Hacker News Reader

Comment by neltnerb | Hacker News Reader

Couldn't comment to the author without paying for an account there, but as someone who has actually built sCO2 reactors -- you can get away with a much lower cost HPLC pump to pressurize the CO2 if you can force it to be a liquid at the inlet side using an eductor tube and slightly heating the tank to guarantee that the slightly cooler HPLC pump head won't cavitate.

It worked pretty well, though I had to stick a TEC on the pump head. $5k instead of $100k+ for a high pressure syringe pump.

My application was to use the sCO2 to solubilize organometallic precursors to deposit thin films on particles in a fluidized bed, and one of the best reasons for sCO2 was that the only likely contamination would be carbon (sCO2 is sometimes used to form carbides intentionally).

View on HN

>My application was to use the sCO2 to solubilize organometallic precursors to deposit thin films on particles in a fluidized bed, and one of the best reasons for sCO2 was that the only likely contamination would be carbon (sCO2 is sometimes used to form carbides intentionally).

I understand some of these words.

Kidding aside, what was the goal of this endeavor? What is the end product?

Particles with thin films on them, of course.

Let's guess. I'm going to say they are dissolving the 'precursors' into the co2, pumping it through the bed of particles with sufficient vigor to achieve 'fluidized' [1] state. As the supercritical co2 rises through the bed, the particles are dumping heat into it, ultimately flashing to vapor and leaving the solute (or at least the metallic part of it) behind as a thin film on the particle that did the trick.

This of course leaves the particle itself cooler, making it less likely to receive additional deposition and resulting in a more uniform thickness across the batch.

These coated particles are probably then going to be used as a catalyst in some other process.

[1] https://www.youtube.com/watch?v=My4RA5I0FKs

(alternate is that the the precursors themselves are the fluidized bed and the co2 is used to help move the coating from the donor to the recipient. If GP called them parts and not particles i would be more inclined to this option)

Almost! Except for the flashing CO2 to vapor, I actually just had a filter at the top to keep particles from flying out but the reactor itself was at 5000PSI and I just used a backpressure regulator for the exhaust stream which was very little since basically every byproduct was a gas. But in backstory, I invented a process, particle supercricial fluid deposition (UMass Amherst previously demonstrated deposition onto wafers and I worked with them on it through STTR), which lets you put down conformal thin films onto particles as a protective coating.

https://www.brianneltner.com/thin-films

The key is conformal films and aspect ratio. If you're familiar with CVD, you might have encountered "duning" where film growth can produce unstable interfaces or find that a trench will get filled in at the top and leave a void underneath.

One common option is using ALD which uses alternating precursors so you can fully coat every surface of an arbitrarily high surface area material with a monolayer of your precursor, and then a second step converts that precursor to metal so you get atomically thick layers.

The downside of ALD is that it's quite slow in terms of growth rate because you can only grow one atom thick layer at a time. Additionally, the precursor has to have a vapor pressure high enough that the often solid precursors can actually be flowed through a reactor under vacuum. But it can get super thin films that can be conformal and which have incredible aspect ratios.

What I did was use sCO2 to dissolve chemicals like those ALD or CVD precursors in an "organic solvent" (sCO2 itself) to deliver it at super high concentrations to the reactor (at like 5000PSI) instead of at super low concentrations under vacuum. It stays supercritical through the full fluidized bed. But the reactor itself is hot so the precursors spontaneously decompose and do so preferentially on surfaces. But! The key difference is that because of the high precursor concentration, instead of the film growth rate being dictated by how fast precursor can get to the particle surface (diffusion limited) it's limited by how fast the precursor reacts on the surface (kinetically limited). The prior is notorious for unstable interfaces, the latter is much better because it's just growing at max speed at all locations simultaneously.

So the real difference is that it let me deposit thin films from a high concentration precursor stream.

