It’s Friday! Let’s talk through Orbital Arc’s progress this week, with a bit of backtracking about molybdenum. Also, strap in folks – this one gets technical.
So, this week I have focused on two things primarily: 1. Evaluating potential ion fuel options. And 2. Figuring out how the heck to deposit molybdenum on stuff.
The impetus for both these activities was the same: a few weeks ago, I started work on TRL 4 prototype ionizers in the Rice University Nanofab Cleanroom, and it took me about two weeks to realize the current plan was moving too slowly; ever since, I’ve been trying to speed it up, which is hard, and requires some backstory to really understand.
Backstory: If you know anything about field effect ionization, you know that you basically need a very sharp tip to do it – the tip provides a multiplying effect, called a field enhancement factor, on the electric field used in the device, which, combined with a very small gap between your tip and your ground electrode, is how you go from a 300 V power supply to the 10 GV/m field you need to yank electrons off of passing molecules.
The tips have several requirements: they must be sturdy enough to survive the vibrations of a space launch, they must be electrically conductive, they must be very heat resistant, and, above all, they must be very, very sharp. More specifically, you need to have a very high aspect ratio between the radius of curvature of the tip and the length of the tip.
This requirement set makes things like carbon nanotubes very promising as field electron emitters or field effect ionizers; those of you who have been following my progress for a while may remember that several of the earliest design iterations were focused on using carbon nanotubes or carbon nanofibers. They are very heat resistant, very electrically conductive, very strong, and best of all they can be grown free standing at aspect ratios of 50:1, or in combined fibers that taper to final aspect ratios of 500:1, with tip radii as small as 4 nm. For a single tip, they are basically perfect,
However, Orbital Arc doesn’t just want one tip, we want several million tips in an array, all performing at roughly equal levels. This adds an additional requirement that the fabrication of the tips be very highly uniform – if one tip is 50 nm taller than another, or one tip has a tip radius 20 nm bigger than another, that can affect the performance of the device, because it can increase or decrease the field enhancement factor. Worse, tips at different heights will interfere with each other, causing a field screening effect. The result, which has been played out in dozens of different research papers, is that nearly all the activity comes from just a few tips in the array, which wear out very quickly, while most of the array does basically nothing. Performance degrades over time, and there’s a ton of wasted space on the device.
This failure mode (we at Orbital Arc believe) is why FEEP thrusters never lived up to their potential – They use liquid metal or ionic liquid fuel fed to the end of a volcano-shaped tip through a capillary, and the field deforms the liquid to make an atomically sharp temporary tip (called a Taylor Cone) which emits ions. It works, but if any of the volcano-capillary-tips is a bit taller than the others, all the ion current will come from that one, because it will form Taylor Cones more easily at a given field, and those cones will stretch taller still, causing them to screen the field from the other tips and prevent those tips from emitting. Emission current density becomes a function of the tallest tips and their capacity, not a function of the array, and performance is very limited as a result.
Orbital Arc’s solution is to deliver the fuel in gas phase; no Taylor cones, which means we lose the automatic atomically-sharp tip formation process, but we get no amplification of screening effects beyond the baseline from tip fabrication variance. And, gas phase fuel delivery means that if one tip is more active, it just ionizes the fuel that reaches it more quickly; it doesn’t produce more current than any of the other tips, because current only flows when gas does, unlike when fuel is delivered by capillary action which feeds fuel to tips more quickly the more quickly the tip consumes it.
So, we started out trying to make nanotube-based tips. But, as our research continued, we concluded that process control of the growth of sparse vertically aligned carbon nanotube arrays is just not good enough, even with the most modern techniques. Nanotubes grow to widely varying heights, and widely varying radii, and even the best approaches in the world can’t seem get the variances down below ~20%, when we want them to be closer to ~1-2%. Every fuel molecule that passes an inactive tip and isn’t ionized is wasted fuel, and also will increase the erosion rate for the whole thruster (a topic for another time), so we need ALL the tips to produce ions. So, we switched up the design, and started working on fabricating our tips from molybdenum.
