 you can follow along with this presentation using printed slides from the nano hub visit www.nano hub.org and download the PDF file containing the slides for this presentation print them out and turn each page when you hear the following sound enjoy the show so this is lecture 38 and we'll be talking about modern MOSFET modern meaning the design considerations which are about let's say 20 years old what I told you is when I was born this many of these things when I told you and but now these things maybe when you were born and we'll see how it works so the first thing is the short channel effect you know the transistor is may getting smaller and smaller every day started with several microns of channel length source to drain nowadays 20 nanometer 20 nanometer I mean 20 nanometer is probably on the order of 100 atoms or 200 atoms it's very very small and how they make transistors out of it you know I have worked in this field for a long time still I cannot believe that they can make these things happen now I'll talk about control of the threshold voltage and now when you make short channel then how you lose control of the threshold voltage and how you bring back the control we'll talk about how making transistor smaller is no longer working very very much anymore and why people are sort of trying all sorts of ways to increase the current of the transistor by mobility enhancement and then I'll conclude so we have talked about the threshold voltage let me talk a little bit about the channel length and what effect that has on the threshold voltage looks like the threshold voltage should not even know about channel right threshold voltage depends on face of s the bulk doping why do I care about the channel length let me show you how it works that's your MOSFET on the left at the two blue regions are source and drain the bulk is this magenta or I don't know maybe light red colored region and a light blue region you see that's the depletion do you see why the depletion is two words more in the body rather than in source and drain because source and drain are very heavily doped do you remember that if you have a heavily doped source and a lightly doped region depletion goes into the the body region so substrate region so it's that's why I have shown the blue primarily in this in the bulk region now if you look at the threshold if you go ahead and measure this threshold as a function of channel length so you make various transistors of different channel lengths now starting from very long to gradually make them small you will see a strange thing and the strange thing is that first of all you might have expected that your threshold voltage will be equal to 2 phi sub f and if the bulk is n type I'm sorry the bulk is accepted type na then you put the corresponding sign over there for charge qb bulk charge q na wt what is the wt wt is the depletion right how far it has gotten depleted c ox you know how to calculate so you'd have gotten a threshold voltage but you see you'd have expected that this is there is no channel length involved here that the whole no matter of big 10 micron or 500 angstrom you make channel length you make threshold voltage you'll be measuring by the cv characteristics remember they will be all the same that's what you expect but if you really do this measurement what we'll see that that the threshold voltage will remain more or less independent up to a point and then it will begin to roll over and that's what is called the vt roll roll off very important problem now this is something we did not expect you didn't expect but this is a big problem and what I want to calculate is that how much is that and why that happens that's what I am after okay now let's see how how where that comes in now another thing is again for the short channel and I'll explain how the roll off comes in but the consequence of roll off there are many but there's another effect that when you apply a large drain bias then in the long channel you have this id vg characteristics do you remember sub threshold slope it cannot be seen is less than 60 millivolts per decade henrik shown got into trouble if he tried to do it less than 60 millivolt per decade there are ways to make it 60 millivolt per decade by the way but for the time being for semi classical transistors you cannot make it less than 60 millivolt per decade okay so you have that characteristics and if you make the channel too small then what's going to happen the gate will lose control because essentially what will happen that the current will just just like a bipolar transistor gate is trying to control on the top so he says okay fine the electrons pays respect to the gate on the top but sort of underneath through the depletion region it just bypasses the gate altogether as a result what will happen that your current even when the gate voltage is low but by just by using a large drain bias you can have a current flow which is not very good you see this is something very bad because we even when you turn the transistor off you'll have a certain amount of current and that's really power dissipation in your computer that's not good we don't want power dissipation when I don't want the transistor switching on and off right so even with vg equals zero and short channel transistors I have this punch through effect and I'm in trouble so that's something I do not want by the way do you also see the threshold voltage has shifted the threshold voltage is the point where about a micro amp per micron that current flows and you can see the rate has that same amount of current at a lower voltage lower gate voltage so that is what threshold voltage is so your threshold voltage has rolled off this is the consequence of it okay so where does it