Particle ALD: From R&D to Commercial Scale
Dr. Markus Groner – Forge Nano
How do you take a promising lab-scale particle ALD process or application and take it to industrial scale? We’ll discuss everything from chemistry development all the way to high-throughput spatial ALD for particles. Forge Nano has years of experience and a diverse set of PALD tools for processing anything from grams to tons of powder.
0:00:01 Good morning. My name is Markus Groaner. I'm a senior scientist here at R. And D. Department at 0:00:12 Fortune nano. And I'm here to talk to you about how to scale up particle A. L. D. And as you may 0:00:19 have noticed there's been a lot of exciting news in the colorado L the community where we now have 0:00:28 three local L. D. Companies under one roof forge nano on the outskirts of Denver and we're all very 0:00:37 excited to be able to working with each other and joining forces and resources to do um all kinds of 0:00:46 L. D. And provide all kinds of equipment including semiconductor equipment. Now 0:00:58 today I'm gonna be talking about the process we go through when we want to take an exciting new lb 0:01:08 process or exciting new particle LD application um to code powder and make better materials with it. 0:01:18 Which then it can be used to make exciting new improved products including batteries and catalysts 0:01:25 and many other applications that you've been hearing about all morning. So I'm gonna talk really 0:01:37 more about the process and the methods and the resources and the the tricks we have to scale up 0:01:45 particle A. L. D. And it's basically a force that process um Where we first look into the standard L. 0:01:52 D. Process development to validate a chemistry. Um Then we scale that too inert powders as the 0:02:02 second step next is actually doing the particle alia on the customer's powder substrate typically on 0:02:12 the order of grams to kilograms before we then go on to commercial production. Typically on the tons 0:02:20 kind of scale. So of course this is the ideal process that we like to follow but that's not always 0:02:29 the case. Um People always want shortcuts or they might be a reason to skip a step in some cases. So 0:02:42 to go from a promising new A. P. L. The application that's been reported in the literature for 0:02:47 example to commercial production. And this is where fortunate comes in. And of course we do the 0:02:54 standard L. D. Process development an investigation into precursors and and all those kinds of 0:03:01 things that everybody is probably familiar with here. But more importantly we are very focused on on 0:03:09 how to apply these new processes to the customers substrates and make it work for their application. 0:03:19 And that's where we're substrate compatible limitations may come into play. We have to think about 0:03:26 cost, timeline and more important the customer's own testing. So when we scale L. D. Um to two large 0:03:41 batches of powder we of course have to consider powder behavior. Um precursor amounts that we use 0:03:50 are much much higher orders of magnitude higher than for away for LD. So we have to consider that. 0:03:56 And and there's multiple tools we can pick from two scale the process to particles. Of course we 0:04:04 have the standard types of analytical techniques that you might think of for a L. D. Including some 0:04:09 in situ monitoring. Um lots of particle characterization and of course we care a lot about 0:04:16 reproducibility since that's important for for industrial scaling. Um At the very large scale the 0:04:24 batch versus continuous LD processing options have to be considered. And I talked about that a 0:04:33 little bit more later. And of course powder handling of large volumes of powder become very 0:04:40 important and even more important is actually to consider what kinds of processes the industry is 0:04:51 used to working with are willing to accept to introduce them to their into their chain so that comes 0:04:59 into play as well. 0:05:03 So going back to the plb scaling process, four step process that I want to talk about. Uh The first 0:05:11 step is just doing the standard L. D. Development on silicon wafers on Q. C. M. Or we can look at 0:05:20 the mass cane uh while we run TME water for example 0:05:29 and a standard um L. D. Tool where we look at temperature and growth rates and the South claiming in 0:05:38 nature. And also we have, we often want to consider multiple different precursors for a certain 0:05:45 process to give us options down the road. A. L. D. On silicon wafers can be done very fast in the 0:05:55 new tools that we are now now selling uh in collaboration with Sun do um These uh ultrafast wafer 0:06:05 tools are on the order of one cycle per second. And then we'll be talking about this new L. D. To 0:06:13 the X. Presentation tomorrow in the Tuesday am session. 