Session Notes:

0:00:01   Hello. My name is Simon Elliott from Schrödinger computational chemistry companies specializing in
0:00:07   atomic level process simulations. Thank you for coming to my talk at the particle Vale de event. I
0:00:13   thank you to forge Nano putting on this event like Teoh. Start my presentation. I'll be speaking
0:00:21   about the computer design of thermally stable precursors for L D and C V. I told my co authors in
0:00:29   Schrödinger who helped with this study
0:00:34   what we're doing in a computational screenings, the same as other screening approaches where
0:00:40   surveying a wide volume wide area of chemical space and we'd like to find that one nuggets those few
0:00:47   exciting new precursors that fulfill the property. So we're looking for So this is quite a difficult
0:00:55   feast. In the case of chemical screening, uh, the space is very large. When we're looking at organa
0:01:03   metallic precursors. It's all about choosing the Liggins. And here is a selection of some of the big
0:01:08   guns that are frequently used in the area of A, L, D and C V. And combining all these Liggins I can
0:01:15   lead to very, very many precursor compounds. So we want to use computers to help select the best
0:01:21   precursor compounds screen and provide the best candidates for an experiment study. In this study,
0:01:31   we want to look at hetero elliptic precursors. Hydroelectric means were combining different Liggins
0:01:36   in one molecule. So shown on the left is a home electric precursor of zirconia. The SL vessel, a
0:01:44   might on shown of the rice, is a naturalistic precursor that combines CP Rail on three messily
0:01:53   rights on the set. Reelected Precursor, called Zuelle, is in use in high volume manufacturing
0:02:01   because it combines some of the advantages of the two different vacants stability of the CPI rig on
0:02:07   the re activists in the A my groups. So this is a profitable strategy and but one of the problems is
0:02:16   is a very large space of ways that you can combine different Liggins together into one molecule. And
0:02:22   that's what we're going to address in this job. So how many possibilities are there? The formula is
0:02:28   given there on the left. If you have a complex which is cute, Negan's on you have a library of
0:02:35   Liggins, maybe, and Liggins in your library. The show on the X axis. This formula gives you the
0:02:40   number of possible complexes that you should consider, For example, if we look us Tetra Vaillant
0:02:47   complexes like Cicconi, um, the topic of this talk. And for example, if you have 11 different
0:02:53   ligaments, you can combine them together in 1000 and one different weights around. There's a cobia
0:02:59   center. So this is a large problem, a large number of structures of large number monikers to be
0:03:07   considered. And clearly you couldn't synthesize all 1000 molecules testing experiment Ah, on the
0:03:14   computer, running this many molecules is very, very doable. Such calculations typically take about
0:03:21   one hour per complex, so that would be 40 CPU dates on a typical computer. And if you run in
0:03:29   parallel, you could get through those many calculations a few days. So computational screening it is
0:03:39   much, much faster than its experimental partner, and it could therefore be used to narrow down the
0:03:47   space just a small number of molecules that can be looked at in more detail. So in this study were
0:03:53   actually going to just look at the combination of six different ligaments. Now, the shoulder on the
0:03:58   left plugging an equal six into this formula, gives US 126 unit complexes on of them were going only
0:04:05   going to look at the ones that are up to six coordinates. So that's 81 different molecules that are
0:04:10   six courts. So let's look at the Ligand Library. We have three mono dentists. Liggins on three by
0:04:16   dentist Liggett's Oman. Identical twins attached fire one after by dentition touch fire to Assam's
0:04:22   two teeth to the metal center. The big guns are all nitrogen bearing and oxygen free, so I selected
0:04:30   two. Primary A minds dimethyl a mind and I f l A M I. Just so they differ in the length of the Al
0:04:36   Qaeda group the Scion No ligament and then three by Dent eight Liggins to originates, one with two
