We're sorry but this page doesn't work properly without JavaScript enabled. Please enable it to continue.
Feedback

Unravelling the atomic scale chemistry of atomic level processing

00:00

Formal Metadata

Title
Unravelling the atomic scale chemistry of atomic level processing
Title of Series
Number of Parts
34
Author
License
CC Attribution 4.0 International:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
Identifiers
Publisher
Release Date
Language
Producer
Production Year2023
Production PlaceFrankfurt am Main

Content Metadata

Subject Area
Genre
Abstract
In modern semiconductor device fabrication, the dimensions involved means that Atomic Level Processing, exemplified by Atomic Layer Deposition (ALD), is widely used for film deposition. Further scaling and use of complex three-dimensional structures means that Thermal Atomic Layer Etch (tALE) will start to take centre stage in etching. The key chemistry takes place at surfaces which drives the self-limiting characteristics and other advantages of these atomic level processing approaches. I will present examples to show how first principles atomistic simulations based on Density Functional Theory can be used to predict the chemistry of atomic level deposition and etch processes. I will first discuss the key chemistries involved in atomic level processing chemistries and how these can be accessed by a range of atomistic simulation tools, together with challenges that we have identified in this exciting area. The first scientific topic is the simulation of plasma enhanced deposition (PE-ALD) of metals, using the example of cobalt for next generation interconnects. This is the first example of an atomistic level study of the full PE-ALD cycle for Co metal and show that the process requires use of ammonia or mixed H2/N2 plasma. Calculated energy barriers for key steps give guidance regarding the temperatures required for the process. Finally, we show how substrate pre-treatment can reduce nucleation delay and therefore allow selectivity in deposition of the target film. The second example is MLD of hybrid materials, using alucone and titanicone as the prototypical examples. Using aliphatic ethylene glycol and glycerol results in less-than-ideal growth per cycle (they lie flat) and poor ambient stability. Therefore, we developed functionalized benzene rings as rigid alternatives and show that the molecules remain upright, which provides high GPC and stability. Subsequent work on titanicones with both DFT and experiment, using these aromatic precursors, confirms the enhanced stability of MLD films which also show high growth rates. Finally, I present our work on self-limiting thermal atomic layer etching (ALE), highlighting how simulations can (1) predict the window of self-limiting etch (2) unravel the difference between amorphous and crystalline substrates and (3) probe the impact of surface orientation on tALE chemistry, all of which are important for future thermal ALE processing on complex 3D substrates.
Keywords
AtomclusterCopperMoleculePrecursor (chemistry)Chemical vapor depositionCork taintInsulinFunctional groupRiver sourceAtomic numberChemistryFunctional groupMaterials scienceErdrutschAtomSea levelMoleculeComputer animationMeeting/Interview
ChemistryAtomic layer depositionAtomic numberHybridisierung <Chemie>Azo couplingInfrastructureDeep seaAtomic layer depositionSimulationMaterials scienceTranslation <Genetik>Computer animation
Diamantähnlicher KohlenstoffElectronic cigaretteWine tasting descriptorsAreaComputer animation
AtomSystemic therapyFoodMaterials scienceBedrockAreaSpinning (polymers)Computer animation
CrystalHybridisierung <Chemie>SiliconAssembly (demo party)Phase (waves)BiophotonicsChemical structureNitrideEpitaxieQuantum chemistryDiamondMemory-EffektInfrastructureCarbonate platformPhysical chemistryPotenz <Homöopathie>Organische ChemieHuman body temperatureHalogenChemical reactionSurface sciencePauli, WolfgangBulk modulusReactivity (chemistry)MoleculeElectronPrecursor (chemistry)ArzneimitteldosisSubstrat <Chemie>ChemistryAtomic numberReaction mechanismEnzymkinetikAlp (name)SauerstoffatomAtomic layer depositionHafniumChemical reactionPauli, WolfgangWursthülleIndustrial etchingSubstrat <Chemie>LigandGene clusterAction potentialController (control theory)ÜbergangszustandSurface scienceHydrogenSiliconPrecursor (chemistry)ChemistryCalculus (medicine)Reaction mechanismActivation energyReaction rate constantSilicon tetrafluorideMetalOxideSilicon tetrachlorideSea levelHydroxylMoleculeLigandAtomLeft-wing politicsChemical structureCrystallographic defectAageGesundheitsstörungMachinabilityElimination reactionOzoneOxideActive siteChemical vapor depositionWaterIonenbindungEnzymkinetikReactivity (chemistry)SimulationBinding energyChlorideMonolayer <Biologie>Functional groupQuantum chemistryHuman body temperatureSetzen <Verfahrenstechnik>FluorideFluorineLactitolCHARGE syndromeErdrutschBase (chemistry)Metabolic pathwayBreed standardTitaniumRadical (chemistry)AmmoniaSpeciesComputer animation
Alp (name)Growth mediumChemical propertySetzen <Verfahrenstechnik>AdenineMetalElectronic cigaretteChemical vapor depositionChemical structureAtomStuffingIndustrial etchingGap junctionAreaConformational isomerismComputer animation
SaltAlp (name)MeatFrench friesPilot experimentMetalPrecursor (chemistry)ChemistryAtomic layer depositionAle MunicipalityCopperRutheniumCobaltChemical vapor depositionAldosteroneCell growthAtomic numberHydrophobic effectIslandZellmigrationSurface scienceReaction mechanismChemical reactionRadical (chemistry)Chemical structureHydrogenAtomSea levelMixing (process engineering)MoleculePilot experimentStiffnessMaterials scienceHybridisierung <Chemie>Chemical vapor depositionBinding energyCopperZellmigrationRutheniumIslandChemical structureFunctional groupAmmoniaHuman body temperatureSpeciesStickstoffatomHydrogenActivation energyPrecursor (chemistry)Calculus (medicine)OxideSurface scienceFood additiveLigandCobaltAtomic layer depositionRadical (chemistry)MolekulardynamikMetalWine tasting descriptorsNitriteResistenzColourantGoldOrange juiceLeft-wing politicsAreaIndustrial etchingComputer animationEngineering drawingDiagramMolecular geometry