0:06:25 And of course we have standard analytical techniques to look at thin films on waivers. Now the main 0:06:34 purpose of this is not to uh necessarily publish a paper or or uh explore the chemistry in detail 0:06:44 but rather to kind of get a down selection of the process conditions. Find that we might then want 0:06:51 to use for the subsequent particle L. D. Scale up mm. So step two is transferring this alien 0:07:04 knowledge to particles and we typically use inert powder substrates um in a small fluid ized bed 0:07:13 tool. That's a nice tool for this kind of work it allows us to do in situ R. G. A monitoring of the 0:07:24 gases that come out of the exhaust the top of the fluid bed tool and do all kinds of chemistry 0:07:32 studies and and make sure that the L. D. Coding is coding going down as expected precursor delivery 0:07:41 becomes a big deal once we scale to particles because we can have orders of magnitude more surface 0:07:48 area than for typical object coding type LD tools. So, vapor pressure and long term stability to 0:07:59 hide that high temperature is important for the for the precursors that we're considering. And of 0:08:04 course the for the Q. C. The primary tool here is elemental analysis using ICMP. We can we can also 0:08:14 compare the calculated amount of deposition deposited materials to the actual and we also look at 0:08:26 surface area moisture and sometimes other things like powder flow or or imaging the particles and so 0:08:34 on. So really the main question we're trying to answer here is does the L. D. Process translate well 0:08:43 two doing LDL particles? Um Certain non ale ideal LD behavior maybe Okay if you're running wafers 0:08:55 but sometimes these issues really show up when you try to do them on particles and sometimes vice 0:09:02 versa as well. 0:09:06 For example uh to L. D. Tends to work quite nicely on waivers would take a long water but we found 0:09:15 it to be very amenable to particle processing just as an example. 0:09:25 Mhm. So we'll talk here a little bit more about fluid ization of particle beds. Um Since since we're 0:09:34 primarily using the fluid bed tool at this stage of the process, it's it's really nice way of 0:09:41 processing particles. Um It's kind of a commercial processing tool that we applied a particle A. L. 0:09:49 D. And run at low pressures and it really has nice mixing um Solid gas mixing, great heat transfer 0:09:57 so it's easy to heat the powder uniformly. And we can characterize the politicisation behavior of 0:10:05 the powder by by graphing the pressure drop versus the gas flow to get our fluid Ization curve has 0:10:14 shown on the right here and basically as soon as you hit the minimum fluid ization gas flow velocity, 0:10:22 um the powder starts behaving like a liquid the bad expands and if you add more gas beyond that than 0:10:32 you get um bubbling fluid Ization, which is probably what you're seeing here on the right were 0:10:39 operating above the minimum fluid ization and that's usually ok. And globalization is a whole 0:10:46 science and also somewhat of an art, I would say, with lots of tricks that we're going to talk a 0:10:52 little bit about later. And of course, uh not all substrates behave quite like this carbon over here 0:11:01 does. 0:11:06 So next, I'm gonna talk about the wonderful things we can do and we combine a mass spec with the 0:11:12 fluid bed reactor L'D tool. So we're looking at the exhaust gases that are coming out of the top of 0:11:20 the fluid eyes bed. And we can tell all kinds of things about the chemistry and the processing going 0:11:28 on in that fluid ice bed uh starting with substrate dying and outgassing. That happens when you 0:11:35 first start heating up and pumping down the the particles. 0:11:44 Then of course there's the L. D. Mechanisms that you can figure out by looking at the various 0:11:51 species that 0:11:54 are generated when you do the L. D. Um And then even more importantly perhaps is that we can detect 0:12:03 when an elder reaction has saturated. That works especially well for for fast ale reactions. So on 0:12:11 the right here you can see that when you start tomatoes, you've first, all you see is uh the methane 0:12:21 signal in red. And then after seconds minutes hours that methane signal starts starts to decrease as 0:12:29 the surface is becoming saturated. And you see a rapid rise in the T. M. A. Signal and that tells 0:12:39 you that the reaction is saturated and then you can move on to the purge and then the water dose 0:12:45 which does pretty much exactly the same thing at first. All you see is the methane product. And then 0:12:51 when all of the surfaces in the bed have been coded 0:12:56 you get water breakthrough. And the nice thing about all this is that you can even automate the 0:13:02 process 0:13:05 with with with software and and run hundreds of cycles of a process like this. And and more 0:13:13 importantly perhaps for a scale of is that we have very high precursor usage. Um we can easily 0:13:20 achieve greater than 90 precursor utilization and a lot of these processes with the mass spec. And 0:13:30 that's very important when you're going through tens or hundreds or even kilos of a certain 0:13:35 precursor doing a run. 0:13:46 Okay so once we get the L. D. Process working well on a nerd substrate, we go to the third steps 0:13:53 which is dealing with the particles that the customer has provided for their application. And that's 0:13:59 where we run into challenges with powder handling and fluid ization sometimes um including 0:14:06 agglomeration and loot creation and all kinds of issues, things that can happen. And for example on 0:14:13 the right here we have what looks almost like a paste, 0:14:19 fortunately when we started processing this material in the fluid bed, it actually started um fluid 0:14:27 izing quite nicely. Um So sometimes all it takes is drying of the substrate. 