop (for a small vacuum chamber) within a minute or two into the few-10-2 mbar region. After a few more minutes, the diffusion pump will reach its operating temperature and cause a further pressure drop, in a matter of seconds, below 10-3 mbar, which will max out the AGP3000 gauge.
It is time to start the ion pump. It is first connected to a low-cost 1000V 300mA lab power supply made by a number of Chinese suppliers such as Rainworm and HSPY. Beware that despite their innocent appearance, questionable choice of output terminals, and the total absence of warning labels, those units are capable of delivering a LETHAL current and you must be VERY careful when using one of these. Set the maximum voltage (1000V) and a ˜50mA current limit, then apply the high voltage to the pump with the correct polarity for your pump type. If the current reads zero on the power supply, sadly it is not because the diffusion pump pulled an excellent vacuum, but because the Penning discharge failed to ignite inside the ion pump. Hit the pump with the handle of a screwdriver or heat it with a blowtorch (away from the magnet and high-voltage feedthrough) until you notice a sudden increase in current.
At the beginning, the power supply will enter constant-current mode, with the voltage dropping to about 450V. As the vacuum improves, the voltage will increase, and then the current will drop with 1000V being continuously applied to the pump. After the ion pump current has dropped to about 10mA, it is time to check if the roughing section can be valved off and shut down. Restrict the flow between the UHV chamber and the roughing section by gently closing the all-metal valve with your fingers (the vacuum chamber should of course have been earthed to eliminate the risk of electric shock from the ion pump supply), and look at the ion pump current. If the current exhibits a large increase, re-open the valve completely and try again after a minute or so. If it doesn't increase much or even decreases when the valve restricts the flow, then proceed to closing the all-metal valve completely using a wrench. Shut down the diffusion pump heater, leaving the fan on. When no more boiling noises can be heard from the diffusion pump, shut down the mechanical pump and vent the roughing section gently through the Swagelok needle valve. The KF40 roughing hose can then be disconnected from the UHV section if desired.
Leave the ion pump on the 1000V lab power supply until the current drops well below 1mA. Then it can be transferred to a more suitable higher-voltage, lower-current supply. One that works well and costs very little money is the CX-50 1-10kV 5mA model from Chengdu Chuangyu Xinjie Technology. This is a clever hack that repurposes a cheap mobile phone charger circuit, with a transformer that has two secondaries, a low-voltage one used for voltage regulation by feeding back to the mobile phone charger IC, and the main high-voltage one used to drive the output. This provides an isolated, adjustable and regulated high-voltage output at extremely low cost. Its main disadvantage is it produces copious amounts of EMI - it is definitely a cheap design where they are cutting corners - so it needs to be kept away from sensitive electronics. Set it to 5000V and connect it to the ion pump with a low-side 10 kiloohm current sense resistor, which can be used with a multimeter to estimate the vacuum pressure from the pump current. Add a 100nF capacitor and perhaps even a gas discharge tube in parallel with the current sense resistor to avoid killing the multimeter and/or the resistor from high current transients, which may happen with ion pumps. I destroyed two multimeters while using them to measure ion pump currents, before I figured out the correct technique. Adding a 100Mohm high-voltage resistor in parallel with the CX-50's output may also help with voltage regulation and respecting the manufacturer's instruction that the power supply should not be run without a load.
After a few weeks of pumping, the pressure as estimated from the ion pump current was in the 10-9 mbar range, which is sufficient for many modern experiments in atomic physics. Of course, with a bake-out the pump-down time can be drastically shortened and even better pressures achieved; heater strips at low voltages (24-48V) can be used to avoid the hazards of running mains voltage all around the vacuum chamber. It is also interesting to note that shining a 365nm LED lamp with many watts of optical output (wear safety glasses and avoid skin exposure) through the vacuum windows produces a sharp rise in pressure in an unbaked system. Such powerful UV sources are amazingly inexpensive and compact nowadays. While it is probably not as effective as a bake-out, it is very easy to do and may be another tool to reduce pump-down time.