ChemistryReaction mechanismChemical reactionRadical (chemistry)SpeciesAcetoneLigandStickstoffatomChemical vapor depositionHydrogenAtomic numberSurface scienceAluminiumSauerstoffatomKohlenstoff-14DiolEthyleneGlycerinCell growthGel permeation chromatographyAlkoxideThin filmOrganische ChemieHybridisierung <Chemie>TrimethylaminAlcoholMethaneAdsorptionDimethylaminReactivity (chemistry)AluminiumSurface scienceGlycerinMoleculeChemistryCell growthRiver deltaChemical vapor depositionWaterMercury (element)GesundheitsstörungEthylene glycolMethanisierungElectronic cigaretteStickstoffatomDiolEthaneSetzen <Verfahrenstechnik>NitrideTrimethylaluminiumHydroxylHydrogenActivation energyZinc oxideMolekulardynamikBinding energyReaction mechanismRadical (chemistry)Precursor (chemistry)SpeciesAmmoniaIonenbindungAlcoholFunctional groupMethylgruppeMetalGlykolsäurePyridineSauerstoffatomCobaltEthyleneLigandBy-productMetabolic pathwayThickening agentHybridisierung <Chemie>Calculus (medicine)ErdrutschElimination reactionSimulationThermoformingComputer animation
TrimethylaminCell growthGel permeation chromatographyBarr bodyTransformation <Genetik>Chemical vapor depositionMoleculePorosityDimethylaminReactivity (chemistry)Reaction mechanismChemical reactionPrecursor (chemistry)Organische ChemieAromaticityStiffnessBioavailabilityFunctional groupSurface scienceHydrochinonPhenyl groupAtomic numberThin filmAmineAlcoholCork taintHybridisierung <Chemie>Silicon Integrated SystemsHydrogenBinding energyPolystyreneIndustrial etchingGentamicinChemistryBiosynthesisKhatLigandAle MunicipalityRiver deltaRutheniumPrecursor (chemistry)PolymerHafniumBinding energySubstrat <Chemie>Industrial etchingMoleculeAluminiumFunctional groupAreaIonenbindungHydrochinonSauerstoffatomWattHexamereChemical structureOligomereSurface scienceStereoselectivityChlorideTitaniumTin canChemical vapor depositionCell growthLeft-wing politicsFood additiveChemistryWalkingAromaticityOrigin of replicationAtomWaterMetalOxideCopolymerHybridisierung <Chemie>Setzen <Verfahrenstechnik>Human body temperatureAbbruchreaktionPolystyreneBiosynthesisMaterials scienceBenzeneHydrogenStickstoffatomCalculus (medicine)TocopherolStarvation responseComputer animation
ChemistryLigandBleitetraethylPrecursor (chemistry)Chemical reactionIndustrial etchingSurface scienceMoleculeBulk modulusSpeciesStop codonSubstrat <Chemie>SauerstoffatomHafniumFluorineHuman body temperatureAdsorptionMetalHydrogenMolekulardynamikInclusion (mineral)Surface scienceSetzen <Verfahrenstechnik>Precursor (chemistry)Industrial etchingReaction mechanismSubstrat <Chemie>Octane ratingMolekulardynamikChemical reactionData conversionRiver deltaMetalFluorineMoleculeSodium dodecyl sulfateSauerstoffatomCobaltElimination reactionBulk modulusButcherWaterAbbruchreaktionAtomic layer depositionSimulationLeft-wing politicsHuman body temperatureCalculus (medicine)EnzymkinetikChemistryOxideWursthülleFluorideGasZirkoniumdioxidComputer animationDiagram
ChemistryMagmaComputer animation
Transcript: English(auto-generated)
Okay, thank you very much, Max, for the kind introduction and the opportunity to present today at this month's Bolstein talk. Can you let me know if this first slide is visible,
please? Yeah, we can see the slides. Great. Okay. So thank you to those who joined the presentation this morning. So at Tyndall, I head up the Materials Modelling for Devices group. And as Max mentioned, this involves simulating with first principles, density
functional theory, molecular dynamics, the processes involved in atomic level processing of materials. So I'll start by acknowledging the people who do all the work in the group, as you can see in these photos, and the funding and infrastructure support that we receive
for this work. In particular, our computing is supported by the Irish Centre for High-End Computing and previously by PRACE and DECI, because these are pretty large scale computationally intensive simulations. So a couple of words on Tyndall just for reference. So we are
Ireland's leading deep tech ICT research centre, close on 600 people between PIs, PhD, master student engineers and support staff, with significant income, most of which comes
from competitively won projects. And we work a lot with industry, which is one of our missions, try to translate low TRL fundamental research into a higher TRL work that our industry clients
can then incorporate into their production or their products. So we found over the last five, six years that many companies in the materials processing space are interested in using simulations to better understand or design materials and also the processing
of those materials, in particular, better atomic layer deposition processes, since there are many sustainability questions around atomic layer deposition. So we're in Cork and we
have some exciting developments coming. We're expanding our footprint into a brand new building in the yellow box here. And the current campus then is in the red box. So that should come on stream before the end of this decade. And we are very much at the heart of a very
vibrant semiconductor landscape, both down in Cork in the south west of the country, but also in Dublin and in Limerick in the west of the country as well. So significant interactions with many big players in Ireland. These are the kind of areas that where we're
applying ICT into, AgriTech and food security, health life sciences, MedTech, energy in the bottom left, and then kind of our bedrock work in information and communications, including