0:14:36 Another issue that arises that is that the L. D. Precursors sometimes can interact um with the bulk 0:14:46 of the powder rather than just with the surface. And that's that's often a challenging situation 0:14:52 that you can perhaps get around by using different precursors or doing different things during the 0:14:59 new creation phase of the process. And of course, some powders have very high porosity, temperature 0:15:07 limitations or they have ah inert loading requirements or certain safety issues to deal with as well. 0:15:17 We also have different types of particle ailed coding tools at the small scale. We use not only the 0:15:24 fluid ice, but also the rotating drum. Mhm. Both of these can be done at grams to kilograms types of 0:15:34 scales. And just to give you a brief introduction of rotary mixers, blenders, whatever you want to 0:15:42 call them reactors. The powder is basically cascading in a rotating drum and this allows you to do 0:15:52 static dozing. You don't have to constantly uh fluid eyes, the powder with with gas flow. And it's 0:15:59 also very useful if you want to do high surface, very high surface area porous powders or perhaps 0:16:05 slow Ellie processes. And here, instead of looking at fluid ization has shown previously, you're now 0:16:15 dealing with powder tumbling around in a drum. That can that can assume many types of forms. Most of 0:16:22 these work for for doing particle it'll be processing but of course you don't want the powder 0:16:29 sticking to the walls. 0:16:35 Yeah, as far as analysis of these powders, really more, the more important part of this is now 0:16:45 application testing by the customer. And the question often arises as to does it matter if it's 0:16:52 really good informal L'd where you don't exaggerate any particles together? 0:17:02 And the answer is kind of a yes and no. Um even if the customer doesn't care or the application 0:17:10 isn't impacted by by agglomeration for example. Um It's very useful to maintain good L. D. 0:17:18 Processing because when you have nice con formal ah self limiting L. D. It really helps with 0:17:26 reproducibility and that's one of the main reasons too. To keep striving for doing proper LD, even 0:17:35 when it might not matter in some sense. 0:17:49 So the next thing to talk about is how to deal with difficult powders. So for example, on in the 0:17:57 middle here we show slugging behavior that happens with particularly sticky powders and that can be 0:18:05 seen if you just put flour into a fluid. Ized bed for example, and there's different ways of dealing 0:18:13 with that. Oftentimes just drying the material or doing some sort of pre treatment can consult the 0:18:19 fluid Ization challenges. We also have various full realization aids on our various tools including 0:18:27 and vibrating the whole the whole bed, impacting micro jets and even stirs. 0:18:37 And the other thing is that often times fluid ization changes as you introduce the different LD 0:18:45 precursors. So on the far right here we have phosphorus stuck inside a rotary drum and we were 0:18:56 discouraged about our ability to be to be able to process this until we hit it with the first team a 0:19:02 dose. And over the course of this dose, this this powder that was completely stuck to the walls and 0:19:13 turned into nicely rolling powder. And we were able to process this material 0:19:22 and as I mentioned earlier, we have multiple reactor geometries at that. Work better or worse with 0:19:30 various powders, not just the fluid bed and the rotating bed, but also occasionally, it's actually 0:19:37 okay to do a packed bed coding. And of course we're going to talk about LD when we spatial ale D. 0:19:42 When we do talk about scale up. 0:19:47 The other thing to consider is that we have to those hundreds of grams of precursors sometimes just 0:19:52 to do one ale decoding run on a kilo of nano powders for example. Mhm. So 0:20:04 yeah of course gases tend to be easy to dose but liquids and solids are more challenging as are 0:20:11 certain precursors like ozone with a limited lifetime. 0:20:17 So we can do direct dozing of precursors we can use um yeah precursors heated in bumblers. We've 0:20:25 developed a liquid injection technique and we've also dissolved precursors. And sometimes you can 0:20:33 also generate precursors in situ. And there's different ways of of doing the dozing. You can either 0:20:41 have insert your feedback control as with the mass spec or sometimes for a well established process. 0:20:50 You can just those a predetermined quantity of precursor each for each cycle and how you quantify 0:20:59 the amount of precursor you've introduced is also a bit of a challenge and has different um 0:21:06 solutions including flow metering and even grab a metric measurements. 