15. Transient Enhanced Diffusion (TED) - Simulation Examples, TED Calculations, RSCE in detail

Key Takeaways

started uh a couple of announcements there are uh two handouts for today there are the lecture notes for today which are um this handout 25 and you also have an additional handout 26 which we're going to go through it's a little calculation example I was hoping we could go through uh during today's lecture and have you do some calculations so it's not a homework problem we're going to do that during the lecture um uh let's see as far as homeworks I don't have any to pass out quite yet the TA is still working on those uh and she's out today but I do have the uh clipboard which I will have up front here if you want to sign up for your final project uh we started circulating that last time where you can I'm asking people to put your name down and even if you don't know your topic if you can check whether you want to do a written report or an oral presentation to the class that would be helpful and then uh once you know your topic that would be good to fill that in as well all right so I'll leave that up here for now and you can come up uh towards the end and sign up okay so um today's lecture is actually um going to be the final lecture on ION implantation and Transit enhanced diffusion uh let me review here um here we on handout 25 the first page review what we've talked about so far we talked about ion imp Planet profiles and we said you can model them uh very simply as a gal scene um more accurately as a Pearson Ford or dual Pearson distribution I'll show you some examples of real implants today and uh or they can be simulated by numerical techniques such as Monte Carlo we last time we talked a lot about the damage modeling and that we introduced something called the plus n model where N is a small number about one and this is the model for residual damage that says that there's roughly n excess uh silic interstitials injected per primary ion where N is a small number on the order of one we we then last time we talked about how these excess interstitials cluster very very quickly into defects called 311 defects and uh later these defects then dissolve during a kneeling um and their their evaporation rate um is really what determines the kinetics of transient enhanced diffusion determines how long teed lasts and also determines the magnitude of teed um in fact we had a clustering SL evaporation model that we talked about last time uh and that was used to explain the time and temperature dependence and to a certain extent the dose and energy dependence of teed in a in a rough way so what I wanted to cover this time is to actually give you some real examples of iron implantation profiles data and a little bit of simulations uh I want to spend a few minutes maybe 10 15 minutes if we have time in class you know a lot of people are sleeping in because the Red Sox won last night the World Series uh but for those of you are here I that's what the handout number 26 is is all about we'll calculate together uh a little simple calculation on Transit enhanced diffusion and then I want to finish the lecture spend the rest of a Time on the effect of teed on devices now that we have these models for teed we can go back and revisit the reverse short Channel effect and talk about that in more detail okay so let's go on to um slide number two and um what I'm showing here these are uh some actual uh Supreme 4 outputs output profiles modeling in imp plant profiles and just to show you a practical example of how you might use Supreme for war along with a dual Pearson a dual Pearson means you take two Pearson 4 distributions and you add them together and you can use this in a quasi empirical way to simulate the ion channeling tail so there are four different plots here four different panels look at the first one marked a these are all for the same condition uh in terms of energy this is all uh Boron bf2 implant and the energy the primary energy is 65 Kev uh and then what we're looking at in each panel is a different dose so before we get started with looking at those let me just make a note uh the total uh atomic mass of a molecule of bf2 is 49 you just add up two flines and a boron you get Mass 49 the mass of boron is 11 so we as I mentioned last time people sometimes use bf2 to uh get a lower energy or more easily am morphies the Silicon for example look at this uh little simple calculation 65 kolt bf2 is equivalent to 65 * 11 over 49 so you assume that the energy is partitioned according to the ratio of the masses uh uh for a boron implant so if I do 65 K bf2 it's the same as if I implanted the Boron atom itself at 15 kV so in especially in the old days when it was very very difficult to build ion implanters that could ion implant at low energies now they have implanters that can get you down to one two 3 Kev but um the beam current tends to suffer when you go that low so instead of doing very low energy Boron people would do a higher energy uh molecule bf2 um so you'll often see bf2 uh implants and so now let's go ahead and look at these uh these bf2 uh uh implants and there a couple things being shown here the plus signs or the symb are the actual data and these data was taken from alash tash's work back in 1989 these are actual Sims profiles um and the um the lines the smooth lines are Supreme simulations so for example in in part A or our plot Number a we have a dose that's very high of 515 and so what what that means is throughout with such a high dose bf2 throughout most of the implant of the time of the implant you're really going to be amorphized you're going to be imp planting into an amorphous uh uh Crystal uh because it's going to Amor the bf2 will amorphize the Silicon so this is the profile that you get you notice it looks pretty much like a standard Pearson 4 and there's this little small exponential tail but that tail doesn't really become obvious or evident until you get about one two about three hours of magnitude down from the peak and then you start to see the tail so the way this profile is constructed is is what dual Pearson means it's two Pearson fors added together so there's one Pearson 4 here which is with this this dash line that's labeled as the amorphous profile and then there's a second Pearson 4 here which is has also unfortunately a dash line but you can see it has a slightly different slope and that's labeled the channel profile those two Pearson fours are added up with a certain ratio this ratio here in this case is 969 and you get the total profile and you can see the simulation uh obviously if you pick the right number for the ratio the simulation then fits the the entire profile including this region here at the near surface and the amorphous and the tail rather the channel tail let's go to a lower dose say one and a half E15 now you do see again we see two Pearsons this one here and the one that's associated with the channel profile but now the channel profile is a higher fraction of the total and you see this um exponential tail coming in a little bit more uh more obviously and if you go down to panel number C or plot Number c um the dose is even lower 4E uh 514 