0:04:44   methyl groups. The other was to Ethel. Groups swing in their differing in the size of the Al Qaeda
0:04:48   substitute int on ago annotate that has an extra nitrogen added to the backup. So, as I say, putting
0:04:57   together these six different Legans, all the different possible had to adapt precursors. We get 81
0:05:02   different complexes as an oxygen free ligand. It's interesting to see what kind of oxygen free
0:05:09   compounds this ligand sesh could be used for. Of course, mostly the A, L, D and CVD literature and
0:05:16   Cicconi, um is dominated by zirconium oxide and this very, very many studies of that process for
0:05:23   zirconium nitrite we find a lot less relatively simple. Uh, precursors have been used in each case
0:05:31   the whole dialectic eletricas. Any mines, you know, the thermal or plasma guilty. There's also a
0:05:37   published study on the CBD of zirconium carbide. Just using a single source contain CP Rick on day
0:05:43   mites. So plenty of space for innovation here. Let's see what kind of costs full compounds could
0:05:50   fulfill what's needed. Frailty. So our strategy is start with the metal center to find our ligand
0:05:57   library that have introduced to you and then that the computer systematically generated all the
0:06:01   possible structures. That's cause of liberation. We have our 81 different complexes and then submit
0:06:08   that to quantum mechanics for optimizing the structure. Andi, for computing a property, so we need a
0:06:14   proper C by which to decide which are the best molecules of which other less good molecules. So this
0:06:20   isn't metric or screening criteria. And again, quantum mechanics, or indeed, molecular mechanics can
0:06:27   be used to compute these properties, and hence screen out the best molecules. What we need here is a
0:06:34   robust workflow, right across the whole strategy, so this we can reliably run hundreds and hundreds
0:06:40   of calculations that was happening crashing, and that's really where the skin of this kind of thing.
0:06:49   So what kind of properties might be we'd be interested in for a precursor? These are the main things.
0:06:55   The synthesis should be cost effective. The precursor itself should form a the Quirt, ideally at
0:07:02   room temperature and should be volatile. Three. A port. It should be thoroughly stable for a lady
0:07:09   reaching the surface of the substrates without decomposing. And in CBD, on the other hand, we'd like
0:07:15   it to decompose of the temperature when it's heated in the gas phase of produce active species that
0:07:19   react with surface about, obviously at the surface. In both cases, we want the target material to
0:07:27   grow and not an impure material and not a material that is contaminated by byproducts. So some of
0:07:35   these properties can be computed easily, and some our little more chicken listed there. Some of the
0:07:42   factors that we have to consider for now, let's just focus on the main property that we're going to
0:07:47   consider in this talk, which is durable stability. And for that we're going to use the bond
0:07:52   dissociation energies of the molecule as a way to estimate the terrible stability.
0:07:59   So thestreet she here is that when the molecules the organa metallic precursors air in the gas phase,
0:08:06   they can be dissociated perfectly by the home electric cleavage off their constituents Bonds that
0:08:13   this is that a problem in a l D. But is perhaps what is desired at CPD. So we compute the energy
0:08:22   required to cleave all the bonds in a molecule. And we re cleave the bonds when politically, as I
0:08:30   say so this generates radicals. And these are what needs to be computed to get the bond dissociation.
0:08:36   So this is actually another level of in your operation and our problem. We have 81 complexes of
0:08:41   those complexes contained over 1200 bonds. So I listed there the numbers hundreds of hundreds of
0:08:47   carbon carbon bonds, carbon nitrogen bonds, carbon hydrogen on the metal ligand bonds themselves,
0:08:53   zirconium, attrition, bonds that actually are the center of the complex. So this would also be a
0:08:59   very tedious set of calculations to set up by hand, many many of the fragments are actually the same.