photonics. And we take an atoms to systems approach where we go from very fundamental work, right up to building prototype systems, spinning out research into companies, some of which have gone on to be extremely successful. And so simulation of materials and their processing
is actually at the heart of these research areas shown in this slide. So just an example of how this can work in terms of photonics work. So you can go from basic fundamental physics, quantum information on the left through to materials, devices, integration, and all
the way out to systems and prototyping. So an important activity we offer as well, and please feel free if anyone's interested to contact me on this, is the AscentPlus access program, which is funded by the European Commission
and provides free of charge access to Tyndall, INL in Portugal, CEA Letty, IMEC and Fraunhofer on fabrication, characterization, virtual access to all these facilities here that you
see on this slide. And it's fully funded by the access program. And the website is down there on the bottom. And for this, you write a short proposal to access a piece of kit or people for a particular project. And then you can come to the relevant institute for
a few days or a week as needed to work directly with the people involved and to reiterate that's free of charge to the researcher who's applying. So we'll get on to some of the science here. So there's an old quote from Wolfgang Pauli
saying that God made the boat, but the surface was invented by the devil. And the idea was well a surface shares its border with the external world. And lots of things can happen
on surfaces. You can have different structures. You can have defects forming easily. You can have molecules absorbing at the surface, some of which you want in your chemistry, some of which you don't want. You can have different surface facets exposed in your material,
some of which may be useful for your target reaction, some of which may not be reactive, depends on temperature. You can tune the surface acidity, et cetera. So there's lots of phenomena that can take place at surfaces. And we take the view that that's what makes them interesting.
So this works for us. And so in atomic layer deposition as the kind of prototype, oh sorry, example of atomic level processing, what you're doing is you're starting here in the top left with some sort of substrate. And it's usually got to be terminated with something like OH
groups or NH and NH2 groups or something that can react with incoming precursors. And so we're going to try and deposit some layers of our target material. So where the pointer is here, we bring in some precursor molecules that will react at the surface. And what will happen is if it's hydroxylated, normally a hydrogen transfers to the precursor molecule,
which has some ligands on it, you form a stable elimination product, which then goes away and the precursor molecule binds to its metal site to the surface. And the key thing
in ALD is once all those initially available sites are used up through this absorption process, nothing else can happen. And so the reaction self limits. Any more precursor you add will not lead to any further deposition. And so with that precursor covered surface, you can then either bring in a co-reactant through a thermal process where you might
do this at 250 Celsius. For an oxide this could be water or ozone. That oxygen containing molecule will react with the precursor fragments on the surface. That will eliminate more ligands of the precursor and make new metal to oxygen bonds. So again, a hydrogen from water could
transfer to a ligand of the precursor and that goes away as a molecule. And then the oxygen that's left or the OH that's left will bind to the metal. And you now have a layer of your metal oxide that's being formed. The alternative approach is to use plasma assisted
ALD. And there you'd make a plasma of maybe from ozone or oxygen or ammonia. And that gives reactive species like radicals or ions that can then interact at the metal precursor covered surface to eliminate the ligands and give you the, in principle, the same metal
oxide layer you grew in the thermal process. And so that's where the pointer is. And then you repeat the process as many times as you need. And in principle, you can control layer by layer. But of course, the amount of material deposited per cycle is always less than a
monolayer in most cases. So you have to understand the details of the process and what affects the amount of material deposited per cycle. In self-limiting thermal atomic layer etch would start with a substrate without a precursor, these green balls, which usually tends to
be HF. That will modify the surface layer, but not the whole substrate. So you get this green layer here in my picture. That will then react with another precursor. It could be some sort of metal chloride or trimethyl aluminium. And that will then undergo some
ligand exchange, which will produce two volatile species, one of which contains the target etch atom that you want to remove. The one we've been looking at the most is hafnium, your high-k dielectric. And so what's fluorination like on hafnium? So you get this hafnium fluoride
type layer eliminating water sitting on HFO2. And then that can undergo further reaction with something like silicon tetrachloride to release hafnium chloride and silicon fluoride. And the hafnium has been removed and you repeat. And again, you'd like to understand what the
energetics of this process is like and how it proceeds to help you then maybe make better atomic layer etch processes. So simulations can explore detailed atomistic mechanisms of a potential process chemistry, explain experimental findings, which is something
we do with collaborators. Ideally, predict new processes and look at what if scenarios for process design. And we'd like to move towards a technology computer-aided design and framework for atomic level processing. Should always be upfront about the limitations.