not that much amorphization going on so you have a large very prominent Channel tail so the second peeron is really dominating and finally at a low dose of 2e3 there is no amorphization so the implant that really dominates in the to dual Pearson is primarily that associated with the channel profile uh and the ratio is actually zero so it's just just the channeled one so that's how dual Pearson works you can see it is two peeron fours added together um one of which takes care of the channel profile the other which which takes care of the Amorphis and you add them together depending on uh a certain number called the ratio so it's it's quasi empirical um so that was for um uh bf2 let me just show you some uh simulations these are uh the different symbols here are not data so I apologize these are actually different symbols represent different Supreme simulations just to give you an idea when you do a simulate depending on which model you choose you'll get slightly different profiles and it's up to you to figure out by comparing to data or by looking in the literature which one of these is closest to the truth uh for example just as a comparison if you look at the open circles here the one this profile marked gaussian um it's a little bit hard because there's a lot of symbols but there's an open circle profile it's it's quite symmetric as it has to be because it's gaussian it has no Channel tail um if you look at the um open uh open triangles will it's the Pearson 4 a single Pearson looks a lot like the Galaxy not much difference a little bit skewed um the Dual Pearson is the uh these open boxes and it does a little bit better job you would think well we don't know actually we haven't seen the real data here um it has the channel tail uh Incorporated in it and look at the Monte Carlo now presumably the Monte Carlo has the most physics built in uh into the into the simulation and it looks a little uh noisy because of course we only followed a certain number of ions but in fact the Dual Pearson looks reasonably close to the Monte Carlo the mon Carlo is attempting here uh and to include a little bit of the uh of the ion channeling so you get different different models give you different answers in Supreme and you have to figure out which one suits your uh particular application best so now what I want to do is I want to take a few minutes here in class and uh if we go on to slide number four in the handout and this is also the first page of your of your handout number 26 if you want to look at that they're identical and what I was hoping you could do is get together uh as a group there's there's not enough people here to really split up into individual groups uh because everyone again sleeping in because of Boston one um is to kind of go through this and make some simple calculations for the next five or 10 minutes uh together on how you would um do the simple calculation let me just read through it before before we start um so we have an engineer wants to form a shallow Boron a dope Source drain for an advanced technology and the question the manager is wondering whether to buy a batch furnace uh or um uh to use a rapid thermal anal and the furnace andal they are considering for this implant would be 800 Degrees for 1 hour so that's relatively long time at low temperature or should they use a rapid thermal processing machine or RTA at 1050 for 1 second and the implant that they want to activate is boron and it's 20 kilovolts relatively low energy 20 KV and a 5 14 dose so what we're asked to do here in a is to make a rough estimate uh you know using a square root of DT kind of estimate of how far the dopin move during an 800 degre c 1 hour a meal versus 1050 for uh 1 second um again it's not necessarily Gan diffusion but you can calculate the root DT and this is without te effects and this is pretty simple because I've printed at the bottom of the slide the intrinsic diffusivities under equilibrium conditions D at 800 is this number and the diffusivity of boron at 1050 is this number so we have those numbers written right there so part A is pretty easy now Part B what we want to try to do is include the effects of teed so we want to use the charts that we handed out last time and in fact they're in handout number 26 if if you didn't bring your old handouts I've included them again in in handout number 26 um use these charts to to figure out the expected enhancement of the diffusivity remember we said the diffusivity is now the equilibrium D star times CI over CI star that's the effective diffusivity and how long teed lasts to calculate the real square root of DT uh assuming uh including teed effects so these are the two things I'd like you to uh to try to do and I was hoping you could kind of work as a uh as a team on this maybe get together with a couple of folks who are near you uh somebody here hopefully has a calculator if uh if you need one to do this I'm not so interested in exact numbers I just want to get an idea of how the answers um play out so why don't you I'm going to give you five or 10 minutes to get together and work on that and uh we'll stop the lecture now and then we'll come back and and we'll see who who has an answer that that looks reasonable and we'll kind of talk it through okay so that let's uh we can stop the recording at this point as well because it won't be that interesting to uh record you um punching your calculators but 1300 what can you say about the first 1300 seconds how what's the how high is the diffusivity during the first 1300 seconds it's going to be 7,000 times than it is in the the next 1300 seconds or or 2,000 seconds so who cares in some sense about the next 2,000 now if if this instead of being 3600 seconds was you know you know orders of magnitude more seconds yeah then at some point you actually have to take into account the normal diffusion so you you don't completely ignore the actual time the actual time that the thing is in the furnace is 3600 seconds it's just that for on the order of almost you know a third of that uh it's enhanced diffusivity by 7,000 so we can ignore the rest of the time but you you want to keep that in in the back of your mind so then did were you able to calculate um either a DT enhanced or um or Square Ro of DT enhanced either one what' you get for that the DT that you got okay I end up somehow with h 3.5 * 10us 10 cm squared but doesn't me I I multip oh I multiplied this so if we multiply the ordinary diffusivity time that I got 7,000 * um 3.7 * 10 -7 so the actual diffusivity I got was 2.6 time uh 10us 13 cim squ per second during the transit during the time when the diffusivity was enhanced and then multiply that by 13 oh you know my my mind's numbers prob I had 1333 here maybe I had was carrying of few more significant digits num was in the root oh root TT okay so for the square root you got what 18 right 10 the minus 5 so in angstroms to put it on the same that's 1860 just to put it on the same units so not even close in the intrin intrinsic diffusion so really it's going to the the uh Tedd clearly dominates in that case and there's quite a bit that's quite a bit of motion now that that you can easily see on a Sims