My actual application was trying to coat copper flake with chromium metal to increase the oxidation temperature for use as a replacement to silver conductive pastes used in solar panel manufacturing. Silver was (might still be) like 10% of the total cost of making one, so this was of keen interest and would be pretty awesome. Ultimately, I was able to make it so that copper showed no oxidation at 300C over a minute while uncoated copper did show a lot of oxidation.

However, the amount of coating I needed to keep the oxygen away from the copper was too thick (or at least I couldn't make it thinner with that tech) and the chromium was expensive enough that it'd have ended up more expensive than just using silver.

So I moved on =) A project worth doing for sure, everyone I talked to about it simply could not guess if it would work. Sadly it worked but wasn't commercially interesting in the end.

Thanks for taking the time to post this. It was a fun read.

Super interesting. Comments like this are what keep me coming back to HN.

Holy moly this is cool! Thanks for all the detail and the link to your blog post. The process sounds really cool, i bet it was awesome seeing it work the first time!

Also:

>Almost!

You’re quite generous with your knowledge and compliments haha!

Thank you very much!

Exciting read!

Does cavitation depend on temperature in like submarines too?

CO2 is super weird. In a CO2 tank it's actually two-phase, CO2 gas and liquid in a steady state equilibrium that is exclusively a function of the temperature. Since the outlet was connected to the HPLC pump inlet, if the temperature between the tank and the pump head was exactly equal, when the pump piston draws liquid in it every so slightly reduces the pressure, which results in the liquid CO2 boiling inside the pump head producing bubbles that make metering the flow rate impossible.

So the cavitation in this case is indeed temperature driven (albeit indirectly) but it's really just a side effect of the pump type. A syringe pump style system just is well, a syringe, so the problem goes away.

Excellent use of the HPLC pump, we used to order them by the pallet.

Here's the phase diagram:

https://en.wikipedia.org/wiki/File:Carbon_dioxide_pressure-t...

I'm no engineer, but I like to work with engineers that are good at not just established "engines" but imaginary ones too.

I always thought any engineer should be able to understand this well-established graph just by looking at it, regardless of whether they had any ChE background.

Engineers already exposed to material phase behavior sometimes still take a little while, but there's not much there, it's a pretty straightforward graph.

Studying it alone should bring understanding, not of the full implications of the dynamic physico-chemical properties of course, just the states of matter that exist under the conditions indicated by the data.

Some engineers can never get this, I don't get that.

While I would prefer an engineer with an appreciation for what it takes to turn the imagination of some PhDs into apparatus capable of creating drawings like this, using commercially available components when possible, based primarily on drawings like this.

I'm no capitalist without other people's money (I personally owned two of the automated positive-displacement high-pressure [syringe] pumps), but a good capitalist to work with would have enough imagination for an intuitive understanding of the leverage used to assemble advanced equipment into money-making machines, when the majority of operators must accept the high-dollar components as unrecoverable costs.

I sort of agree, but I think you're too harsh. Having taught phase diagrams to freshmen, very smart freshmen, as with everything else it comes down to familiarity and practice. I tend not to look down on people for not knowing stuff they're not familiar with, I'm sure they could all get it with someone who can actually guide them through it.

But I do appreciate a simple, inexpensive design of course. But they're so hard for people to make if they don't understand the deep details of everything in the system, and once the knowledge is split between two brains then it's a lot harder to spot inconsistencies. But I also can hardly blame people for throwing in the towel on deeply researching the sheer vastness of the McMaster, Swagelok, and Digikey catalogs. Much less ASTM standards or the ASM handbook.

I work on building a lot of surprisingly easy things like electromagnets and battery charging circuits and low noise filters and precision temperature controllers and such like that in-house because it's shocking how bad and expensive stuff made commercially usually is. Just yesterday I was replacing a flow sensor in a water pump to water cool a vibrating scanning magnetometer my office mate designed. It's a fun line of work. I also do stuff on the side as a consultant along these lines to scratch the "let's make something people can use in industry" itch (http://neltnerlabs.com/).