come from a few lines of algebra again very simple just pay attention you will see you see the gate essentially has control only over that part that part of the channel do you realize because what about the part which is directly under the rectangle of the gate but in the light blue the reason is that of course the source and drain voltage has more control over that region so the gate they are every point the gate and source are fighting for control every point on the drain side the drain and the gate are fighting for the control for the electrostatic control who controls it and of course in the gate therefore wins in some places where it's closer and loses in some places when it's further out now therefore the charge over which gate has control is this qb short short channel meaning in short region you can see the yellow region will be slightly different had it been the long channel compared to the long channel I'm coming to that in a second so that region I wanted to say that sort of gate doesn't have control the drain has control that little red triangle that I just saw showed you how much area does the gate has have control you can easily calculate let's calculate the area of the trapezoid do you agree z is the width width in the in the other direction l is the length right l prime l plus l prime divided by 2 do you agree when you have trapezoid don't you take the two sides and divide by the two and what is the height height is wt that's the height and then you take the average and q na and the z is the width in the other direction perpendicular to the flow of current so you do the do the math and then by the way c ox do you see from the c ox I will also pick up an l so correspondingly here you have the corresponding charge so I have l plus l prime divided by 2l if my l was huge and l prime almost comparable then it would be 2l divided by 2l it would be 1 and I get back my long channel limit do you see that so that is exactly what will happen because you see in this case the trapezoid and the rectangle is about the same and again you can see in here you can do the long channel 1 in the long channel 1 l is almost equal to n prime if you have 10 micron and then 10 micron minus 40 nanometer you know because of the depletion all right that's about 10 micron so therefore in this case you will get back the original one so how is the difference between the threshold between these two well there you can just subtract if you subtract then you will pick up a l prime plus l divided by 2l on the right hand side if l prime is smaller than l then you can see there will be delta vth and therefore there will be a roll-off so the physics of roll-off is you are losing control of the part of the channel and that's why the threshold voltage is no longer what you calculated for the bulk semiconductor now I'm not done yet l prime is something I have to calculate how do I do that well this is a very simple geometry if you just pay attention you will find out I want to calculate l prime do you realize that this the difference of the two trapezoid it will be l minus l prime and I have half of it so I'll divide by 2 that's the difference let me show you here let's look at the green triangle rj is a junction curvature right that depends on how you have deposited your substrate I'm sorry source so that's rj and we can calculate it just from the morphology ws can you calculate it of course you can calculate that's the amount as soon as you know the doping you know what the depletion is so the hypotenuse will be rj plus ws right so you can see on the left hand side of the equation I have it's a Pythagoras system so rj plus ws square what is the height wt right so that's the perpendicular wt square and do you see that the base is rj because that's the circle that goes through plus l minus l prime divided by 2 right because l minus l prime is the whole difference it is shared between source and drain so l minus l prime divided by 2 that's it and so you can calculate l prime from here do you see do you know everything of course rj you know yes from the curvature it will be given wt is a depletion you know how to calculate wt right the band bending at 2 phi of f so you know how to do that and you can calculate it insert it in this expression and you know how much threshold voltage you are going to have in order to for a given structure for a given rj and for a given channel this is how much you would have now what will happen that you want it small right because I don't want threshold voltage to roll off I want it to be flat and now if I want that then I want that let's say 10 percent I will allow only 10 percent roll off so I will put this delta vth let's say original delta vt was 60 millivolts 0.6 volts let's say and I want 10 percent of it so I will put 60 millivolts here and try to optimize everything on the other side so that my it never goes above 60 millivolts that's what I want so you can see immediately what will you have to do even before I go anywhere in order to uh in order to keep it small first thing is you'll have to keep rj very small if you could that means very shallow source and drain therefore on all modern transistors they will tell you that we have a modern process in which instead of using boron I am we are using a composite of boron which doesn't diffuse as fast or they are using laser annealing so that it doesn't diffuse much they are just trying to make rj as small as possible doesn't diffuse too much and so you'll have to and when we are going to talk about it but you get the idea that we are going to work on this formula so that I never exceed alpha naught so that uh my roll off doesn't happen I have good control over my transistor so there are many ways so first of all now