profile and that could definitely affect your device performance a junction depth uh difference of that much so how about at 1050 exact same formulation uh applies but the nice thing is how about what can we say about CI over CI star at 1050 it's a lot lower right I get about 5 50 so the enhancement of the diffusivity is is only 550 times larger that's the diff and and how long does it last teed at 1050 if you go back to your magic curve phosphorus doesn't look like it's going to last very long again you have to multiply by five and by 8 over 06 the RP ratio so the amount of time it looks like teed lasts under this dose condition I got about 67 seconds 2/3 2/3 of a second basically 2/3 of the anal remember the total inal time was 1 second so again the last one third of the anal I'm I'm going to anal I'm going to ignore just because this this is still a large enhancement factor it's 550 so I'll ignore it if it had been 1050 for hours then obviously we can't ignore a normal diffusion okay so then in that case the root DT at 1050 I ended up with something about 420 angstroms is that close to what you guys got 4.2 * 10us 6 again putting it putting it uh back onto all onto angstroms so again we see this kind of interesting you know anomalous and maybe initially somewhat non-intuitive result here we have uh at higher temperature admittedly in much shorter time it's only a second um we can get uh a lot less root DT a lot less motion or broadening if we do a 1050 rapid thermal eal compared to putting in the furnest you might say oh 800's really low it should be a safe temperature it's only you know a couple of uh angstroms motion but in fact you're much better off for this implant doing uh according to this calculation uh of Ted you're much better off doing a 1050 anal uh for for a fraction or for a second um so that's just sort of give you a feel for where the numbers come from and um you know how the how the whole thing works I think once you work through an example like that you have a much better feel for Te okay good well that's that seemed like everybody had a good handle on that so let's go back to the uh regular handout hand out 25 and I think you have um you've got a good feel for how simple calculations and how powerful it is just to use those two or three charts with with two or three charts you can you can say a lot about uh rough back of the envelope calculation for Ted anything anything more sophisticated than what we just did in the last 10 minutes you probably should use be using Supreme at that point or some simulator I should say all right let's go on so um uh just to remind you ourselves where where those charts just came from if I'm on uh slide number uh five now on the handout they they came from the this observation that was made at B labs and confirmed by a lot of different workers that the um time scale of teed was the same as the time scale of the shrinkage of these famous now now famous 311 defects and that the at low temperatures the 311's hang around a lot longer than at high temperatures and of course that's why Ted um lasts a lot longer uh at low temperatures okay so um just for to remind ourselves and from the little hand calculation we just did this is exactly represented of what we just did in class here what we the general picture of teed that we showed last time we have a certain enhancement Factor CI over CI star Max again that's that's a ballpark that's the maximum enhancement so we assume it's just a constant enhancement throughout the entire period of the steady state while the 311's are decaying and then we suddenly say there's a rapid exponential decay right after T some period of time called tow enhan that we just calculated in our example so that's the simple model the critical parameters that determine teed well we just the amount of teed we just said was the superat saturation level which we write as I over I star or CI over CI star and we know that's a function of temperature and this is the functional dependence or you can read it right off that plot and the duration of the steady state condition which is the so-called towel enhanced time and now that tow enhance depends linearly on dose and doses have a wide range so this is the key it depends linearly on Q so if you implant 20 times or 100 times more dose you get 20 or 100 times more interstitials um so so it last last that much longer so notice the dose unintuitively somewhat the dose doesn't determine CI over CI star the dose determines how long the thing lasts how long the transient lasts and there is there is a dep certain small dependence on RP and the energy dependence comes from the dependent in the equations the energy dependence is represented through the dependency on RP okay so that's the general picture and I think that example helps us understand it now let's look at anything a little more complicated I said we should we should be using a simulator so the next couple of slides here on slide seven starting with slide seven I'm going to show you some Supreme four um examples um using sub more and more sophisticated models so the first model I'm going to use Supreme 4 using a gaussian implant assumption which of course is not very sophisticated but and I'm going to be kneeling this um uh arsenic implant at a th000 degre c for different times and uh if you look at the the open boxes that's my initial ion implanted Arsenic and what's being shown here is a series of different curves different symbols for different times and what the model that was used in these particular simulations is called the fairy model in Supreme when you do your next homework will there'll be different models you have to invoke the one here is called fmy as the name suggests it takes into account the fmy level and therefore the concentration dependence of the diffusivity it does not take into account any Ed so when you run uh when you tell it to use PD method equals FY um you're not taking into account Ted so this is just normal ordinary diffusion and the junction depth proceeds according to a square root DT type of Behavior Uh as you would expect it does take into account the concentration dependence and you see the box- like profile as a result now we go on to a little more sophisticated model um this is that same implant 30kv 414 but now this is on uh slide number eight what we're doing is we're using a Monte Carlo implant model that changes the aen planted slightly uh but the reason we use that is because the Monte Carlo model in Supreme keeps track of the damage because it not only does it track the incoming ions but every time an a silicon gets displaced it can keep track of that uh so it can it can keep track of the damage and then I'm we're using a model in Supreme called the fully coupled or the full coupled model it takes into account these following uh factors the impact of the FY level it also takes into account point point defect injection and 311 cluster dissolution it has a model there for 311's and and their kinetics and it also because it's fully coupled defect gradients can drive diffusion so if there's a if there's a two-dimensional uh situation like in the reverse short Channel effect you can