0:09:05   And so you could do the job more efficiently on. We have to compute to do that as well. So the
0:09:10   computer works through all the possible structures and fragments on puts together the energies that
0:09:15   we need. So here are the details, the calculation and it's done density functional theory to be
0:09:21   three out by p. Functional good basis session Aren't Spitko organization So here is a molecule to
0:09:29   illustrate to you what we mean by the bond dissociation. This complex is how to roll Epic it is too
0:09:37   uh, dime Ethel A might begins d m a r and any to and it has to die s away. My Lincoln's see the one
0:09:47   that the charges like ethylene has to effort groups attached to the night And all four Liggins are
0:09:53   bonded to the central zirconia. So coney amassing which is in pale blue. So the different bonds in
0:10:00   this molecule could be dissociated. I plus the bond dissociation at the y axis the to the yellow and
0:10:12   orange discs show the energy required to break the vessel to a right barked method. Nitrogen look
0:10:20   zirconium on the day, so we find 272 to 54 cabbages. But bowl and that's what's closet here. Another
0:10:29   bonds that come broken is the nitrogen carbon bond, and these data are shown US crosses and diamonds.
0:10:35   And you see the cross is shown here for this particular molecule. The carbon carbon bonds could be
0:10:41   broken to yield at methyl radical on. This is also shown with across 3 29 Villagers Brummel. The
0:10:49   carbon hydrogen bonds could be broken. They were very, very many hydrogen atoms in this molecule,
0:10:53   but actually most of them are symmetrically identical since in the end, just three different bond
0:10:57   energies ranging from 372 300 to 450 carriages promote. On these air shown is green crosses. Finally,
0:11:06   there's a mode of dissociation called pizza hydrogen elimination. But the beater hydrogen from one
0:11:11   of the ethnic groups can transfer to the nitrogen but former stable, close shelled species again, in
0:11:16   this case, which means this is often quite a low energy decomposition pathway, especially for a
0:11:22   mites, and the resulting byproduct is the thing. So for this molecule, we can see this one as an
0:11:29   energy difference of just 205 Villagers probable making us the most favorable off the bond
0:11:35   dissociation reactions in this particular article. So a prediction for this molecule will be that is
0:11:40   decomposes. First of all by this beast or hydrogen elimination passed away. And so that was just for
0:11:48   one molecule. Now I allowed in the data for the other 81 molecules. This fills up the graph, but
0:11:56   I've done here is just sort it the molecules from 1 to 81 by their lowest bonds association. So the
0:12:03   ones on the left inside How fee? Uh, their minimum bond association energy is high on for the ones
0:12:10   on the right. Inside their minimum bond dissociation energies, lows there more easily decomposed. So
0:12:17   that no stable ones on the left, most on the stable ones and right inside. And the numbers 1 to 81
0:12:22   or a simple rank ordering a lot of data. It's maybe a bit hard to take in so I can show different
0:12:32   parts of the data to help you understand this. These are just the missile ligament bond dissociation,
0:12:39   energies. So just from the zirconium to the nitrogen on, you can see a huge range, even though it's
0:12:44   always a zirconium nitrogen bond. The energy required to break its ranges from best in 150 kept just
0:12:52   normal. Up to over 450 captures befall three times stronger. We can see some Trans The Scion Oh
0:12:59   Bonds, the metal scion of arms tailed blue, is always the most stable, and that's expected for
0:13:06   Sacconi. The vessel to a mite on identity. A might is generally the least stable within this class
0:13:17   of compounds. You see, there's a range over maybe 100 colleges provoked depending on which complex
0:13:23   we're looking us. We can see this, whether it's methylene night or Ethel A might really doesn't make
0:13:30   a great deal of difference. One could maybe say that more of the dinette Soleil minds are more
0:13:35   stable and more of the diocese reminds our best stable. Basically, the to Diggins behaves the same
0:13:42   of the are a group doesn't have a huge difference in between. We see the energies for the by
0:13:47   dentists figure. This is a shown in dark blue green grey, and these are much higher. So it is
0:13:53   already an indication that the by dentists Legans are bound more strong. I suppose consistent with
0:13:58   the fact that they have two bombs. Looks, you know only when you compute these energies do you find
0:14:05   the way it's Quantum mechanics works, so those are the metal ligand bonds, and each complex has up
0:14:12   to four of these measures licking the bonds and therefore up to four dots shown on this graph. Here,