So the quality of your density functional theory set up in your model is obviously important. Tends to be a trade-off between the accuracy of the calculation and the cost. In terms of computational time, that comes with a real cost. You're not guaranteed to maybe capture all the relevant details or explore the relevant chemical space completely. So there's interest
in going beyond standard DFT calculations here. Can we use machine learning, for example, to help with these surface chemistries and explore a wider chemical space more quickly? So ways to overcome these might be better HPC resources, bigger computers, so we can
do larger models, higher coverage, more precursors, et cetera. And when we're doing the calculations, we do these kinds of reaction pathways. So this is just a very basic calculation of a silicon cluster model of the surface, which has some, so those are the grey balls, it
has some hydroxyls with the red oxygens. And then there's this titanium containing precursor has absorbed, some hydrogen is transferred to give you a bound precursor, which binds to the surface through titanium to two oxygens in the Moss cluster model of the surface.
And you can then follow all that via these reaction pathways, calculating in particular transition state, so activation barriers, and then the thermodynamics of various potential products. And so that will tell you if the process may proceed based on thermodynamics,
but also if it's going to be feasible in terms of the kinetics, because you don't want too high an activation barrier, because you won't overcome that under reasonable process conditions. So all this information comes out of simulations and is really important for helping to understand
if a process may take place, given a starting substrate and potential precursors for your target material. Transition states in this case are highlighted here. OK, so I mentioned we could look at exploiting machine learning to more rapidly evaluate
quantities like activation barriers, but obviously that's not a trivial piece of work. But if you had those, you could feed them into kinetic Monte Carlo simulation for your rate constants to then do a better simulation of the A.L.D. or A.L.E. process.
So what motivates the work are these kind of roadmaps that you see here. And these are the typical things that the likes of IMEC would produce showing the scaling and the nodes as a function of the year. So this one goes out to 2036. The end is basically nanometers
and in the A's angstroms and then the types of structures that may be there. And the kind of thing that comes out for me is you start to move towards these complex three dimensional
structures. And those are where atomic layer deposition, atomic layer etch can be really powerful because of their the properties that make them useful in terms of uniformity and conformality being able to do deposition or etch on 3D structures.
And then, for example, just recently ASM published some material from their investor day looking at the types of areas that are of interest to this particular company around A.L.D. So you can see Hi-K, dielectric gap fill and metal. And then with things like chipsats
and pilot lines that are being proposed. So we'll have atomic level processing is going to be crucial in the processing of the within all these pilot lines and hybrid materials
for flexible devices. Lots of interest in using non-rigid substrates. And so you may have to do deposition with hybrid materials, mixed organic, inorganic, for example, and molecular layer deposition is really useful for this.
So I'll run through three basic examples that we've been doing. First example is A.L.D. of metals. So we're looking at cobalt. Second example is mixed organic, inorganic materials. And the third one then is thermal atomic layer etch of metal oxides. So the motivation for work on metal A.L.D. was a project we had looking at copper interconnects.
And when you get down to the transistor, the dimensions available or the volume available for the interconnect is really, really small. And you have this kind of structure here where
you're putting down a copper line. So the brown area is the copper wire. But you have to have this liner and barrier presence. So the barrier stops copper migration into dielectric. The liner helps to deposit conducting copper films rather than copper islands, which aren't conducting.
And as you go down towards these very small scales, copper tends to have a higher resistivity. And it also makes those non-conducting structures. And so there's two approaches that we were looking at in this project. One was alternative interconnects and the other one then was looking at the barrier and liner material. And so I'm going to focus on the alternative
interconnects exploring cobalt. On the barrier liner, we basically did some very large DFT, molecular dynamics using DFT and activation barriers to modify a tantalum nitrite barrier layer. So this is a tantalum nitrite surface here where tantalum is the
big gold balls. And if you put a copper structure on it, it incorporates to make these kind of 3D like islands with low energy barriers for migration. If you incorporate some ruthenium into the tantalum nitrite, instead of having an extra
ruthenium layer, it actually prevents that formation of isolated ions, makes a high energy barrier to migration. So that's one angle, right? You can make a combined single material that does the job of both the barrier and the liner. But I'm going to talk about looking at cobalt ALD, particularly plasma assisted ALD.
