 Thanks Nicola. So I'd like to really thank the organizer. It's very nice to be in this session I mean with all the speakers. I don't really have to introduce high-troop with computing the bad news is I hope you Not fed up at this time with with anything that's high-troop with So interestingly you're gonna see some common themes in this talk. I mean with what previous speaker talked about So the Of course as I said, I'll go very quickly on the principle of high-troop with computing Then I will show some examples and try to to see if we can learn some some general things about High-troop with computing in these examples and this example are transparent connecting oxides and and something a little more recent And that's more a little more exotic which is Electrides And I will make some points if I have some time on the challenges of scaling up in Automation especially in terms of software and methods and workflows So I don't have to show that for a long time you guys know all about high-troop with computing now And and the specific example where we apply this type of technique is transparent conducting oxides And I mean you can also ask yourself. I mean, why do you need a transparent and conducting material? You actually do have a transparent and conducting material in your pocket if you have a smartphone There's a touchscreen the touchscreen is a transparent conducting material and it's an interesting problem for the material scientist because usually the Transparent materials are not very conductive and vice versa So there are ways to make transparent conducting material. This is known for decades now You basically take an oxide with a large enough bang-gap that will give you transparency And then how do you get conductivity? You just dope it. You can dope it n-type. You can dope it p-type This is the these are the options It is these materials are known to produce the mass produce actually on the n-type TCOs I mean, this is a big industry indium oxide. For instance, it's a very big transparent conducting oxide The problem is more in terms of scientific challenges is more than p-type TCOs where actually There's not widely commercialized p-type TCO and the reason why is its properties are just poor One of the property which is the mobilities of the carriers is roughly one order of magnitude difference between the best P type in the lab and the best n-type that are known This is a gap. That's a problem for certain applications. The most fancy ones being having transparent electronics Maybe more relevant for the planet is is that solar cells would I mean you can have new device and new type of solar cells If you had the p-type conducting layer and transparent So if you get back to the basics what's what driving conductivity is the amount of carriers and also the mobilities of your carriers And if you get back to mobilities What's driving mobility is scattering which is typically not Very easy to get from ab initio and then you have a big component of effective mass What I what I mean is that a material with the lousy effective mass is not going to be a high mobility material This is this is a necessary condition And the good news is that effective mass is something you can get Right, I mean relatively easily with with simple DFT band structure and in the effective mass at least for screening I'm not that bad with DFT So this is what you want you know what you want you want you want to material with specific band structure You want a low whole effective mass. I mean it's something like lower than one would be great you want a bandgap that's higher than 3v to get you transparency and Something important also you'd be able you want to be able to make this material p Okay, I don't really care about what's happening at the connection band I really focus here on p-type oxides because this is where we can make new materials This is the work for me. You've seen a lot of these funnels I mean, this is an horizontal funnel and the mobility we assess by whole effective mass Then we go to transparency through bandgaps and then carrier through the mobility and we've done that screening on known Oxides and this is an important point. I mean we're not making predictions here These are materials that are known that are supposed to exist that have been reported to be synthesized Most of the time you just don't know their property You don't know if they're transparent You don't know if the car is will be a highly mobile or not In terms of technique and codes This is done at the DFT level then we go GW and also GW or HSC actually and so far I haven't seen much discrepancy in terms of the defects computation with both these methods So let's start with the first step I'll get back to to infrastructure But band structures are not the most difficult thing to to to automatize in terms of workflows and we We I mean build this workflow that basically gives you a band structure DFT band structure for any material If you have the band structure, then you can extract an effective mass This is easy to say by the way you can have different definition of effective mass I'm not going to go into details I mean if you're interested come talk to me or or look at our papers I think you have challenges in for instance if you have several bands what which one which bands do you take? You have to be careful with that Um Then you you take that I mean you take to 6,000 material and you have all this database of effective mass And you can start screening you can say give me the very low-haul effective mass That's what I want That's what I want to start with and if you look the 6,000 material and you give a reasonable criterion effective mass From 6,000 you end up with 20 around and this is the point I wanted to make into my title and this is really needle in the a slack problem and by the way You can see that from the other speakers talk Usually you have this type of problem where you start with a lot of possibilities and you end up with with the handful Maybe interesting material and this is really only one property. This is my first property You can be optimistic and the good news is I have only 20 compounds You can actually do much better than DFT and you can for instance you do GW on this 20 compounds This is totally doable in terms of computing computational time So if you combine all of that the first screening on DFT effective masses and then you you get band gaps more accurate with with GW You get this plot. This is band gaps and Effective mass for all the red dots are basically those materials that are known in the inorganic crystal structure database And we've compared them with the known p-type TCO This is the the best p-type TCO in the lab and you can see that By design of course all of them are getting better in effective mass You see that some of them will absorb too much some of them are really interesting and by the way I've put the effective mass here of the n-type and you see there's hope for some of the materials to get as highly mobile as n-type But there are not many materials Haven't talked about something which is dope ability and this is really the very very