actually uh the gradient and the defect can drive diffusion but this is what you would see if you if you do this fully coupled model look at all these these first four curves all lie straight on top of each other 30 seconds to 120 seconds they're all the same now why is that well I mean essentially it's because of teed because um whe whether you're 30 seconds or 120 seconds there's some portion of to enhan there's a certain enhanced time and during that period CI over CI star is so large uh because of this the amount of 311 that end up being formed that it doesn't matter uh whether you're doing 30 seconds or 120 um you still get the diff Fusion is dominated by whatever happens during the the transient when the 311's are evaporating and then after that if you only if you go to longer times so say 600 seconds 12200 and 1800 seconds do you start to see normal diffusion being large enough because the time is large enough that normal diffusion now overtakes the teed and you can actually see that and so this is an example of why people didn't discover teed until they had rapid thermal annealers because in a furnace you can't anneal for very short so you could never access anal times that were this short and you would never would discover that in fact teed was going on so a signature of Ted if you do an experiment uh on a kneeling an implant a very clear signature if you do Sims uh on some different times is you see very short times they all look the same profile that's an immediate signature aha there's some kind of teed going on some very short transing is dominating all the diffusion and then you'll you'll go to long enough times that you'll start to see normal diffusion taking place and this is these are Supreme simulations there's no actual data here the the different symbols are not data I apologize it's for different the symbols are for uh the different uh simulations okay so that was a couple of examples let let's go on to slide nine I I I want to give uh we're going to go back now now that we've talked about teed and we understand the models in much more detail I want to remind ourselves why we want to to limit diffusion and then talk about some particular device impacts of teed well we already know we want to keep Junctions shallow these Junctions here XJ either in the Deep Source drain and particularly in the shallow Source drain extension has to be kept shallow in order to do gate length scaling uh we need a steeper lateral Junction so in this direction in in the L direction we needs the Junctions to be very steep uh in order to lower the series resistance uh at a given effect Ive Channel length we need a very small under diffusion under the gate uh to to reduce the overlap capacitance if this underneath the gate uh this diffuses too far under the gate then you'll have a lot of overlap capacitance and that's going to uh slow down the the circuit speed we talked about needing a retrograde well in order to get better Mobility while controlling short Channel effect we need this retrograde well in depth we need to retrograde the profile and decrease it towards the surface uh and there's these f see little um angled Halo implants these Halos here are shown in the bright red these are implants that are done uh um in the uh at an angle uh to the gate in order to put the dopen just right where we want it uh and we if we we're using this angled implant to obtain the right profile we'd like to not have that diffuse all over the device because then it kind of the uh we don't get the shape that we want so all of these reasons we need to limit diffusion um so I want to talk about now that we know teed uh and go back to this reverse short Channel effect that I uh introduced uh three or four lectures ago and I think we'll have a better understanding this time if you want to uh learn a little bit more about it there's an article by rafy in iedm of 93 or Crowder in iedm of 95 where this was first explained uh and what people were trying to explain is the device physicist in the early 90s were all finding if they PL and the circuit guys if they plotted their VT as a function of the the gate length uh of the transistors um on a chip or or on a wafer what they saw what they expect is the normal short Channel effect VT is fairly constant and you get to short channels and the VT is supposed to roll off due to well-known electrostatics what they found instead there is so-called reverse short Channel effect it's revers to one's expectations in fact the VT the threshold voltage on these devices was going up as you made them shorter and it was going up and peaking quite a bit and then eventually the normal short Channel effect would take place so this so-called reverse short Channel effect was quite bothersome in the early days because people didn't understand when they scaled to these Channel links why VT would be going up how could it be well we talked qualitatively about this um the reasoning for this several lectures ago let me review the qualitative and then we'll do the more quantitative this is an actual simulation from Supreme of what's going on in the reverse short Channel effect so we have our gate uh you see the the side spacers this little region under here under here is is the um Source extension this is the drain extension and the Deep source and the Deep drain and what's happened is that we have implanted these sourc drain regions and their extensions we implanted say with Arsenic and we've generated a flux in fact these arrows are supposed to indicate the interstitial fluxes so these are silicon intertial atoms that are diffusing um and in fact and they're then recombining they can recombine com in the bulk or they can recombine in the surface uh and because of the way the recombination goes at this surface they end up creating a flux that goes like this and now that flux um of uh and the gradient of the interstitials is so sharp that it actually can drive uphill diffusion of the Boron and you can see that uphill diffusion you can't see it in this but if I take a cut right through the center here at x equals 0 and and plot it versus depth so Boron concentration versus depth at the center of the channel um if you have a one micron device on the chip it looks like this if you have a 0.8 Micron device on the chip it looks like the the green curve and what do you see in the green curve well for one thing the surface concentration of born's a lot higher than it is in the one micron device so that explains to the Circuit designer oh I have a higher concentration of Bor on the surface that makes my VT higher so that explains the VT effect how it got to be so high and how did look at even the peak of the Boron profile in the 0.18 Micron device even the peak moved towards the surface again that's totally non-fickian diffusion if I give you gaussian diffusion and you diffuse it in your calculator the peak stays put it just goes down it doesn't Peak doesn't suddenly move over to the left to the right that's not gaussian well what's pulling this peak over is the fact that there's these these interstitials that are that have a gradient and they drag with them because remember Boron likes to likes to diffuse