So we took some inspiration from this work here from Aaron Kessel's group a few years ago using these various plasmas, hydrogen, nitrogen, ammonia. And on the right is just the schematic again of the ALD cycle.
So we're going to work through each of these processes briefly. Now when you do the nitrogen, hydrogen, ammonia based plasma ALD, the starting surface is going to be terminated with nitrogen hydrogen groups. So they could be NH or NH2, mix of the two various coverages.
And that's what's going to interact with the precursor. So the first thing we do is we look at the stability of those things. This basically shows for ruthenium and cobalt surfaces OO1 across the top, OO1 across the bottom, the stability of different coverages of those NH and NH2 species as a
function of temperature. So the different color lines are the different coverages. So if we take the cobalt OO1 on the top right. So we have possibilities of just NH present. So these three here where the pointer is. And then mixed NH, NH2, and you basically walk along the temperature axis
until you hit the temperature of interest. So we took the temperature from an ALD experiment that was published. It's about 575 Kelvin. And we find at that temperature, which line do we meet, which is the most stable. So it's the red dashed line, which is 5 NH on the surface.
So that's submonolayer coverage of the NH species on this cobalt surface. So that's the surface that's going to be present when we're going to do the metal precursor pulse. And we do this then for the other surfaces of interest. So this is the one I want to focus on. And if you went to lower temperatures,
you might expect the blue line becomes more stable. So that's NH plus NH2s. And then down below 400 Kelvin or so, the 6 NH plus 2 NH2 is the stable surface. So as you play with the temperature,
you can, you'll have different surface coverages of these terminating groups will be present. So we get these kinds of structures here. The left is the cobalt OO1 in orange and the NH species are in blue for nitrogen and white. We're going to start with that surface.
And we bring in the precursor. We're using cyclopentadienyl rings here, these five membered rings. We calculate the interaction energy with van der Waals calculations included. And you'll see for all four cases, two metals, two surfaces, those energies are negative, indicating an exothermic interaction.
And reasonably large, and certainly not fizzy sorbed in our view. So we take the cobalt OO1 in the top left and focus on that. We're then going to explore the transfer of hydrogen to the CP ring and see if that's
favorable and what the activation barriers are like, because that will eliminate the CP ring, allowing the cobalt that's in the precursor to start to bind to the surface. So we have a movie here. This was Gee's work. He made all these movies. Cobalt surface with NH is
down here where the pointer is. Here's the precursor. Hydrogen migrates to the CP in the precursor. So there's the hydrogen. It'll bind and then the CPH goes away and the cobalt can now bind to the surface. Looks like this here. And the barrier is less than 0.6 eV.
It's a moderate barrier. So that's useful. And then we can look at things like coverages, right? How we do two cobalt on the surface. We can eliminate one CPH. We can. And then we can look on the right hand side
where we eliminate the CPH from the other molecule. And so we get two CoCP species sitting at the surface. And the barriers for those two processes are about 0.75 eV. So they're still reasonably moderate. And then we can look at potential
to eliminate the CP fragments from the cobalt 100. And turns out that that's actually quite easy to do. And you can get cobalt buried into the surface directly from the metal precursor pulse.
The 100 surface is a higher energy surface that may or may not be present. But you can see there's a difference between the two surface facets that we've explored here in terms of the metal precursor pulse. So we took a pragmatic approach with the plasma step. And so we just said, we look at the radicals, nitrogen, hydrogen, NH, NH2.
Those are produced in the plasma, and then those will react at the surface. So on the right hand side, you can see these white balls. They're hydrogen atoms that we introduced in a finite temperature MD simulation at 600 Kelvin.
And we allow them to just find where they'd like to be. Some of those hydrogen radicals make a H2 molecule. Some bind to the CP ligand and make a CPH, which we can then eliminate. So that's useful. And then we, if we look at this top view,
we had an NH and a hydrogen radical made its way and made an NH2. And we don't want cobalt nitride growing, right? We want cobalt. So we need to eliminate some of those nitrogens and making ammonia is one way to do it. The other way might be, you might end up with NH2s near each other that could make N2H4.
But we looked at the ammonia option. And so we can start to protonate those nitrogen species with hydrogens from the plasma. And so then we can, we can add NH. Yeah, we can add, sorry, N and then H.
So the top one is adding N. And what it does if we look on this plan view, is it inserts into the CP to make pyridine, which goes away. And this is the molecular dynamics simulations. We don't bias it. We put the nitrogen some distance away and let it find its own
pathway to where it like to go. If you have an NH, you make a pyridinium. There's the NH. It binds in to make a pyridinium. So we can eliminate CPs by making pyridine as a potential byproduct. And then we can add NH or NH2. And those combined to a cobalt,
which had lost its ligands. That then is beginning to recover the NHX terminated cobalt surface for the metal precursor step. So, OK, this video doesn't always play, but it's just meant to show when we add all these
radicals in together, we start to make various species. There's a pyridine forms here. There's another one forms here, and we're protonating some of the nitrogen down here. So, we pull out some products after short. Now, this is a very short time, obviously, because this is DFT-MD. But nonetheless, these are happening quite quickly.