last screening We do and this is important actually many application. I think I think we've been spoiled with with silicon We sometimes have the feeling you can dope everything n-type or p-type. This is definitely not the case in oxides So it's not because I'm coming with a nice oxide and I say oh if you could dope it that will be great There are sometimes intrinsic defects in oxide and this is very well known that will prevent any doping and What I'm really looking at ears because I want to do whole doping is whole caters and oxygen vacancy It's notoriously known to be a whole killer. I'm not gonna go here in details, but you can do defect computation I mean, this is there is a huge our community doing a defect computation And you can actually have a sense of do you the issue material have does your material have a lot of whole caters? If you have a lot of whole caters, that's a no-go for making a p-type TCO And if you do that There are few materials that are actually killed purely by by defects And I think this is important an important point I mean defects are important and we're gonna have to take care of them more and more in high-to-put computing But let's pick up the materials that are in the sweet spot where they seem to have an interesting Large band gap and low whole effective mass, and I'm not gonna talk about all these materials I'm just gonna pick one and I'm gonna pick that one because that one we worked with experimentalist actually to kind of see if we Were right in terms of of the interest of this material, and this is biomebismet tantalum oxide This is a very interesting band structure large band gap and a very curvy Veil's band and this is what you were looking for curvy bands and large band gaps By the way, we did also defect computation that shows that this material has a tendency to be p-type But the defect chemistry was quite complex It's a mixed perovskite So it's a double perovskite bismuth 3 plus and tantalum 5 plus go on the B site And there's something that's wonderful about perovskite You cannot really compute that but for some reason if you go to an experimentalist and you ask them to make a Compound they usually not very happy They're like, oh, I just make a genius say it's a perovskite, and then they're okay with it I don't know if they're fascinated by perovskites So we managed to convince a few a bold experimentalist to actually make this material by the way This was a known material so but still there were still challenges in making it making it really pure And getting tin films of it So this is a collaboration with professor Kim and and also professor Sintovich for from Cornell So they made this material and then you start characterizing it and then you a little bit as a theorist a little bit anxious about the result You will get Good first news is that this material is definitely transparent very transparent Then conductivity measurements were not that good news because as is the conductivity was not very high You could not really measure any conductivity. So we looked into okay Can you entrance? I mean can you extrinsically dope this material and potassium is a novice obvious choice For crystal chemistry, but also from from computation and if you do the potassium on the biome side Then you can get conductivity and the best sample we add in terms of mobility with whole measurements around 38 and and I remind you that that the best p-type TCO around 10 So this is a big step. This is a big step to fan this material. They're not many oxide with such a high Whole mobility and still transparent So this was very nice and why is it this material not in your next? Smartphone, I mean the promise actually carry concentration for a moment So mobility is great But the defect chemistry of this material is so difficult to deal with that we cannot reach higher carry concentration that That 10 to the 14 which is really low It's at the limit of the measurement you can make not very useful for a device So we're working now on getting that understanding better the material and getting that higher An interesting thing is that you can learn something here I think this is the first Bismuth based p-type TCO and Bismuth is really an important ingredient in the story If you actually look at the character of the valence band You would see that this Bismuth hybridizing with oxygen P And this is an important ingredient because if you had only oxygen P You would have a very flat band and because you have Bismuth you start hybridizing and you start making your band more curvy So maybe there are also Bismuth based materials that could be very interesting By the way the field of p-type TCO would not have suggested something with Bismuth. This is not very common On this line you can say okay Is there anything more general than just you you want to work with Bismuth? Actually, you would like to work with main group reduced elements Bismuth 3 plus but also SN2 plus PB2 plus they really show up a lot in our screening and it makes sense What happens with those guys is that they really their orbitals their earth's orbitals is really aligned with the p-orbitals You get hybridization and this is really helping getting a low or effective mass. This is our champions potassium tin oxide This is the most curvy valence band I've ever seen with such a large band gap Okay, I just kind of pointed out that that From 6,000 oxides you had only 20 and there are reasons for that the oxygen piece really usually flat and usually doesn't hybridize with metals So oxides are kind of it's by itself by nature very difficult to find for p-type Mobilities, so can you go beyond oxide and this is one of the nice thing with high throughput computing if you have the framework to do oxides you can do anything else than oxides and and we kept on Cranking the number and computing and we extended our database and we are now around 30,000 materials that are screened including of course now oxides The thing you can do also is do some I would say this is really the most simple that the mining you can think of you Can just do some histograms and compare chemistries, but it's already very instruct instructive This is the distribution of effective mass whole effective mass in oxide sulfides nitride and phosphide in our database And you can see that if your criteria is really getting very low whole effective mass you're going to work with phosphides than oxides So this is really clear The problem is that if you take bang gap into the mix as you go with lower effective mass Your bang gap is actually getting lower and lower and the phosphide will tend to have much smaller bang gaps And so there will be less transparent So you kind of you kind of stop This is nice. I mean this is you can see that in semiconductor simple models like KP theory and things like that There's kind of correlation between effective mass and bang gap So you see it coming from the data, so you kind of screwed if you're trying to do go beyond oxides But there's a way and if you think about it in certain application The TCO would actually be very thin and you if you're very thin You can deal with with