Molybdenum is a well-proven field electron emission material – the first field emission arrays fabricated by Charles A. Spindt back in the 1970s used molybdenum tips. Other options include tungsten or silicon, but tungsten is even harder to work than moly, and silicon is both less electrically conductive and less heat resistant, which means it will have more resistive heating, and fail sooner because of it. So moly seemed like a good option; there were also a variety of processing recipes available in literature that could be readily adapted to the project. We made the decision to switch back in May, and have been working on a moly-based design ever since.
Moly, though, is a harsh mistress.
This is mostly because we need a lot of it; from conversations with people doing normal nanofabrication things, a normal amount of molybdenum to deposit on something would be, say 200 nm. In fact, at the Rice Nanofab Cleanroom, that’s the max you can deposit in one session on the best tool available for moly deposition (a DC magnetron sputtering system). But, our design called for 10 microns, which would mean 50 separate depositions in the magnetron tool, which would take 1-2 hours each, and could only be done maybe once per week, because the molybdenum targets in the tool will run out if used too much (which will damage the tool if not prevented), and the lab only replaces the targets every two weeks. That would be….a year. And that’s just to get one layer of molybdenum on one wafer, assuming I didn’t mess anything up in the process and need to scrap it and start over. And, moly isn’t the only layer I need; the next layer actually was designed to be 14 microns thick. Not to mention the cost – $175/hr, plus $180 day use fee every day I enter the cleanroom. Call it $400 on average per 200 nm, or $2/nm, which is $20k for that one layer of moly. Not tenable.
Therefore, we decided to use a different tool, an e-beam evaporator. Where the sputtering system ignites a plasma above a target and sputters off atoms of the target material with the plasma impacts (very similar to how gridded ion thrusters erode over time from plasma erosion, just intentional in this case), the e-beam heats the material up by electron bombardment, until it melts, and ultimately evaporates. The vapor then fills the vacuum chamber in which the process is run, and condenses on all the internal surfaces, which are cooler than the molten source.
With moly, this is challenging – you need to get the metal up above its melting point to get any meaningful evaporation, which, with moly, means getting above 2,600 Celsius. That’s roughly half the temperature of the surface of the sun. It glows white hot, and you need to look at it through welder’s glass to tell if it is melting. As it starts to melt, it spits specks of molten metal out into the chamber, and spits more any time you increase power or move the electron beam around, which you need to do constantly because you are trying to consume the moly target evenly across its surface to avoid drilling through it and destroying your crucible and possibly the machine.
This process is harder, more manual, and tends to heat the substrate much more than the magnetron sputtering system would, and also be less stable, leading to more stress in the deposited film, but for my purposes it has several big advantages: First, I can open the e-beam vacuum chamber whenever I want to add more moly to my target crucible, which means that the limit on the deposition thickness in one session is however much moly I can fit in the crucible. Second, the cost is $50/hr, instead of $175. Third, the deposition rate can be faster; 1 angstrom per second is normal; I’ve gone up to 3 angstroms per second on my first deposition (which did fail, so maybe that was too fast). The magnetron can only do about 0.6 angstrom/sec.
The e-beam, though, also has some downsides. First, because it is highly flexible, it is one of the most popular tools in the lab. As a result, they only let you book 4 hours at a time. Second, because every usage session involves opening the vacuum chamber, you have to pump it back down every time, which takes the first 1.5 hours of your 4 hour session. At 1 angstrom/sec, the remaining 2.5 hours will yield 9,000 angstroms, or 900 nm, assuming you can keep the deposition rate constant, so 10 microns is still 11 separate deposition sessions, each of which costs $380 including day use fees, so the layer still costs $4,180, and still takes at least three weeks to do (since you can only book 4 sessions per week, again because the tool is popular).
This, however, has a workaround. See, the lab is open 24/7. So, if I book 4 hours from 8 PM to midnight one day, and 4 more hours from midnight to 4 AM the following day, I can get do two sessions back to back without breaking vacuum. Further, you only get charged the day use fee when scan your ID for entry through the lab door; if I stay in the lab across midnight, I don’t get charged the fee for the second session. This means I get 6.5 hours of deposition time, 23,400 angstroms, or 2,340 nm, for a total of $580, bringing the cost of the whole layer down to $2,700. It would still take 3 weeks to do this 4 times and have 1 last session to finish up, but that’s a cost I can live with, especially since that’s cost per wafer, and each wafer can yield 20+ ionizer prototypes.