So, H2 forms. NH3 has formed here. And we find that that's always favorable when you add hydromedicals. And then we've got the pyridine, et cetera, forming. So, the mechanism is that you need both the hydrogen and nitrogen.
So, you can get that from the ammonia plasma and the hydrogen can eliminate nitrogen as ammonia or can eliminate CP and then the NH and H2. Those can eliminate the CP. And so, you need both nitrogen hydrogen in there for this to proceed. So, you get this.
This is the kind of typical picture that people make for the for the cycles. The post plasma surface here where the pointer is, add the metal precursor, add the plasma species, and then you repeat. So, just summarize everything in a single slide.
So, with the activation barriers that you can get from these calculations, you can then do some KMC. So, this is work from Matti Shirazi, who was in the group, worked with Simon Elliott when he was leading the group at Tyndall on hafnia, ALD. And Matti developed a KMC simulation process to actually do that.
And then there was further work on zinc oxide in Alto, building on this. And you get these kinds of pictures here when you run the calculation. So, A is the bare surface looking down on it. And B, C, and D are after you add
precursors for a number of cycles under the deposition conditions. And you can explore then what kind of precursor fragments are present. You can get the change of mass per cycle and get a growth per cycle out of the KMC simulations. And OK, unfortunately, this one doesn't play, but the link is here to the paper.
OK, so that's the ALD work on cobalt. And then we've done hybrid materials, which was part of an EU mercury network project called HighCOTE.
And the prototype here is Alu-Cone, which is based on trimethyl aluminium using organic alcohols instead of water. So, this is the hybrid version of aluminium oxide. And so, you have this kind of schematic chemistry. So, you do the TMA here as usual.
But now you're adding a diol. So, you need two reactive ends. One end makes this new Al-O bond, eliminating methane. Then you have the molecule and then you have an OH present, ready to do the next cycle to react with the TMA. So, people tend to use ethylene glycol and glycerol as prototypes.
The difference is glycerol has a third OH group in the middle of the molecule. And you can get experimental data like this showing the growth per cycle is different for the two diols and they can have quite different deposition chemistry. So, we wanted to explore this, which hadn't been done using our typical surface slab type
molecules within DFT. There was some gas phase models were used previously. So, we can take models of the post-TMA surface. So, we can have an aluminium dimethyl or aluminium monomethyl. Both are worth looking at. And then we can do two of those precursors
on the surface. So, here the aluminium is purple. So, here there's two aluminium dimethyls. Here there's two aluminium monomethyls. And you follow this kind of scheme here, where the diol can react at the precursor terminated surface
bind to the aluminium, eliminating some of the methyls. We calculate the delta E when it binds upright here. And we calculate the change in energy relative to that delta, delta E when the molecule lies flat, because there's two reactive ends. So, it may bind to both ends.
And in all cases, here's ethylene glycol. The upright is favorable. The flat is even more favorable. So, that's in addition, that's minus 1.4 eV relative to the upright binding configuration. And then we can have the same with the depending on whether you do dimethyl aluminium or monomethyl
aluminium matter. So, that's the glycerol on the top right. And then we have the monomethyl aluminium here on the bottom, methylene glycol again. All the delta Es are negative. There's the glycerol, and that's also negative.
So, preference to lie flat, which is what was proposed from experiment. And the difference now is when the ethylene glycol lies flat, the two oxygen ends are used to bind to the surface. When the glycerol lies flat, you have an OH still available. And also, this is thicker than the
ethylene glycol. So, you will get a thicker film over whatever number of cycles you do if you use the glycerol. Because the TMA can interact, interacting at the glycerol, making this new bond here
as the dimethyl left over and the delta Es are all favourable. And we can do the ethylene glycol. What it likes to do, the TMA likes to bind to one of the oxygens that's binding the ethylene glycol to the surface. Makes an aluminium oxygen bond this way. And you can see, though, that it's not
protruding as much as it does when you bind to the glycerol. So, the thickness over a cycle is going to be lower with ethylene glycol than it is with glycerol, which is reasonably consistent with experiment. Now, the challenge with these is they're not
stable in ambient. Once you've watered them, they're not going to be stable at all. Excuse me. So, we started looking at aromatic molecules. There was some work had been done on this, work with Simon Amato Knez had been looking at this.
And we looked at it with our slab models. And we take the benzene based molecules because they have a stiff aromatic backbone. And that may prevent the line flat and binding mode from taking place. We also went from OH termination
hydroquinone to NH2 in this diamine. And then we did in the minophenol. Well, so we get these kinds of structures here, looking at our four options and binding at the post TMA surface. And the delta E's are all negative. So, they're all favourable. You could see the actual delta E depends
on what group binds to the surface. But for now, the important thing is that they're all exothermic. And you can do it by having a second TMA precursor. So, we can do the compared to the line flat.