with interstitials it's diffusing as a pair because of the fully coupled diffusion they drag with it the Boron Peak um so uh this is the qualitative explanation at least of what's going on in the reverse short Channel effect and now let's go through a little bit more carefully the articles by Rafer and Crowder and you can see what it is that they um how they explain this a little more detail um these are some of the key process steps that influence the channel uh profile so the reverse short Channel effect ends up all up about about being what what determines the channel profile so step number one uh we know we do an ion implant and it has some shape this is supposed to me represent if you turn your head sideways the shape of the Boron implant so we do an implant and a couple of energies it Peaks here and then there's another Peak down here that's what we expect it to look like in the as implanted case step number two in making the device well we grow a gate oxide all right there's a certain amount of thermal budget associated with the gate oite 800° whatever it'll diffuse a little okay we can understand that that's normal diffusion step number three we put down a gate and we pattern it okay so it has a pattern like that usually polysilicon gates are put down at 600 not you know 500 very low temperatures not a whole lot of motion going on all right now uh the step number four is called the lightly dope drain actually we don't use that terminology anymore from devices we call it the source drain extension it's this shallow Source drain right here that gets an i imp planted and it gets masked by the gate so when you're doing step number four the side wall is not there Step number five hasn't been done yet so you have the shallow Source drain extension so there you're injecting a certain number of Point defects okay now I take that I've injected these Point defects and now I have to form the sidewall spacer depending on what material I form the spacer of if I form it of silicon nitride that goes down at 800 for about an hour uhoh bad temperature 800 for about an hour we just did that calculation 800 for about an hour you can get a a lot of teed you can have a 7,000 perhaps enhancement in your diffusivity so you got to watch out for that that's why nitride spacers are a little tricky uh if you do a low temperature oxide deposited spacer you can do that at 400 so it's not so bad you don't have to worry so much all right so there's a possible indication of problem and now we do number six we do the Deep Source Trin implant again that's probably going to be arsenic pretty high dose 10 to the 15th not a great thing in terms of teed I'm introducing a lot lot of uh Point defects that can then come in here and enhance the Boron implant uh the Boron uh profile so I just wanted to give you the order of the steps so you can understand how all these things fit together to uh determine the channel profile and if you want to go a little more sophisticated into raf's iedm article that's being shown here on slide number 13 of your handout um what he calculated here so this is the edge of the gate this is a two-dimensional plot out of a a two-dimensional simulator and these are Contours and what he's he's done is he's ey implanted a certain amount of um of uh damage or or uh dose into the into the drain region and he's calculated something called the time integral of the super saturation ratio so it is essentially and he calls it the enhanced time but it's it's it's the integral of CI over CI star integrated over Bal um and and he's plotted in terms of Contours so uh these are profiles of damage caused by this um uh shallow sourc drain implant and um he represents this damage by this integral of CI over CI star um called which he calls the enhanced time so to speak um and a number here that corresponds to 10 to the 6 has a units of seconds so it's as if you did a 10 to the 6 second um anneal essentially uh and that corresponds to a super sat an average Super saturation of about uh a factor of a thousand if the real time if this this were really the real time is about 20 minutes let's get cut off at the bottom here 20 minute real time is equivalent to 12200 seconds so there's huge amount of enhancement but look at these Contours look how they look how they go down um you're going from 10 to the 6th here down to 10 to the 5ifth in this range so again I have a large gradient in this CI over CI star and the gradient um is pointing me pushing me towards towards the gate and towards the the U the gate oxide don't forget he's using um saying that the oxide interface under here acts as a sync for a recombination of interstitials okay so this is the origin of the uh reverse short Channel effect as he explained it on P slide 14 the implant damage from the the shallow Source drains sets up a retrograde CI over CI star profile under the gate and the gradient in this profile so this grad of of ey over ey star because again it's diffusing as a pair uh results in a uh an extra flux in the Boron diffusion that wouldn't be there if it were diffusing alone but because it diffuses as a pair with interstitials so that retrograde causes a boron pile up at the interface and the shorter the channel the higher the pile up because you as you make the channel shorter you bring in those Source drains closer and closer uh into the center of the channel so that explains the reverse short Channel effect why does the the uh the Boron pileup gets bigger and bigger and therefore the VT goes up and up for a shorter device oops and in fact here's an example if we go on to slide uh 15 from his article he is this is boron concentration simulated versus depth now he's doing this at the center of the channel and say a long Channel device looks like this here's the Boron he simulated it basic basally has the as implanted shape these are they're two implants there's a a low energy implant you can see its peak and a slightly Higher One so this Boron at the center looks sort of like this now if you take the Boron um uh that's the center of a two Micron long Channel if you go to a0 four five Micron device it looks like this one there's a huge pile up at the surface and the peak is close to the surface so the surface doping could be three to five times different um when you have a very short chall device compared to a long Channel device because of the influence of these Point defect gradients that get set up and so these very different surface dopings are can then be used to calculate the VT difference between these devices and that's exactly what rafy did and this is taken again from his article 1993 um he calculated here um these uh the what he based on those profiles what he thought the threshold voltage should be as a function of one over the channel length or if you want to read the channel length on the top axis for different uh biases and if you look at his um calculations here this smooth um solid line is including the teed effect he predicted that the channel length would go like the um VT would go up like this as I go to Shorter Channel links and the data these are the experimental data they obtain from devices at Bells indeed you can see the