That’s the theory, anyway. In practice, my first four deposition attempts went like this:
- Deposited 800 nm at about 3 angstroms per second, but adhesion was poor and it built up too much tensile stress, causing the whole moly layer to fracture and curl and turn to molybdenum dust. Scrap that substrate.
- Loaded my crucible incorrectly, nearly breaking the machine. Needed the staff to recalibrate it.
- Loaded my crucible correctly, but did not fill it enough, so the e-beam drilled right through the moly target and almost destroyed the crucible. Had I not caught it, this too could have broken the machine, in a much worse way than #2. Still got ~50 nm out of it.
- Got it to work, but freaked out because chamber heating caused previously deposited layers on the viewing mirror to delaminate and peel, which made it look like the mirror had shattered in the viewing window. Pulled the plug early.
The above was the near-miss to destructive failure from attempt 3. Note the black spot on the crucible bottom. 10 more minutes and I’d probably have drilled right through it.
In total, I now have about 530 nm deposited on one substrate. I don’t know how much stress there is in that 530 nm, or how good its adhesion is; it should be better than the first run because I did the adhesion layer underneath it in a more effective way, but I can’t be certain. So, I don’t trust it yet, and I don’t want to take chances with it, and potentially waste the work. Therefore, I am not going to deposit more moly until I have a chance to relieve the internal stress.
I can relieve the internal stress by annealing the substrate for an hour or so in a high temperature furnace under vacuum. Conveniently, the Rice Nanofab Cleanroom has a furnace available for that purpose. Inconveniently, that furnace has been down for the last four weeks – the temperature has been reading higher than the actual temp, and it needs recalibration. So, I can’t proceed with more deposition, and have been stalled for about three weeks at this point on the fabrication side. Nothing to be done but wait for them to fix it.
But I hate doing nothing. So, instead, I started looking for alternative design and fabrication pathways. And I found some interesting things.
First thing I found is a new design architecture that will be easier to fabricate and require less alignment precision between layers, which can also be completed without any chemical mechanical polishing (CMP – it is the best way to get something super, super flat, but it will likely cost me $4000 every time I do it, and requires shipping the super-fragile substrates-in-process to an outside vendor, with a multi-week turnaround time; painfully slow, expensive, and scary). We will need to change a couple aspects of the process, but the general concept should work almost perfectly. Thank you internet.
Second, I started thinking about thinner layers, shorter tips, and sharper points. If I need a 10 micron tall tip, then I need a 14 micron tall insulator layer, and both those layers will take FOREVER to deposit on the tools at the Rice lab. If I only needed a 5 micron tip, then I would only need a 9 micron insulator layer, and I can reduce the fabrication timeline by about 40%. Seems worth doing if we can get to a design that works.
The thing is, if you have a shorter tip, you need a sharper tip, and the field enhancement factor gets much more sensitive to the tip radius (on a per-nanometer basis). If you can’t get the tip arbitrarily sharp (which we can’t, because of reasons that will stay proprietary), you need a higher voltage to compensate. The result is that when I cut the tip height in half, I needed to basically rebuild our analytical model from scratch – instead of being in the realm of “field is plenty strong, no worries” we ended up in the realm of “you need this math done to significantly higher precision than before.” This took about a week to do (much faster than the first version, which took me a year, because I had no clue what math I was even looking for).
The result is that we now have a much more reliable and flexible model that we can use on a much wider variety of parameters. Also, yeah, it turns out we can get it to work with a shorter tip. But, it comes at a cost: voltage needs to increase about 3x. The materials involved *should* be able to handle that, and from a thruster performance standpoint, it just equates to higher ISP, at a cost of somewhat higher power. Not a terrible thing, though the original ISP was pretty close to the sweet spot for cislunar space, so we probably do want to use longer tips eventually. But, for a prototype, not terrible at all.
Third, I found a research paper where a guy documented a process for depositing 5 microns of moly on a substrate in 30 minutes. With very low residual stress. In a magnetron sputtering system. Now, the system he used was not too similar to the one at Rice; they really are limited to a couple hundred nanometers at a time, and low deposition rates (I asked anyway, but I figured they would tell me what they told me). But, I discovered today that Texas A&M has a sputtering system that is much more analogous to the one used by this researcher….. Drive 90 minutes to save myself 3 weeks of work and a couple thousand dollars? Yes, please and thank you. Next week I’ll be trying to set this up first thing.