You can see generally binding to the NH2. So, you make aluminium nitrogen bond is always less favourable than binding via the OH to make an aluminium oxygen bond. So, the upright structures are here across the top. Then we do the flat line. And relative to the upright,
these are at best either very essentially the same, either energetic or much less favourable. And the main reason they tend to be less favourable is to accommodate this. The ring has to distort, which these aromatic rings tend not to like to do.
And so, we propose the flat line situation is not going to be favourable. And you have this upright, which will give better ordering of your system, etc. So, these tend to promote reasonably thick films. Okay, so we also then explored if you functionalize the molecule by putting some chemical
group on the aromatic ring backbone, will that have any impact? The delta-Es are all very similar with the hydroquinone based molecules. So, the functionalization has no impact on the stability of the molecule when it binds
at the surface. And the same if you bind through the nitrogen. So, you can functionalize the aromatic bit to give you desired properties, but it won't affect the chemistry of the surface. Obviously, what it could affect is the packing and coverage and ordering. But the intrinsic stability of the organic precursor at the TMA covered surface
is not impacted by the functionalization of the molecule. That's okay. So, we worked with Maard Karpinen's group in Alto on looking at a bunch of precursors for titanium-based hybrid organic inorganic films. This is a nanotape surface, which has some OHs, as you can see with the pointer.
And we put a TCL precursor on, eliminated HCl, and just looked at this titanium chloride precursor interacting with the organics, just to understand that the fundamental process is taking place. So, you get these structures here, TiO in A and D, and TiN in B and C. And we calculate the delta Es,
and they're all exothermic. So, that's favorable. We did some quick checks on the flat-lying option, and that's not favorable. And so, you'd expect to get these upright adsorbed molecules,
which should give good ordering and can give good stability. The TCL molecule in the next cycle, then up here, it can interact at the organic bit. And it's also exothermic relative to the previous step. So, it's going downhill as you add each precursor. So, thermodynamically,
it's all favorable. So, we published some work with Merritt's group. They used infrared spectroscopy to explore the formation of various bonds using zinc-based hybrids as a reference.
Basically, the aromatic rings stay intact in the deposition, which is good. What was interesting was the things like growth per cycle and the deposition temperature that gave you the best growth per cycle varied quite a bit between the different types of precursors,
both the number of aromatic rings and the terminal groups that bind to the titanium containing precursor all have an influence. So, there's still a lot more to understand about the chemistry of these hybrid materials and their deposition. And then very briefly, we did some work on
sequential infiltration synthesis of ruthenium with Ghent. The idea here is you build these block copolymer structures here with polymethamethacrylate, polystyrene, you infiltrate the metal precursor. And in one of those materials,
it will infiltrate and bind with the polystyrene. So, there's the precursor. It won't infiltrate and bind to the PMMA. So, nothing happens. And you now have ruthenium containing precursor essentially mobilized in the polystyrene. You can reduce that with hydrogen, which will then give you ruthenium lines or ruthenium deposition.
You eliminate the polymer bits and you get these ruthenium lines based on this template of the block copolymer. And people know how to make template of block copolymers. So, we wanted to briefly understand why you get selective infiltration into polystyrene and not into polymethamethacrylate with this precursor.
So, you do these very basic calculations. Here is three monomers of PMMA. So, an oligomer with three units. And here's the precursor. And we looked at different ways the precursor could interact at that oligomer.
And in all cases, the energy of interaction is not favorable. So, the RuO4 will not interact and bind at the PMMA. It will basically just diffuse. It'll infiltrate a bit. And when you do the purge, it will just diffuse out. When you do the polystyrene, so again, here on the left is the polystyrene three unit oligomer.
The RuO4 here will bind at the ring. Make new bonds, reasonably large gain in energy. And we can introduce hydrogen and it interacts with oxygen from the precursor here to make water,
which is favorable. And we eliminate the water and that's reduced the ruthenium containing precursor. So, this precursor will bind to polystyrene, get immobilized and then can be reduced. And then you get to deposition. So, it's a really interesting area and I'm doing a bit of work at the moment
in understanding the fundamental chemistry of these SIS, also called vapor phase infiltration processes. So, then finally, we look quickly at self-limiting thermal atomic layer etch of metal oxides. So, this is back to the original idea of doing hafnia etch using HF and you can use Tickle.
So, HF will fluorinate the surface layer of HF02, eliminating water. And then you can get this exchange process when you bring in the Tickle, which will eliminate the hafnium from your original substrate and it's etched away.
So, we'd like to look at the thermodynamics of this process and see where self-limiting is going to be more favorable than continuous etching. Look at HF adsorption, you know, what's the mechanism by which HF molecules interact at the surface that are fluorinated. It was interesting work in comparing
crystalline with amorphous substrates and then something briefly on different surface terminations. So, what we're going to do is we built a model. So, Reid and Suresh worked with Simon on this idea here, where you start with your material being in your precursor and you have two options
where you can do a self-limiting etch called SL, which is just again, you know, modifying a layer of the surface. Or you can do spontaneous etch, where you just have continuous elimination of products until you run out of material or you hit an etch stop. So, that's a bulk to gas reaction.