But my biggest pipe dream for the past 9 years has been using AI tools to close the loop around synthesis and testing and predicting properties for heterogeneous catalysts. Equipment is a key part of that dream, sadly Google was uninterested (no ads?) and I've only ever found a handful of companies working on it all of which don't quite seem to get just how big a deal it is almost certain to be. At least the national labs are catching on.

> but I think you're too harsh.

What I found when I took thermo which is where you learn this stuff. Nothing really prepares you for the basic concepts before hand. So it's a steep hill to climb. I think some students get through it via rote memorization. Which means they don't actually understand it. I had to retake the first semester because I got a C-. Which was likely for the best.

I also think that thermo is somehow very deeply connected to how the universe actually works. And the next step down is a doozy. So it's difficult to explain it in terms of something familiar.

I was super lucky that they were taught in freshman chemistry. It made it more familiar seeming later on I'm sure.

I agree about thermo, it has an unexpected amount of symmetry. It wasn't until my third time taking a thermo course that I think I really started to get it.

I had a professor that I just straight up asked "why?" and his response was "because sometimes we want to know numbers that are hard to measure so we come up with ways to represent them using things that are easy to measure". That is by a wide margin the most practically coherent reasoning for the importance of thermo I've ever gotten, and suddenly all the seemingly arbitrarily PDE manipulation made perfect sense.

Although I think the basics like "what is entropy" are pretty generally useful regardless of if you ever touch the PDEs or free energy equations.

Differential forms in thermo can be an eye opener

I don't want to be harsh, I admire and want to recognize the creative designer.

I enjoy guiding people through unfamiliar things which I have experimented with, it's good to see the diverse ways to grasp the same fundamental concept. Some students are quicker than some degreed engineers.

You seem like a person that could make an outstanding collaborator, building systems unique to the direction of the project.

I'm glad you put your website here, I'll pick up the phone and give your company a call, there's not a more promising, businesslike way.

I like instruments also, and have more than one.

Testing pays the bills, see the message for the acetone user.

When retired it might be good to build some robotic type stuff, maybe in a capitalist venture to overcome financial limitations.

Or looks like things would be fine without any equipment, helping companies negotiate the byzantine ASTM process alone instead, just been invited to join the emerging cannabis committee as an independent voting member.

That's only about testing, but extraction operators might best benefit from more supercritical ability on their side, you could surely consider this yourself now since it's very uncommon to have experience safely handling this type equipment, especially as a researcher on an experimental basis. Haven't done EOR in years and this time it could be CBD oil.

Anyway, presenting some findings on heavy petroleum oils to my traditional ASTM committee shortly, they want to get it into the book as soon as possible because it could save operators millions across the board.

Started to meet with some AI people just over a month ago too, inviting them to the lab to see firsthand (& smell) the kinds of crude oil they have been modeling about, some engineers for the first time.

Yeah, I have the deepest respect for the people who do a lot of AI. I took Ng's online course and know that I don't know a ton but I'm more a "learn the basics, read papers after that" kind of person so I just started trying to replicate papers after that a bit.

Now my problem is mostly that I just don't have the energy to scour the research journals and books to put together datasets worth training with (for heterogeneous catalyst performance). Same as everyone I guess. The algorithms don't seem that bad, albeit obviously I want to run it past some friends that do it before claiming anything since I know I'll do it badly and just not know, so maybe in a perfect world it's just a matter of needing more data.

I honestly think it will work because I with a human brain can look at trends in catalyst manufacturing methods and composition and guess at least reasonably locally what changes will do. The parameter space is just too big between atomic composition, preparation methods, support acidity, surface area, etc etc etc. I think honestly a computer could do it because it feels like a problem of fitting the data in my head enough to model it beyond super locally.

But every time I try to improve a catalyst I succeed, which I think says more about how poorly optimized we are today than it says about my cleverness -- literally almost any random change is likely one that is a variable no one has tried yet. It's an awesomely open feeling field, and more exciting now that biomolecules are getting thrown into the mix from fermentation facilities.

Reply on news.ycombinator.com