And fourth, lastly, I started compiling a database of potentially compatible fuels for our thruster. Sulfur Hexafluoride (SF6) is still my favorite, but after a whole lot of research, I have never yet found any record of anyone trying to ionize SF6 by field effect ionization. So, I don’t know how it will behave when I try. The simple answer that “field effect ionization typically doesn’t break apart molecules” is great, unless SF6 is atypical. I can’t leave all my eggs in one basket, so I’m making a list of other potential options, and gathering all the data needed to input them in the analytical model.
SF6 has this great combination of being cheap, super safe, and heavy, which is hard to duplicate, but after assessing about 40 candidate fuel gases this week, I’ve found at least 5 that are good prospects, and with 4 of them, I was able to verify that they can be ionized without decomposing. None of the new options are so safe you would inhale them voluntarily to make your voice deeper. But they check the other boxes, and are all handled regularly and safely in industrial use. (Side note: This process included some interesting history lessons in how many poisonous gases we used to regularly emit into the environment; one of the unsuccessful options got the axe when I found out it was literally a chemical warfare agent used in WWI). Three of the good (non chemical warfare agent) options are actually easier to ionize than SF6, which partially mitigates the voltage increase issue in the smaller tip design.
So, that’s what we do when we have 3 weeks of non-lab time because a furnace is broken.
But, that’s not all that Orbital Arc has been doing. The two outside projects, at Desert Works Propulsion and at Oak Ridge National Lab are also continuing apace. First, at Oak Ridge, Dr. Nick Lavrik has been helping us with TRL 3 and etch recipes for our tip fabrication process; check out some of the results from this week below:
Scanning electron microscopy and plasma etch process research conducted as part of a user project at the Center for Nanophase Materials Sciences (CNMS), which is a US Department of Energy, Office of Science User Facility at Oak Ridge National Laboratory.
The fabrication of these magnificent microtips was supported by the US Department of Energy via the Center for Nanophase Material Sciences at Oak Ridge National Lab! They are very uniform in height, at 13.2 microns, but not quite sufficiently uniform in tip radius in the current iteration. For TRL 3, they should be sufficient – this array will be able to perform field effect ionization nearly any gas other than Helium.
Dr. Lavrik has also been supporting us with COMSOL simulations to try to validate our analytical model, which have led to interesting insights into the design. More on that in future weeks.
Meanwhile, company advisor and ion thruster guru Mike Patterson has been working with his team at Desert Works Propulsion to assemble a thrust test stand and suitable vacuum chamber to test the prototype thruster once it is ready. Check out their progress below:
This is the thrust test stand – it is still a work in progress (they tell me they are two weeks behind schedule on it, though their schedule was so fast that it is still going to be complete long before we need it), but it is done enough that it went through its first calibration test yesterday, which is the process of applying known forces to the thrust stand and taking displacement readings, so that when you get a displacement reading from a thruster applying thrust, you can translate that back to a known amount of force. The little digital readout is the displacement output.
Once the thrust stand is done, it will go in here:
That’s a modest-sized vacuum chamber with some massive pumping capacity – 4,200 liters per second of nitrogen. It’s fast enough to hit 1.7 x 10^-7 Torr in an hour (which, if you are a vacuum chamber nerd, you will know is FAST). This means, hopefully, that as my thruster emits exhaust into the chamber, the background pressure can be kept low enough that it doesn’t cause arcing, spark any plasma, or lead to too much rapid charge exchange erosion. Meaning, I can turn it on for more than 5 seconds at a time without melting anything.
We do know that there will almost certainly be some sputter erosion of the chamber walls, which will back-deposit material onto the thruster, and eventually cause it to short out. The downside of testing in vacuum chambers instead of in orbit is that all vacuum chambers have walls, so if you shoot out plasma in them, it always hits something; space has no walls, so this is much less of a problem. But, even with this limitation, we expect we will be able to operate long enough in this chamber to get good data.
And that’s it for this week. Lots of excitement – probably much more than future weeks will hold, as I’ve been playing catch up with these updates, so this is really about two months of work that all kind of came to fruition in the last few days. But, excitement or no, we keep going.
Cheers,
Jon