And then the self-limiting is taking place on the surface. So, we do these kinds of reactions here, right? So, here down bottom left, we do bulk oxide plus HF molecules gives you gaseous metal fluoride and water eliminated.
The self-limiting then is going to take the same surface with HF, give you a layer of metal fluoride and then eliminates water. So, these are the types of models we have. These are looking down in a plan view. This is our crystalline half-neo-monoclinic one-one-one surface. And then we can fluorinate it to different
degrees. OK. So, you see here, we have all the fluorine replace, replacing, sorry, all oxygen replaced by fluorine in the right ratio. Some oxygen and then half the oxygen. And we look at the impact of all that extent of fluorination. We can tell you then which is favorable.
So, we can do a zero Kelvin delta E for those three conversion reactions. The SLs are all pretty favorable. We did some delta G calculations of finite temperature are still generally favorable. And the spontaneous continuous etching is less favorable than the self-limiting
conversion reaction. But we focused on this guy here, because he was the most favorable delta E. We do these plots. And we're doing them here on the left for crystalline half-neo and on the right for amorphous half-neo. The idea here is the brown line is the
calculated delta G for the self-limiting process. The blue line is the continuous etching. And you can see as you walk along the X axis, the brown line always lies below the blue. So, the self-limiting is more favorable. There's a point down the temperature, which the continuous etching becomes more favorable.
And so, basically, if we stay below that temperature, which in this case is about 650 Kelvin, then the self-limiting etch should be the dominant process. And that's reasonably consistent for half-neo and zirconia with the experimental data.
On the right-hand side, we do the same for amorphous half-neo. And what we find is the temperature at which continuous etch takes over lies at a higher, is higher than on the crystalline material. So, basically, the amorphous material will be,
will have a more favorable self-limiting etch up to higher temperatures. So, this is useful because now you could take a potentially new process. You could do this type of simulation, make these pictures, and then say, if self-limiting would be thermodynamically favorable, and if it is up to what temperature,
will that happen? And this is just the thermodynamics. These are relatively quick to do, but can be very, very useful. Now, we can start to look at the interaction of HF molecules on the right-hand side. We went from one molecule, one HF, which dissociated to these guys here, fluorine blue,
and there's a hydrogen, right up to 34 HF, which should be covering the whole surface. And you find mixed adsorption, not every HF interacts. So, from that, you get these plots here of x-axis is essentially theoretical coverage based on the number of HF molecules. And the y-axis is the actual coverage
that we got when we relaxed everything. And it kind of plateaus in both cases. On the left is crystalline, on the right is amorphous. And the amorphous, we can get more fluorination compared to the crystalline. And that enhances the rate at which the etch takes place.
So, we can calculate an etch rate here based on these covers. It's just an estimate, right? So, you know, as always with DFT numbers, you know, we have to be careful. But it's smaller than the etch rate for continuous etch mono that you'd expect. And the amorphous etch rate is higher than
the crystalline etch rate, which is at least consistent with the experimental results that have compared amorphous and crystalline hafnia. So, you can do calculations of various products. And we can get water, so there's water here. And we, you see here down,
there's two water molecules are formed, bottom left, and then on the bottom right, we made a peroxide. And those waters and peroxides, they're the elimination products as you try to fluorinate the surface. And the reason that you don't fluorinate the whole hafnia is that fluorine is pretty big.
So, it's very hard for it to get down into the bulk of the material. And the amorphous material, what we also find is because of its non-crystalline structure and voids, et cetera, that can be present, you can get more fluorine incorporated compared to the crystalline material. And then we briefly looked at a higher
energy, higher surface energy cut of this monoclinic hafnia, 001 in this case. We calculate these thermodynamic plots. This is the one we saw already on the left and on the right. The temperature at which you change from self-limiting to continuous
is higher than on the more stable surface, which I think seems reasonable for a higher energy surface because they tend to be more reactive. And we get different etch rates as well. So, this is larger than on the 111 surface. So, again, we can accommodate more fluorine into this higher energy surface.
So, this could be important because if you polycrystalline material, this may give different etch rates for different surface facets that are exposed. And so, your etch is not necessarily going to be uniform, which may not be desirable. So, there's some challenges.
This is bringing us to the end and opportunities. Our models tend to be static in many cases, including dynamics. It is actually very important. And we saw that when we did the work on the cobalt plasma ALD that to really see what's happening, we needed to do some ab initio MD
simulations and be useful to be able to move between scales in a seamless manner. So, from the atomistic DFT, say, to kinetic Monte Carlo, exchanging relevant information that widening the search of the reaction space in an efficient manner.
So, lots of interesting work going on there. And plasma chemistry is quite interesting. And, well, and of course, we'd like to make properly quantitatively predictive simulations. So, thank you very much for listening. And thanks again to the Bodlestein team for the opportunity to present.