weak adsorption and you can play then with undirect versus direct gap And then you make your criteria a little more complicated I just want my direct gap to be transparent to be larger than 3v But I can live with a smaller undirect gap and if you do that And you especially look at Phosphides because first fight showed very very low whole effective mass you look at this You can do this type of plots This is based on GW calculation again And this is the best phosphide in terms of effective mass is similar to the plot I showed But now you have two bang gaps. You have undirect and direct bang gap And if you put the criteria you want a direct bang gap That's higher than 3v and a low effective mass. You see this material born phosphide really popping out of the data Born phosphide is really interesting. It's a very simple zinc blend very low whole effective mass large direct gap It's actually p-type doble. We've done a lot of defect computation in there And we said, okay, this is impossible. This is a binary 3v. This must be this is studied somewhere So we started looking in the literature. This is a material that has been heavily studied in the 60s 70s So you can get some data and actually all the experimental data agree with what we we see computationally shows very high Mobilities for some of the samples p-type dopa ability relatively low absorption due to the indirect gap and Conductivities that are really promising for p-type transparent application and and so this is this is an interesting It's not example where a material kind of has been abandoned And we hope computation might revive it because you say, you know this material might be interesting for that application That that may be never was never thought of So this is this is also one of the things you can get from from high-truple computing Okay, sharing is very important We may I mean we're building this very large databases all this data on effective mass and actually more transport Related measurements like c-bex or computation like c-bex are useful for many application including custom talk about Photovoltaics this is really important to have effective masses there also and we have a scientific data paper So basically can download the data and do that a mining do searchers maybe for your application Some of the materials we disregard are really interesting So I think this is a movement in the in the community and this is important And and this data is going to be soon available materials project too So you will be able to get the raw data, but also like a nicer web interface around it Okay, I don't know how much time I have Okay, I want to talk quickly about something else. That's a little more recent It's it's it's this electrolyte materials. I mean how many of you guys know about electrolytes. Have you heard about it? It's actually interesting because the chemistry community is crazy about that So it's but I think they're really interesting materials and and the concept is is that you have electrons That are basically not localized around the nuclei, but I'm look localizing pockets in 2d layers and in very of nuclei Positions and and the chemists usually say Electroys are basically materials ionic materials where yeah anion is an electron. That's that's a very Intuitive way of saying it These are this is important in many application fields And it's also very interesting in terms of the type of physics you could get with this type of Electron localization They're actually only a handful of materials that are electrolytes that are known and I'm citing them here One of the most interesting one in my opinion is calcium nitride. You can see these layers here These are basically layers of electrons between the calcium nitride layers So these are really cool materials, but they're only very difficult to find so we said, okay Can we do high throughput computing? And it's a very good example where you can do high throughput computing you can look actually at the square of the wave function Around the Fermi level and this is and see if that's localized around the nuclei Or do you have localization of electrons around pockets or more complicated shape? I'm not going to go in details, but you can do that screening on 30,000 compounds and From that we discover more than 60 new electrize by the way We found back the ones that are known which is always nice and reassuring and you can do start doing Classification between you can some of them make zero d electrolytes I would say the pockets of electrons then you have this very nice 1d channels of electrons But the point I want to make is one of the interesting thing is this field has this rule that's kind of put a little bit bits everywhere in the in the in papers that says that Partially field D shell transition metals cannot lead to electrolytes Basically, if you have like a fee tree plus it will not make an electrolyte It would just capture the electron and the electron will not be want to go in this one of these cases And we actually found the exception to the rule and I think that's really the nice thing with high throughput computing It can also challenge rules and this is as far as we know the first partially field D shell Transition metal Electride this is this is DFT computation, but we've done all kinds of things with on hybrids with on self consistent GWU Calculation and that's really shows the same Localization of the electron around family level in this nice tunel and it has chromium So I think I want to make a point that high throughput computing It's a little dumb somehow you do the screening without sometimes putting in rules But it also helps questioning the rules that might be in the field Okay, I'll quickly mention that going from one computation babysitting computation To many many of this computation you have to deal with is a big challenge I mean if you have if you have children you can imagine that and so a lot of the work It's actually spent in trying to make these workflows and trying to make automata automatic things And it's really exciting since I started we started with easy workflows And we want more and more work with more complicated things and things methods like recently We work on high to put GW we are starting working on high to put defects computation Which also is very challenging if you think about charge defects where you have corrections and so on But you really do feel this moving in more and more complex workflows Which require more and more complex software to handle that and Quickly I want to mention with recently work on the high to put phone on workflow Which is close to to what they call I talked about and this this led to this is done with abinit And this led so far to a thousand five hundred phone and band structure that have been compute entirely automatically handling arrows and so on And this is again in a Soon in scientific data, we hope and and this is going to be also available in materials project. Okay on that I'll conclude I'll thank all the the people in my group that really worked very hard on this project the collaborators also and And the funding agencies. Thank you very much