0:14:18   then, are the remaining days of these roll the carbon, carbon, carbon, nitrogen, carbon, hydrogen on
0:14:25   the beach, hydrogen elimination squares and actually the beater. Hydrogen and elimination is often
0:14:31   the lowest of the bond association entries, but not always. It's not always the way this is kind of
0:14:37   a might based molecules will decompose the carbon hydrogen bonds. The green crosses are generally
0:14:44   the most. That's stable bonds, and
0:14:48   so now essentially, these data can be used to the first level of screening the ones on the left hand
0:14:55   side of the ones with the highest energy required to make some decompose. These are the most
0:15:00   thermally stable on we would suggest. These are the candidates molecules for Atomic Claire
0:15:05   deposition. On the right hand side, these molecules decompose much more readily, and in fact you can
0:15:13   see we can predict the way they're decomposing in a Well, yeah. Uh, well separated fashion. It's the
0:15:21   A might or the Beecher hydrogen elimination that will be the decomposition pathway. So these might
0:15:26   be candidates for CVD, where we understand the decomposition through these calculations. So let's
0:15:33   look in more detail at the best. The top 10 or top 20 Reality, um, just stand there is a blow up of
0:15:43   the 20 most stable here again, it's just their metal ligand bond breaking, so you can actually see
0:15:50   which Diggins Air involved in each of them on have also out of here the label of each molecule this
0:15:58   piss cumbersome and confusing to look at. So we'll go through them molecule by molecule and identify
0:16:03   some trends. But even without looking at the chemistry, we could just pass this list to the
0:16:09   experimental colleagues on, say, choose some of these molecules. They should be, sir, very stable.
0:16:14   We've done our screen. Or we could go on and do a second level of screening on Look at some of the
0:16:19   other properties that you want to optimize in the night and l d precursor we look at that's towards
0:16:26   the end of the talk. But for now, let's see. Are there trends and is there chemistry that we can
0:16:31   understand from this data set on? The first thing about the highlights are and the difference
0:16:36   between the simple A mites such as DME dimethyl any on the fly on a group CIA on shown here circled
0:16:45   are the home electric chantry kiss molecules was get tricked a trick. It's saya No, on the Tetrick
0:16:51   is a might on the cyanide molecule is the most stable one we find the whole fest on this is expected
0:17:00   to the zirconium. Scion of armed is very strong, and it's unlikely that Sino relations molecules
0:17:09   would be used in a bloody because they aren't volatile. But we can include them in our computational.
0:17:14   That study, and so is to find out what kind of chemistry they, how they before the chemistry is like,
0:17:22   what effect? This kind of molecule hand. This kind of licking tiles on the molecule is a hole. So
0:17:27   looking here A. These days, sir, we can see that the A might normally has its upon dissociation
0:17:32   energy for the mistletoe. My bond of about 250 cabbages, Pro Bowl, but adding one the Scion Oh,
0:17:39   group is a spectator makes that bond makes that yellow dash go up. So makes us message way might
0:17:46   bond and nearly 50 colleges from all stronger on adding another siamo makes it stronger again. And I
0:17:53   think 1/3 sandal have makes this, Ah, even stronger, almost a strong as the zirconium SIA nobody. So
0:18:03   this is an effect Scion a group is having on the other Liggins here the same day suggested tabular
0:18:10   form. We're going from the electrical system. It's a mite on the left, the Tetrick. It's cyanide
0:18:14   content with rice, placing one leg with the other, and you can see this vessel to a might. Bond
0:18:19   strength is increasing across the road. What this means is that other bonds take over as the least
0:18:26   stable in the molecule. You see them shown here in bold. That's a in these middle compounds. It's
0:18:32   the nitrogen to metal bond that becomes the association actuated. His favorite, or even the carbon
0:18:39   hydrogen bonded this molecule. And so what we're seeing is that the scion no group has a spectator
0:18:45   ligand, unstable eyes, the zirconium nitrogen bonds of the other. And while it's Diana might not be
0:18:52   used. A nail d maybe other electron withdrawing groups will be interesting if this is the kind of
0:18:57   effect you'd like to have, making an otherwise unstable, partially molecule, much more state. And we
0:19:05   can also look at the by dentists. Liggett's If you remember, half of the Ligand library was by
0:19:09   dentist. And here we see one of the guanine Denyce complexes with three Sino groups on the dimethyl.