the alpha naught I have brought it to the other side and the l I have taken to the uh taken to the left hand side this is the same expression as before now this one tells me that even in this case for a given structure if I know an alpha naught it tells me what is the minimum channel length I should be able to use for a given technology right given maximum no problem minimum is this much so this is what I was telling you that how would you make you want element to be as small as possible if you could you would make it zero so that is flat all the way through how would you make that you will make rj as small as possible that's the shallow junction technology the finfet and others that people will talk about and I'll discuss it in a little later all controlling r sub j is a very important thing you do that second is what are you going to do about wt if you want to make wt small what do you do you increase the doping therefore every technology generation that the doping of the substrate is gradually increasing right you want wt to be small but then there is of course a problem in terms of junction breakdown because if you put too much doping do you remember the electric field becomes very high and then the impact ionization may start so you don't want to go overboard and you want c ox to be large right large so then therefore epsilon at a over d d is a thickness you if you want large capacitance then you want smaller thickness and that's what the therefore you people are keep making the gate capacitance smaller smaller smaller started from maybe on the order of 2000 angstrom 3000 angstrom in 1960s today it is 10 angstrom size of your DNA one one strand of the DNA that is how thick it is these days and then of course higher k dielectric because epsilon if you increase that also increases capacitance that also makes l mean small and we want real small in me so i'll go through them one at a time so one thing people do in order to control threshold because in any way you want i just a few side thing is something they do that instead of having one gate many times people would put two gates you see on the top gate do you see blue is the body thickness of ta and then you have an oxide t ox and the vg that's the the gate is in black but instead of just having the gate on one side you have gate on both sides why do i have that because i can have better control over the channel this is called a soi silicon on insulator so the bottom left side could be an insulator could be an gate now in this case what happens that i just told you about this part and then there are regions outside the control of the gate that is causing us problem right we would discuss that and i will come back to that problem again but if you have a sort of a substrate first of all that region the insulator wherever that stops depletion cannot go any further right otherwise it would have kept going so instead of wt being the relevant depletion it cannot go beyond t silicon whatever is the t silicon it will stop there because there's no charge in the insulator so therefore it will stop there so you can immediately see why the wt will now become replaced by t silicon and that's good because that can be very small you can make it very small without the help of doping right and so that's a good thing and also if you have a bottom gate in addition then what will happen that the bottom gate will want to control from the other side it's like having two kings and the same land so now the remote lands are also being controlled by the other king sort of and as a result you have this entire region now under the control of the both gates an uncontrolled region is much smaller and as a result this would have a much better threshold characteristics do you see that now of course these type of things are often done i promised you in the beginning of the semester that by the time we are done remember that i had showed you all sorts of pictures and i said that by the time you are done you will understand them all so this is my last promise and last picture do you see that from here this is called a fin fed this has different names omega fed x fed and all those things physics is the same they know have different names because they everybody has their own every company has their own patent and if they use fin fed the names if in fact they will have to give each other royalty so therefore the names are different but the essential the details are exactly the same it's a lot of money by the way so i understand why the change names source on the left side drain on the right hand side and it is like as if a car is going under a bridge and so the electron goes from there to there source to drain and the gate is on the top like a like a bridge surrounding from all three sides so i just so showed you a two-sided gate but this is sort of a three-sided gate surrounding you from all sides and as a result what happens when the electron goes from the source to drain this is a perpendicular cut from source to drain the green is the fin you know it looks like a fin of a shark so it's a fin and the electrons are sort of controlled by all three sides so therefore it has a much better tracial voltage control and as a result you can scale this much smaller than compared to planar transistor so you see that right from three sides you are sort of surrounding it omega is almost like symbol omega so it's almost like from all four sides you are sort of trapping it so the source and drain doesn't have much control at all but there are problems if you make this very thin again this quantization and other things comes in because if you mix very thin then what will happen do you remember this formula long time ago that you thought you never need again those are