VT going up so look at the VT for a two Micron device is over here uh the thresh voltage is about a volt and for a 045 Micron device which is I think right about here the VT is about 1.3 volts and that agrees very well so that rule up of the VT uh is agreed very well with the simulation when he included the Ted the dash line is a simulation without the teed and indeed uh you you see the normal short Channel effect going down like this so the only way people could really understand the reverse short Channel effect was to uh was to really understand in detail what was happening with the Boron Ted and the influence of the damage on each side of the channel on what was going on inside uh underneath the gate so that's kind of a famous paper on on how um understanding these process models end up influencing the device model which ends up influencing the circuit uh the circuit model so there's a lot been a in simulators there's been a certain amount of work in the early 90s on how to model reverse short Channel effect what are important param well obviously the magnitude of the initial implanted damage well you know the dose of the source and drain extensions and the source drain so that's important but so but you need to to figure out exactly how much I over I star there is we did it in our book our textbook using a very simple analytic calculation we calculate the maximum CI over CI star remember that was a ratio of K reverse to K forward in that equation and we put some estimates down for that but it's not clear exactly how accurate so depending on your Simulator the way they calculate it will be slightly different so the these are different clustering factors in this simulator you can they have a parameter called the cluster factor that will adjust essentially the CI over CI star Max and um depending on how that is that CI over CI star Max you'll get different amounts of reverse short Channel effects so the the uh I apologize here the vertical axis here is VT that got cut off somehow so this is the threshold voltage and this is the gate length calculated and you can see in this particular simulator this is not Supreme this is a Sako uh simulator but um for a cluster factor of zero there is they didn't predict any reverse short Channel effect when they allow clustering and they allow these 31 ons to come in um they you do see a roll up a reverse short Channel effect and depending on the the magnitude of the cluster Factor reverse short Channel effect can be more or less uh prominent uh so there will be usually depending on the simar you use there'll be a couple of parameters one can tweak to affect both the initial implant damage and the recombination rate uh this is a two-dimensional effect so not only how much damage and how many 31 ons do I end up with and how many inter what's the interstitial uh concentration that's important but how those those fluxes diffuse and how they how they recombine at that oxide interface will determine the actual gradient of the interstitials so the K Subs factor is also an important the combination rate of interstitials at a at at the gate oxide interface in the channel and that's a parameter that one can adjust hopefully it's fairely well known but one by adjusting that parameter you can change these curves as well uh aeling temperature is important as you know again this is from that silvaco simulator as well again VT versus gate length and the different color curves here are for different analing temperatures and as you might imagine low temperatures give you more reverse short channnel effect because we know at low temperatures teed is more prominent the CI over CI star hangs around longer the the the time of the enhancement is worse so here at 850 in the red you see more a little bit more of a rollup then if you were to anal at very high temperatures uh in this particular anal I don't know what the dose was uh there wasn't much reverse short Channel effect so lower temperatures um tend to change that and that's just because of the CI over CI star uh term as well as the tow enhanced term uh uh depend on temperature uh and in fact the next uh slide in your handout slide number 19 is just that same plot that we've been us we've used to do the example a few minutes ago uh to remind you that CI CI over CI star is increasing as we go lower in temperature and that's why you see more of a reverse short Channel effect uh as you lower the temperature so not only does the VT change but just to give you an idea other device parameters that can change depending on the channel profile that Boron pile up at the surface underneath the gate can decrease the channel Mobility Channel Mobility at very high concentrations of channeled doens you get more scattering of the electrons more scattering of the carriers so the mobility can go down and in fact this is a plot of a calculation of the mobility that raford made in that article here on on slide 20 so he's calculated the mobility as a function of uh the channel length uh for two different doses thist dose the EXT stands for the sourc drain extension so that's the dose that gets implanted right next to the gate right after you cut um and Define the gate uh poly you implant the sour drain extensions H and here this is for a 312 a relatively low dose not that many interstitials are injected the blon profile looks pretty much as implanted and you don't get much so the mobility stays pretty high until you get to very short channels if you do an extension dose here that's about uh uh three times that say about 812 what happens is you you you get this reverse short Channel effect you get the Boron being drawn to the surface a very high amount of boron in the channel means you have a lot of ionized acceptors and the electrons feel those ioniz exors they get scattered by cool lumic scattering and in fact the mobility then can go down so for a higher extension dose different profile Clauses pile up and you get a lower Mobility lower Mobility can mean you can end up lowering your current Drive compared to what it should have been um so that and that affects the the overall circuit speed so again just just subtle changes in the in the channel profile which have nothing to do with how I implanted the Chann it's how I implanted the the neighboring regions the source and drain end up impacting not only the VT but the mobility um of the device and therefore the current Drive oh this was a neat paper now several years later uh 1995 Scott Crowder Crowder came up with um an interesting idea knowing what he knew about 311's and what we know in our class he said okay well 311's they kneel out a lot faster at high temperature so let's say I have to do an 800° step because I'm going to have to put down those nitride spacers maybe I can still do that if I do before I do it I do a th000 dee short anneal so let's do a very short anal 1 second at a thand evaporate all those 31 ones and get rid of them then I can put the wafer in the in the furnace and put down my 800 degree nitride all the Ted will be over with so I use a high temperature to cause the 311's to evaporate get rid of all those excess interstitials in a very short time um I should have