0:19:18   Imagine age three science groups on these come in as molecules Numbers three and four. So they're
0:19:23   some of the most thermally stable off the complexes. Study This shows us. And if you see the green
0:19:30   darts in the dark blue dots, they're actually the energies required to break those metal to ligand.
0:19:37   Advance to the identity Liggins. So these are very, very stable Legans more stable in fact, that the
0:19:43   scion of groups in these complexes uh and
0:19:49   this is a trend, as I said that we see right across the
0:19:54   study that the Biden taste any minds. Imagine eight square originates are more stable than the mono
0:20:00   devastating. Here's another three examples again from the very extreme left hand side, very stable
0:20:06   molecules on in this case, we've two of the by identity begins to Siam groups here. The structures,
0:20:13   those three molecules, they have very high bond dissociation energies for the message taken parts.
0:20:21   So we conclude that of identity. Liggins themselves bombs strongly to zirconium. Yeah, but what
0:20:27   other effects? They further down this list you can see now we're still in the top 20. But some of
0:20:33   these molecules are a bit less stable in terms of their vessel a might, bomb and strengths of the
0:20:39   yellow dots where the maybe the parents molecule with two a mites and to sounder groups as a kind of
0:20:46   an energy. Around 300 Villagers from all this is lowered to 50 when you add in one of identically
0:20:52   guns or even to buy identically on. This might be an indication of the crowding that the large by
0:20:57   identically gets calls because they pushed their way into the coordination sphere around. You're
0:21:02   some more examples. And these ones, I also have the by dented Liggins. I don't want her to of the so.
0:21:11   The conclusion here is that either by identity licking itself, is very strongly bound. It
0:21:15   destabilizes the neighboring on a dentist, so this would have to be taken into account of designing
0:21:21   one cook with identically. If we go to the other end of the set of complexes and going down close to
0:21:30   generate 81 of the least stable, we can have a look at what causes the molecules to be unstable and
0:21:36   again here by downtick Liggins are having a big effect. All of these complexes have to buy dented
0:21:43   Liggins and to modern technique. It's on. That is what is destabilizing. Well, no Jen, take chickens
0:21:49   so the orange yellow dots are pushed right down to nearly 150 chemicals.
0:21:57   And so we won't dig into the chemistry anymore. There's more. We can look at their in terms of the
0:22:02   carbon hydrogen, the peace are interesting elimination so on. But as I say, really, the screening is
0:22:09   don't we've classified the molecules according to this one property, and now there are other
0:22:13   properties that we really want to optimize for the A L D. Application. Let's look at those parties
0:22:22   again. If we were to try to him to produce molecules that could be synthesized, I think this is not
0:22:30   something that you can just compute with a single number but instead we'd probably take advice from
0:22:37   synthesis and build it into the way the enumeration is done. For examples we might constrain the
0:22:42   integration toe only have two different begins in. That's probably the best way to build in
0:22:48   synthetic requirements. Ah, liquids, uh, precursors, volatile precursors. This all comes back to
0:22:58   inter molecular forces, quantitative and computation off. The partial pressure, for example, is not
0:23:06   something that is straightforward. Do you did push his state of the art before we could do? But
0:23:12   there are quicker on qualitative ways to assess. But there's something is a liquid or whether it's
0:23:17   volatile based on the introduction of forces. We can't, of course, close the mass of a molecule. We
0:23:25   can calculate its entropy quite easily. It's polarize ability and because he's of energy of a liquid.