there will be bound levels quantization will occur square well this is in silicon green regions are silicon and as a result what will happen that if you make things very small your band gap will now be from e1 this bottom of the ec is no longer available the first level available level is at e1 and your band gap will keep changing so if you make a little error in the processing let's say by one nanometer then your band gap is sort of varying all over and if your band gap is varying all over your tracial voltage is varying all over and therefore you will not have a good technology so therefore it's a challenging thing but of course it helps out okay so this is the new band gap by the way do you know what will be the intrinsic carrier concentration now in the presence of e1 and e1h so this will be ni square will be nc nv yes e to the power eg but that eg is not your old eg this is eg prime because things have quantized and electron cannot sit in the bottom of the conduction band and there is another problem these are all modern transistors problem that one has to solve the other problem is that these days transistors are very small let's say 20 micron by 20 micron by this now if you multiply if you have a doping of 10 to the power 17 and if you calculate how many how many atoms you have many times you have just on the order of hundreds or so physical atoms because 20 is a very small right and so 20 by 20 let's say this is the s rams are the minimum size transistor this is the type of memory and those are very small area so every transistor have a different number of dopant atoms if they have the now different number of dopant atoms that means they're doping density if you divide by the volume is different if your doping density is different your tracial voltage is different right every transistor therefore nowadays have different tracial voltage because it's so small these are all consequence of short channels right if your channel was 10 micron long 10 on the other side who cares you have millions of atoms it's only when you try to make things small short channel effect these are the consequences of it and there are lots of people make it make a lot of work on this this is an empirical formula from IBM and those red points that you see those are not electrons those are actually the dopant atoms the blue you see those are the dopant atoms and you can see in a small structure those are countable number and as it's getting smaller this fluctuation and this is the Gaussian curve associated with this fluctuation so the mean threshold voltage is 275 millivolts but you can see some transistors have 175 millivolts and some have more than 325 millivolts on the x-axis so if you try to make a transistor I'll circuit out of it then you see you'll be in trouble because everybody wants to turn on at a different point and that's the problem because every transistor is different and if every transistor is different everybody has a different drive current and short channel transistor it's a it's a very difficult problem to work on. Now very quickly I will tell you about how to control this how to control the short channel effects the first one is that in the standard one you cannot really do much except you know the first surface potential is 2 phi of f in this particular case if you don't have a drain bias right you remember that point how much you need to bend the band in order to get inversion one thing you could do that if your dopant number is sort of uncontrolled in that case you can get use the bag gate which is a vbs you can use the bag gate to control the threshold voltage because if you apply a so far we have not been talking about the bag gate it was just sitting at grounded at zero potential however if you apply a bias on the bag gate then fp fp will be moving right so this or here it's written as I'm sorry yeah the fp is it will be moving that so it has been sort of if you apply a negative bias less than zero then the fp the whole fp would be moving up and that's why what you see right fp do you see that has been moved up from this equilibrium position because it's a negative bias as a result you can control the voltage at which the transistor inverts by controlling the bag gate bias so by doing the bag gate bias you can counteract some of the effects of variation due to due to doping you can also do it by control the threshold voltage by choosing appropriate metal right this you know that this is the previous one probably I needed to explain a little bit more but this one you you know already if you choose a particular work function and that is how the intel most recent intel transistors are that you choose a metal you see it's no longer silicon silicon polysilicon but rather you choose a metal and by choosing a metal do you see that you essentially control the vbi because this metal work function is variable and therefore you can also control your threshold voltage that way because threshold voltage depends on this flat band voltage and therefore you know you can all almost bend it in the very beginning as it is here or you can move it to the other side so depending on the control you want instead of looking at the substrate which is getting difficult you can just look at the gate and that gate through the flat band voltage of the gate you can control the threshold voltage so that is also a modern way of handling short channel effect and the final point I wanted to make was this thinner gate dielectric and higher gate dielectric constants now if you make it you can of course what I what I want to show here you can of course do that I'm showing you the problem with it you can of course make it you say okay I'm going to make it five angstrom two angstrom whatever I want to make