less of the reverse short Channel effect less total interstitials and that's what in fact where he showed the um this star the starred region is when he did the high temperature rapid thermal anal first before he did the 800° C longer time anal and you saw he had less rollup less reverse short Channel effect when he did um the uh the open squares represent the case where he did the 800 degre c um anal in the furnace first and then did the thousand and you see a lot of reverse short Channel effect interesting exact same amount of time in the furnace and in the RTA in both Wafers he just changed the order of the operations so this is kind of interesting that you know what we know about 311 defects and how they how they anal it's important for us to think about the order of the steps in which we make a mosfit where we insert the the anals um now you don't always have a choice you have to cut the gate before you do you know the the sourc Trin extension so it's self aligned and all that but this was kind of a clever experiment just to show that uh the process uh order can have a big effect because of things like 311's um and again this is just a reminder we already saw this several times on slide 22 of what the amount of time and what he did was this thousand degree 1 second in eal he popped it up um and so he can get basically within a second or so he can get rid of all the 311's uh the tow enhance is only a few seconds and then he could go on later and without any 31 ones around put it in the furnace or with very few and and go down to 800 so that was that was his proposal another thing that he showed in that paper this is from that same paper by Scott Crowder iedm of 1995 he did an interesting comparison he also compared devices made by very similar processes on bulk so on regular choski Wafers to the similar device made um on uh an SOI wafer remember we said there's this technology called SOI where you can have um single Crystal um silicon layer on insulator very high quality material um and you can make mosfets in that material uh and what he did was um now he found when he did the source and drain uh implants indeed you of course even in a silicon insulator you do inject interstitials up here but interestingly remember um the interstitials tend to get injected both down and then they go up the ones that went down were going up here now against an insul an interface between oxide and silicon and it turns out that interface is a very good Sync It's a good Rec there's a lot of Rec combination that can take place uh at uh at this interface Cas ofs is fairly large between oxide and single Crystal silicon um so a lot of interstitials can be sunk or can be absorbed there and uh whereas the um in the bulk we get these interstitial fluxes that come in and they don't get absorbed because there's no there's no oxide down there they they can just you can end up getting profiles that look like the fluxes that look like these arrows driving the Boron to the surface so what he what he show was in fact on an S SOI wafer you don't get much pile up as much pile up of boron underneath the gate and on thei wa for um this plot the y- axis didn't get showed up here but the y- axis is the threshold voltage VT again the roll up of the VT is a lot less on an SOI wafer subject to the same kind of a kneeling compared to this uh bulk wafer shown by the solid line which had a lot more reverse short Channel effect interesting idea use S SOI use the property of the fact that there's a lot of uh there's a Sync here for interstitials to sync out a lot of the interstitials that you in implanted um and and get less uh motion of the uh the channeled open so this tells us right away though if we're doing a process in SOI um you know and we can do the exact same process we can get quite a different result in insulator uh in SOI compared to bulk because the uh the channel doping profiles will not be the same uh and so the simulators need to be able to simulate these effects um and another thing I want to talk about is uh I just want to remind you what's the usual order in which we form the channel the gate and the source drain uh in in a moset and and how people are we we are doing clever things with the order of things um this is what uh this is sort of a a cartoon in PowerPoint and uh so this is supposed to be my wafer um these green regions here on the left and right are going to represent the um the uh shallow trench isolation so green is my STI it's just my isolation region and then typically after you do shallow trench you implant what they call SSR super steep retrograde it's just in implant regions if you're making an nmosfet you have usually a shallow implant up here uh or more lightly doping near the surface and you have the peak of the implant a little bit deeper um so this could be a boron implant so that Boron implant usually goes in pretty early in the process and it's that Boron implant that's going to determine your channel profile and therefore your VT and things like that so it really usually goes in fairly early right about here after that it sees the thermal budget of the gate oxide growth that could be typically 800° so it's got to go through that diffusion and then you make the gate the gate is usually a very low temperature process and it's just ing so that's no no no motion of the Boron there then you implant the shallow Source strains and you use the gate as a mask so now I'm doing implants of Arsenic and you're introducing a little bit of um certain number of Point defects here on the left and the right of the channel now I put it in the furnace and I per spacers these green spacers if they're oxide I do it at a low temperature lto goes down at 400 it's not usually a problem if they're silicon nitride spacers typically nitri lpcvd goes down at 800 so I could get a fair amount of diffusion at 800 of teed especially because I have the implant damage introduced from the shallow Source in Grain so watch out for nitride spacers 800 for an hour to make these spacers could really as we saw in our example could really cause a lot of motion at that Boron then we do the Deep Source strains using the spacers as the mask now again Sant uh uh and then usually right after that there's a final thermal anal so one idea people have had is well don't put the ptype SSR implant in at the beginning why don't you put at the end of the process and a very radical idea is to put it in even after the gate has already been formed U this is not being done in production but it was a neat idea that people had in research put okay you you have to deal with all these Point defects well don't put the Bor on in until the until you've already KN knel out a lot of those point defs so here's an alternate process give you an idea how device Engineers were trying to to get around Ted to a certain extent uh on slide 25 an alra process integration scheme for forming a mosfet and this was published back in 1998 by Phillips uh Corporation um called Channel profile engineering 0.1 Micron mosfets by doing through the gate on implantation so they were publishing uh U proposing a flow that goes like this a conventional flow looks is on the right