0:21:16 So next we're going to Uh huh. Right, so here are the two lab scale particle LD systems we have for 0:21:25 sale. Prometheus is kind of a fully featured um coding development two That has um powder volumes 0:21:36 anywhere from about 10 ml to a leader. It allows high temperature processing and bumblers with 0:21:45 precursors that can be heated 250 200 C. You can do ozone LD processing and of course the Argha is 0:21:54 the very useful process development two. And Furthermore This one allows inert loading and unloading 0:22:03 of your substrate. The pandora rotary tool is a kind of lower cost, simpler to operate tool in some 0:22:12 ways. Typically we use it for smaller powder volumes down to Miller Leader. You can also do code 0:22:21 things that don't necessarily fluid eyes that well in a fluid bed such as extra dates for potala sis 0:22:28 or even small objects. And it also allows you to do static dozing since you no longer need uh fluid 0:22:40 izing. Gas, two fluid eyes the powder. So that's useful for very high aspect ratio um structures for 0:22:49 example. And the really nice thing about this tool is that the view port allows you to see the 0:22:55 powder inside as it's tumbling around and you can make sure that it's processing well and change to 0:23:04 those in conditions if necessary to to deal with that. It also has options like the Argus which of 0:23:10 course I highly recommend uh you can even put a Q. C. M. In there and there's a GMP compatible 0:23:19 versions of this tool as well. 0:23:23 All right. And finally we have stepped forward in our process where we go to industrial scale 0:23:32 particle L. D. Depending on your application that might meet tons per year tons per day or even tons 0:23:40 per hour of particles being coded. And the nice thing is we have lots of options um for for batch LD 0:23:51 you can scale up your fluid bed to to quite large sizes if you wanted to. But we also offer a large 0:23:59 rotary blender tool ethos. And this is a small prototype version of it on the right that we have 0:24:08 running right now. And we're doing some um actually some some rocket fuel development basically in 0:24:17 this tool. And there are larger drums available for that. We also have continuous and semi 0:24:24 continuous particle of the processing the oceans which I'll show in the next slide and deciding 0:24:31 between the different types of tools is often a question of LV film thickness and the surface area 0:24:40 of powder as well as of course um precursor properties and delivery and and cost 0:24:50 but oftentimes what people look at it is the cost of the process that is dominated by the capital 0:24:58 cost. We use commercial powder handling equipment of various types to convey the powder into and out 0:25:06 of the reactor. Um So that's usually not that extravagant. But precursor costs can become quite 0:25:17 important if you have a thick l fillmore, a very high surface area powders. And at other times 0:25:25 operating costs such as time and labor and utilities, especially the heating and cooling of the 0:25:30 powder can be the dominant driver. And and like I said, it really varies from application to 0:25:37 application, 0:25:40 and your added cost for coding a kilo of powder might be since cents per kilo or my bt dollars per 0:25:49 kilo or even even higher spending. It just really depends on on the 0:25:58 factors involved here. Also, when you're doing industrial processing powders, things like safety, 0:26:05 the reproducibility of the process and various regulations come into play and may also affect affect 0:26:13 the type of tool that you choose. 0:26:17 Uh Most importantly, the industry itself often determines what type of processing tool is most 0:26:26 appropriate based on how it how it fits into their existing processing line. 0:26:36 Finally, I want to show you special particle LD. So this is basically taking the special ale D. Kind 0:26:45 of methods that you may have seen on waivers and translating them to powder so we can actually drop 0:26:56 batches of powder through zones of precursor and purges and this pot system over on the right um 0:27:10 and do multiple ale. These cycles like that by by building multiple pots. And this other tool here, 0:27:20 Sir C is a continuous ailed process that is almost like a world or old tool for for powder in that 0:27:30 powder is continuously moving through different zones of purging and precursor A and precursor P. 0:27:39 And both of these have been shown to produce powder on the tons per day and even tons per hour type 0:27:47 of rate. Especially if you're only looking for a few LD cycles, which is often the case for for for 0:27:56 battery applications that we are heavily involved with. 0:28:03 So again, this this is showing the diagrams of the three different 0:28:09 large scale powder processing tools. 0:28:15 Since the previous pictures may not have been as clear. 0:28:21 Yeah, for more information on this kind of processing, definitely take a look at James. James has 0:28:28 talked from last year. That's in the PLD Summit Video Library. Oh and finally, to summarize, I hope 0:28:39 I have shown you how fortunate, you know, can provide that phase two in the South Park colorado, mm 0:28:50 for the gnomes who in this case no how to collect L. D. Coded powder but don't quite know how that's 0:28:57 going to be commercialize, able. And I'm Marcus Groner. I'd be happy to answer questions now or 0:29:05 online or hopefully see you in the future in our new location here at uh, thorton colorado 0:29:16 or at some other conference. Thank you.