it that will make my l main smaller that's good but the problem is that if you try to do that you know you already know the problem that you'll have a huge amount of tunneling current yes you got the short channel it's like jumping from the frying pan into the fire the that's that's exactly what the situation is yes you got rid of the short channel effect but you made the oxide so thin in the process that then essentially you have a huge amount of tunneling current flowing through and then that therefore that's not acceptable that's too much too much power dissipation so the final thing is then the my only remaining option is to increase the dielectric constant because if I increase the dielectric constant I can still keep x not big right I can still keep x not big so that my tunnel current is not a problem but by increasing the dielectric constant and that's why we use high k dielectric higher than silicon dioxide silicon dioxide dielectric constant is four so hypoquinium oxide is about 20 right so you can if you put it in then you can also do smaller l main and these are all good things except that high k dielectric has a lot of defects it's again the other problem high k dielectric has a lot of defects and if you put high k dielectric exactly on the silicon surface then the electrons get scattered by the high k so that's the problem and I'll talk about that in the in the next class but for the time being what I wanted to emphasize stopping here is is the issue that short channel control is paramount you have to do it now you do it by various ways you know junction junction curvature oxide thickness and everything that you do then there is a price to pay you have to understand the tradeoff and once you understand the tradeoff then you understand the limits up to which you can push a particular strategy in the end you'll have to use a combination of all those and that's what keeps people employed in the industry so the next topic that we want to discuss is how people are working on mobility enhancement to see whether they can get a better drive current even at short channel transistors and this is when you have exhausted all possibility related to thinning of the oxide because if you thin it too much then there'll be too much gate leakage that's the problem high k dielectric yes you can put it increase dielectric constant but there is just too much charge strapping in in and voltage instability associated with it so the only remaining thing is then trying to go for the mobility and see whether you can do something increase the mobility somehow so that you can have an enhanced drive current remember drive enhanced drive current means that you can charge capacitors very easily and that means you can have higher speed higher speed of the transistors now the first thing i want to mention that mos mobility is a little strange this is not something you might have expected from a typical device mobility on the left hand side the blues are source and rain and the yellow is the depletion region i haven't drawn the inversion region but we'll assume that there's the inversion region right there underneath the gate and typically you might think that the electron will go from source to drain and therefore and as they go along they will be scattered by the ionized impurity you know this is depletion region lots of ionized impurity they'll be scattered by phonons and if you wanted to know what the mobility of this region is you could go back to the first part of the course where we actually discussed these particular curves so i hope that you remember that this is probably too small to see on the x and y axis this x axis is doping and the y axis is mobility and on the top one at the top graph we have the temperature as a function of mobility so let's say you are at 300 degree you could just stay with the bottom plot and if the doping is let's say for example 10 to the power 18 for electrons let's say this is an NMOS in that case you'd just come around 10 to the power 18 and correspondingly read off the mobility values and you might have expected that that's the value that you have to put in your calculations not really because this electrons actually cannot go straight i mean electrons never go straight they scatter back and forth anyway with the three-dimensional lattice but here we have this additional complication that the gate voltage is asking the electron to get out of the gate right which has a very high electric field it is asking the electrons to get out of the gate now the insulator is showing you saying that you cannot go and so its effort towards the gate will be repeatedly be frustrated by the gate insulator so it will bump along repeatedly along the gate oxide before it can get out and that's the rate curve i mean the black one is also bumping around but without an oxide then you don't have that additional problem of quote unquote surface scattering because now it's sort of being pulled very close to the surface and therefore there is an additional scattering now you realize that if your gate voltage is high then the electrons will be pulled closer and closer to the surface effectively and so the effect of this oxide scattering surface scattering that will be enhanced and that is what you see in terms of a mobility field characteristics like this let me just explain how these curves are done the top sets of curves are done at 77 degrees c and each symbol let's say open circle the field circle open square these are done at different densities and there are two sets of curve the top set is at lower temperature 77 degrees c at 77 k and the bottom one is room temperature 300 k so these are the two sets now the first thing you will notice in comparing between that mobility on