conventional usual sequence to make a mosfit is you form the P well you implant the channel Bon I profile all the way up here then like we just said we do gate oxs we form the gate uh we make spacers we form the sourc strain extensions um uh oh I'm sorry that's that's some oxide deposition here's the sidewall spacer here's one of the killers um the nitride dep at 800 a lot of teed can happen there then the Deep sour strains and then we do the RTA so what they were saying is instead of putting the channel uh implants in here where they can diffuse during this nitride spacer step take the channel implant and put it in at the very towards the end after you already have the nitride spacers in the thing that's weird about that though think about it now I'm I have a topography that looks like this now I'm going to implant the channel I have to make sure I give it enough energy to Boron so it can get through the gate so you have to calculate that energy and in fact your Boron Pro profile now um is going to look sort of like this it's going to have this shape to it it's going to be a little deeper here where the gate doesn't exist in the source and drain and it'll be a little shallower here that might not that might be okay from electrical point of view um but it is kind of strange but the advantage they have is that it doesn't go in all the only thermal step it sees is the last high temperature step um it never sees all the T that would happen during the the um sidewall spacer at 800 so that was an idea they had one thing you might think though what what do you think about if you're electrical engineer what do you think about Ina planting through a gate oxide it's a little scary because that gate oxide could might only be 20 anrs thick right either an implant a high energy ion through it's not clear what damage takes place right at the interface between the oxide and the silicon and interface States and things so although this was a neat process I don't think it was ever accepted in production I'm not sure people thought it was reliable enough but they did show and on on the next page they did show um in their iedm article that they could achieve a dramatic improvement over the Boron control this is dopent uh Boron concentration versus depth and um the black line is the reference device so that's the device that went through the ordinary flow where they put the boron in at the start of the of the process and it goes through everything it sees the sidewall spacer nitride depth all that 800 lots of Ted the Boron is essentially almost flat you don't even see the much of the implant the initial implant whereas when they did the through the gate as implanted and after processing so look at after processing here um you can see there wasn't that much diffusion at all so they were really able to control much better because it didn't see any of the teed from the nitride uh all it show saw was the last high temperature th000 degree step um but as I say for reliability reasons I don't know if it was ever really accepted they did show that they could reduce the reverse short Channel effect um this is now on on the slide 27 uh BT as a function of gate length and they have different um different uh things here here's u a reference uh process the black that's when you put the boron in at the very beginning the standard process flow you see the rollup the reverse short Channel effect and through the gate is TGI and they had several different doses TGI you see you see no roll up in VT the VT is very flat until it goes to the conventional short Channel effect sort of effects um so they were able to eliminate the rollup uh because they had better control over the of the profile they essentially eliminated a lot of the teed in the Boron profile just want to mention uh before we finish um chapter8 some new diffusion modeling issues that are in literature right now that are people working on today um uh the conditions where we have a very high dose being implanted the 311 model that we talked about um is not is is good in intermediate dose regime but doesn't really work for very high dose doses so we need to model the the type of damage that takes place in very high doses is being investigated by people today uh very wide energy range there are people doing implants less than a kilovolt today there are some really crazy people trying to do implants very shallow and some very uh deep ones even greater than a megga volt the physics of stopping and the physics of how to uh model those implants is not really all that well known in in this in these two ranges uh we talked about pre amorphizing the substrate prior to in producing Boron we said well hit it with silicon at a high enough dose that you can amorphize it and then you can avoid channeling um but what kind of damage is produced by a pre amorphization by uh an amorphizing implant uh and how what effect exactly it has is not all that well understood there's a whole bunch of new analing techniques that have come out something called a spike anal where you take a RTA machine you zap up the lamps really fast and you zap them down immediately um and the wafer never even spends time at any one temperature it just sort of goes up and down um you're accessing sub one second time machine the KET exactly what happens during those ramps is not completely understood in Spike analing laser analing where we take a laser and we scan it across and uh the we only heat the area for a nanc again the kinetics of defect evolution in the nanc regime has not really been very well understood so that's a very hot topic these days um so that's process conditions are changing from what they used to be new mechanisms well people already know 311's that's kind of been beaten to death I would say at this point a lot of work has been done but end of range dislocation Loops which we we never completely get rid of and these are very important if you do an amorphizing implant you can't avoid them their effect on defusion has never been completely understood uh the clustering of dopant with interstitials which may eStore a defect uh and may affect the electrical solubility of the dopent how many electron or holes you end up getting has not really been understood so that's that's a new topic and um at interfaces how what the recombination rates are in interfaces um as we put more different materials we're putting High K into the problem now what happens when you have a high K interface um how do Point defects um Rec combine at at an interface between a high K and silicon as opposed to sio2 a lot of interesting um new research topics that they're not covered and I won't go through the summary in any great detail I think we've have gone through all this um uh I encourage you to read through chapter 8 carefully and I think we're at a stage now and next time I'm going to talk about Supreme 4 in detail where we have enough tools that we can really understand to a pretty reasonably high level how people put processes together to make make devices and to minimize uh dope and diffusion okay that's it um if you haven't signed up yet on the clipboard I've got it up front uh I'd be happy for you to sign up what