0:23:32   And so these are ways that we can attack those properties and we've looked at thermal stability, and
0:23:38   but what I'd like to just giver as another example of someone compute quite accurately and easily
0:23:44   onto mechanics is the reactivity off the A L D. Process itself in terms of thermodynamics, and here
0:23:51   again we can identify one number that encapsulates the reactivity in terms of actually making the
0:23:59   targets material. So here again, are the complexes. No, I've I've left out the scion. Oh, precursors
0:24:06   on the basis that they will never be useful. Eso from the 81. We've now gone down to 35 that have no
0:24:12   scion. Oh, in the there's all the data But that's just focus on the minimum bond association energy
0:24:18   of each of molecules. They are not showing on the left hand X axis and on the right and ex I'm sorry,
0:24:25   Why access the left by access now on the right hand, why access? I'm showing a single reaction
0:24:31   energy, which is a measure of how likely the molecule is to form zirconium nitrate product. So this
0:24:39   reaction energy includes the bond strength of Cicconi, um, to nitrogen, but also the the difference
0:24:47   in bonding and eliminating the Liggins with protons on. We want this reaction energy to be as low as
0:24:54   possible. So a negative number, if possible or small positive number. The higher the number, the
0:24:59   less likely it is to be a good l D process with no barrier. The proceeds a low temperature. So for a
0:25:05   low temperature, easy ale de process. We want a no number. So looking at that, the black dots reach
0:25:13   the precursors. We could see this. It's actually quite good news. Most of the low energy favorable
0:25:20   processes are on the left hand side, so they correlate with the thermally stable molecules. So let's
0:25:27   look at the actual identity of the ones that are the best. Thes five molecules are the most reactive,
0:25:34   and you can see that they're all involvement mono dentist mines ranging from the trickiest I assault
0:25:41   image on way to the Tetrick. It's like math Olumide on all the possibilities in between. So those
0:25:48   molecules, since there the left hand side of the graph that certainly stable. And since there on the
0:25:54   lower part of the graph
0:25:57   meconium nitrate process, there are also likely to be viable. No temperature, certain processes, and
0:26:04   the next few molecules we can see all have just one by dentist uh, ligament shown here and ground.
0:26:14   And then a mixture of the modern identically seal. The three eso these fall into a class of
0:26:19   molecules on. There seems to be a clear trends there that again, these molecules should undergo a
0:26:25   lady quite well, having more by dentist begins, pushes up this reaction energy and makes it less
0:26:31   favorite. So it's pretty good news and a the most thermally stable molecules are also the ones that
0:26:38   are most reactive in terms of a teacher. Zirconium nitride. So we've looked at two properties there
0:26:45   and used them to screen a large sesh of possible precursor molecules on identify Plea once the best
0:26:54   fish, those two properties. So to summarize high throughput computation is a useful strategy to
0:27:04   screen in a very large chemical space on narrow down that space into candidates molecules that could
0:27:12   be looked at in more detail. Computational E with really fine detailed studies or can be passed over
0:27:18   to experimental studies synthesized and tested in reactor. We could also guess chemical insights by
0:27:27   looking at the whole range of molecules and seeing the trends. In this case, we saw that the
0:27:32   electron withdrawing groups like Sanoh had a strong effect on their neighboring Liggins, making
0:27:37   stronger where the Biden Tate Liggins made the weaker. So we saw quite surprising effects within the
0:27:45   organ, a metallic complex where the thermal stability really depended on the particular mix of legal
0:27:51   that were present. So this is a useful technique for all manner of materials. Were trying to shift
0:28:00   the
0:28:05   level of difficulty to the computer on have robust and easy to use work flows for people to use the
0:28:12   computer to do screening more quickly. Guess to a small set of compounds that are gonna be
0:28:17   successful experience.
0:28:21   Thank you for attending the talk. Do please ask questions asses. The P A L D event proceeds and
0:28:30   please. Also, if you want to reach out, fire our websites www dot shorting ordered dot coms or by
0:28:38   email to me on Do I look forward to discussing this with you further? Thank you.