the left hand side plot and these are by the way all for electrons and the mobility plot that we have seen before is the following let's say we want those two so the one is four times 10 to the 15 and the other one is seven times 10 to the 16 so these two doppings open square at 300 degree and open circle at 300 degrees so these are two different doppings you expect that at higher doping the mobility will be lower right because there are more charged ionized impurities so as the electron goes along they'll be scattered more of course you expect a little bit low and that is something you've seen on the right hand bottom side as well now if you compare however and look at the value these two red dots if you look at the value you'll see that approximately the trend is right in terms of that at higher doping the mobility is a little low but the factor is off by a factor of two or so because of that surface scattering right and so that is very important the other thing that is very important is that you can see at higher field essentially the car begins to deviate from the dotted line dash dashed line and begins to go down very fast mobility means to go down very fast and that happens because of very strong surface scattering at very high fields right so surface scattering in MOSFET is very important wasn't important bipolar why not because the current was flowing vertically there is no surface to think about so as a result we have this additional thing you also realize that at high fields the doping dependence disappears right you see that all the curves are merging together and this happens because at high field essentially surface scattering dominates as a result you no longer have the dominance of ionized impurity scattering why didn't i see it on the right hand plot because remember here the voltage applied on the drain side sort of was very small and there's no gate field here on the right hand side so the entire issue about the gate field actually are absent in the right two figures remember these two are two terminal register characteristics whereas that's the three terminal MOSFET characteristics so the mobility is very different bottom line you have about on the order of maybe several hundred or so mobility centimeter squared volt second that's the mobility you have in silicon 100 surface you remember right why it's 100 because the number of defects on that plane is relatively low so although you could have higher mobility in other planes that's what people have historically that's what they have preferred okay so the MOSFET mobility is different now what are you going to do how are you going to improve it improve this mobility so that you can have more current right how are you going to do it oh by the way so this formula essentially tells you that how the mu naught which is the standard mobility changes modifies with the gate voltage vg theta is some parameter some constant that people people determine now what people do in order to get better mobility is something called a strain straining the channel and this is how it is done on the upper side what you saw where you see the red squares this is supposed to be silicon with a certain lattice spacing you know this is an idealized version a square grid on the bottom you have silicon germanium let's say a material with slightly lower lattice constant not too much more not too small maybe 1 to 2 percent at most that's how it's how the blue squares are different by the red wires the red squares now what happens when you put silicon or the red on the blue and you essentially grow the red ones on top of the blue since the blue is quite few large number of blue there are and a little bit of little bit of red so the red will then try to conform to the lattice spacing of the underlying substrate right and what it will try to do then it will essentially be squished squished laterally so that it can fit into this and of course in order to when it gets squished on the laterally it will expand vertically that's called this Poisson ratio that if you squeeze something in one side it goes out on the other side so that's called the Poisson ratio the bottom line here is as a result of squeezing this is silicon yes but not really because this lattice spacing is a slightly smaller than silicon right as a result when the electron flows through the red its band structure will be little different the effective masses will be little different and therefore when you apply an electric field there is a possibility but not always it doesn't happen always for all materials but there is a possibility that the effective mass will be reduced and therefore q tau divided by m star that's the mobility and the mobility will be enhanced right so this is called a biaxial strain because you are sort of straining it from both sides on the x and y plane so that's why it's called biaxial strain and it gives you a huge amount of advantage by the way if you keep growing the red on top of the blue make it very very thick then what's going to happen red is going to say that I'm not going to go with the blue so it will create a defect in the interface and then it will relax back to its original lattice so therefore you cannot draw it very thick and there is something called this critical thickness beyond which you cannot grow so as a result you have to keep this material thin and then you can have a huge advantage now how big an advantage you know there are lots of work going on but I what I asked you to see on any of this this work is that look at the top layer which says strain silicon strain silicon see and the various types of things that people are trying this is strain silicon dual channel hetero structure in the bulk but the bottom line is for this class is that you see that there is a top layer with strain silicon and a layer underneath which is slightly yellowish orange that has a slightly smaller lattice constant so silicon is forced to conform and as a result its mobility goes up and this is the correspondingly strain silicon you see on the right hand side is schematic diagram and then you correspondingly look at the particular device configuration everything starts with silicon substrate though everything starts there then you grow silicon germanium which has a small lattice constant and then the strain silicon on the top higher mobility right how much higher oh by the way so this is an like a recent experimental device configuration which uses strain here for example the yellow part is a folly gate and this magenta region is a channel region and the greenish region is a silicon germanium underneath under layer which is trying to essentially providing the framework for the silicon to conform to now this has tremendous enhancement tremendous enhancement so let's first look at the extreme right hand plot and this is the mobility as a function of the electric field a plot that I just showed you know three slides before when I talk about silicon surface scattering and all that is exactly the same one but this is I'm just showing the lower effective field part of it the surface roughness part of it is sort of cut out of this out of this figure but you can see here that the strain silicon which is the field black symbols can be significantly higher so for example more than a factor of two here in terms of mobility so this is a this is a great this can be a great device especially for n MOS right n MOSFET but for p MOSFET unfortunately the improvement is not as significant you can see bulk silicon well with strain you didn't get get very much you got from 60 to or maybe 65 to 75 that's not a huge improvement you wouldn't go to a new material for that but for n MOSFET you can you see huge huge improvement right so therefore it makes sense for n MOS and then there are people who are also doing instead of tracing it laterally from the bottom and biaxially here you can see the silicon germanium is sort of on the source and drain side and the channel is in the middle now in this case the silicon germanium is sort of trying to maintain its space and the channel is sitting in the middle so on both sides it's being squished right and it's being squished on both sides so this is called an uniaxial strain because previously it was sort of on the bottom it provided the template the top one was sort of sitting on top and it was being squished on both sides in x and y but here you put this silicon germanium source and drain they have a certain lattice constant and the silicon is sort of sitting in the middle it's like you know too healthy person in the aircraft in the person sitting in the middle row that's the silicon strain silicon and the strain silicon therefore can have very high mobility in this case so your recent pentium will have something like this but so people are trying everything they possibly can in order to improve mobility that's the game now presently that's the game people are trying to play for example this must be a blast from the past remember this was zero one one zero zero planes and all those things so this is how they are playing out for example let's start with the bottom figure on the n mos so that i can walk you through how it works so in n mos the current flows along the green channel so that's the source and that's the drain on the other side the green channel and the yellow is the gate and so electrons this is done on one one zero plane one one zero plane and the electrons are flowing from source to drain along zero zero one plane so in this plane they have a good mobility so that's why you do it in this particular way but for p mos you see which is on the top you have the current flowing in the blue region from the source to drain but you have one one bar zero direction that's where people have experimentally seen that you have higher mobility these are all strain transistors by the way so you already have a biaxial strain and then on top of it you have a structure like this now you do not really have to understand the details of how it works the only thing i want to point out that these days orientation dependence and thereby enhancing the gain is enhancing the mobility is a big game in the town so that's something you should be just be aware of so it's no longer just one zero zero plane old silicon for last 50 years no longer people are trying all sorts of things putting this n mos and p mos in different direction there's a corresponding figure for finfax and again people are experimenting with all sorts of efforts like this so for example here you have the channel again p mos about the same going from the green in the green region and you see the gate but the n mos for a finfet remember this three sided uh device that we talked about and so in that one even a if you put it in a slight angle that gives you better mobility now again don't have to understand the details of it why it happens that we learn in advanced courses but in six or six you should at least know what people are doing and why they might be doing such a thing okay so the silicon is sort of is almost done silicon is sort of running out of steam in significant ways people are therefore trying to do strain trying to do various orientation they are going as far as they can go and steel is difficult for people to see how you can continue for next 15 years very difficult for people to see so therefore they are thinking about getting rid of silicon now if you have to get rid of silicon and invent the silicon industry quote unquote once more what did you have done you said said okay let me go back and see what material has good mobility because mobility is sort of important so you both go around the table you look at silicon well 300 these are of course inversion mobility and gallium arsenide pretty good 7000 but if you really have to make one you'd which material you'd make you'd make indium antimonide right 30,000 you cannot beat it so 30,000 we should make one and this is that's something exactly what people are doing so this would be a structure that was published i think last year's international electron device meeting lots of complicated structure but the bottom line is it should not you shouldn't be worried look at this this is essentially a MOSFET you can see the source and drain right titanium gold source and drain and you have a corresponding gate and the channel where the electron actually flows is that 5 nanometer the magenta region or the red region 5 nanometer indium antimonide quantum well that's where the electrons would flow 30,000 so people are hoping that on n MOS electron side yes this would be a good thing to have very good thing to have now what would you do in terms of holes you know holes you're not holes for you need for PMOS right for PMOS holes goes from source to drain what are you going to do about that well in that case you see none of them are actually giving you much right 450 400 200 well it's not not not great what are you going to do then well if you really had your option then you wouldn't make PMOS with any of this material at all you would make it oh by the way before before that on the n MOS side finishing in on that n MOS side this is something a paper I posted from nature 2009 on the website people are also thinking about graphene which has a mobility people say 100,000 not 30,000 100,000 and people are thinking about making transistors out of that also now for PMOS what you do have done well PMOS at least for the holes you would pick germanium the germanium it has about 1900 a factor of five more than silicon that would be great I put a star on the electron column is not because this value of the germanium n electron mobility is not known bulk mobility is known but I did not find a value for the inversion layer germanium mobility value so that I didn't put in there but the point is that is relatively low far lower compared to entium entium monide but for PMOS side germanium is great so if you had to design all over again you'd make a indium entium monide or maybe graphene n MOS and you'd make a germanium PMOS in fact lot of work are already underway in which they are trying to do exactly that now this is in some way comes back life comes back in full circle because germanium was the first pnp transistor that was made remember that point contact transistor it was made of germanium and it was that's why in all textbooks they always do pnp because what is the minority carrier here in the pnp transistor holes right and it's a minority carrier mobility which makes the difference and so therefore all the original germanium transistors were pnp because that had very high whole diffusion coefficient whole mobility higher diffusion coefficient higher by einstein relationship d over mu s k t over q now but the main problem why this disappeared was this band gap is small how much is germanium's band gap point six five to point seven five something like that and so it was very noisy germanium transistor even at room temperature was ready a lot of generation events recombination generation events so there was a lot of noise in germanium and more over you couldn't grow oxide on germanium germanium oxide doesn't exist so therefore silicon which had a higher band gap and on which you could grow beautiful oxide that essentially took over but now we are coming back in full circle and trying germanium one more time at the end you should realize that this is the oxide not the material itself that dictates the choice of what material you make atmospheres out of it doesn't really almost matter if you don't have a great oxide to prevent the current flow it doesn't matter what material you choose because at the end your interface will sort of eliminate all advantages that you had so to summarize then as i mentioned that short channel effect is a serious concern for MOSFET scaling and for essentially you try different things thin oxide high k dielectric shallow source and rain you do all sorts of things but the problem has been that for last three four technology generations the oxide has not scaled very much at all because too much gate leakage is already one nanometer if you go any thinner then essentially it will become a bipolar transistor lateral bipolar transistor so that's that's the big problem that's why people are trying all sorts of new things now this is people are trying many novel solutions for material for example in teamentium device structures so remember the fin fade planar structures right omega fade all sorts of things and on circuit level there are lots of tricks that people are people are working with now at the end because you have put everything together that is why these days it's possible to think about on commercial scale a 30 nanometer channel length transistor do you know how small 30 nanometer is one nanometer is five atoms right that's about it one nanometer two angstrom remember two angstrom is the spacing one nanometer five atoms 30 nanometer 150 atoms you have 150 atoms between source and rain that's all you have and to achieve that level of integration and at that lower price is fantastic and that only happens because all these innovations in material devices and circuits are going on simultaneously okay so i will end here and we'll pick up in the next class