 This course as I said is something to do with current technologies going on. So I just want to tell you that this course requires understanding or at least some flavor or some interest in physics, chemistry, maths, materials, science, mechanical, civil, electric, every branch of engineering. And actually that is how I like the area because I myself have fabricated the first India chip along with others in TIFR. So we are the first or may be called pioneers in making India's first chip in 1979, 80. So we have actually given us impetus to work in this area and you need to know many things. So actually technology should not be taught, technology should be actually done. Then why this course? The course probably is before you enter the lab these days the technology has become so complex that unless you have a priori values known what to do in the lab, which we say simulations, process simulations, unless you do all your circuit design for the process on a computer, you should not enter the lab because things can go awry if you have no idea where to work with. So this course essentially is trying to, we will see what is my objective and as I say 80 years down the, I mean many years ago I taught, my other students have joined as faculty, now they are professors and they have been actually wanting to teach this course long about of example. He joined in 96 or 98 and he says sir I like to teach this course. I said take it. Then I started a course, analog circuits, VLSI analog and some other person joined and he said I want to teach analog, I said take it. Then I started a course called RF design, RF VLSI. Then they said some other person, Shalab Gupta came last year, he said no, so I want to teach, I said take it. So now I have nothing left to, I have started five new courses in this group. But somehow I was not allowed to continue for long in many except this technology which I taught 16 years at a trot. So it is a pleasure for me to come back to area which is most, what I should say, the area which I like most. But in IIT last for some years now people think I am in circuits, yeah I am doing VLSI design for last maybe 20 years simply because in India the VLSI industry is mostly a design industry. There are no fabs and therefore the students find it very easy if they have a design project or design knowledge to get absorbed in those industries which relatively much higher pay scale than any other branch of engineering maybe except the computer science. So this is how we kept working on VLSI design activities. So even though courses are which I started have mother with N people have started taking that course. So I am left with teaching second years. So I teach analog digital devices to second years. So after many years, I know I did teach two years ago CMOS VLSI design for MTechs. That was the last MTech course I had. Hopefully this may be not the last, maybe they will give me another one. You know after you retire and your chair is not there, no one cares. So that is my problem. So this area as I say today and maybe on Friday, I will only introduce to you what is going on in the world, some kind of history of electronics or history of semiconductors. And please take it, this is a sentence I keep repeating every year with no good results many time, those who forget history, history forgets them faster, okay. So never leave your history because history is the only thing which is linked to you for future. Another way I have found that this year for last three years, one of my students who is now professor, Anil was taking this course and I casually asked him few days ago how many students he was having in last three years. He said 8590, I said fine that is great number, 8590 is a big number. Today morning unfortunately I just opened the web page for this course, I figured 150, I was dazed. Second day I understand 150, okay so as I say for last two years I am retired and maybe another three years I may stay hopefully. So this is a course which I like most and haven't taught many years. So let's see I do justice to what I understand and what you understand. Here is something details about course, my name of course is there, my email address you write down, I say six credit course and my objective for the course is to expose you all to a technology which has revolutionized and provides the world in every sphere of our life. The course emphasizes on silicon integrated circuits. So there are many other semiconductors which are becoming more popular now and hopefully they may also later join the race but today of course is silicon, silicon and silicon. So we will talk only about silicon IC fab and we will try to give you enough information on various processes which are used in fabricating these ICs. Major thrust will be modeling of the processes and also will give you what are techniques to make them. Techniques are not that very important because many of you may never actually work in the fab lab. Very few may be allowed or may be trained to do that but modeling of course CAD is something which everyone does so we stick to that. So modeling is major worry in this course. So silicon solar cells are very getting into very importance in world after almost lull of 20 years, 25 years no one thought of silicon solar cells but it has come back heavily now. Now money has been pumped in by all governments because suddenly they realize petrol has a problem okay. You can see how many wars are going on for that okay. So that is the issue, how much corruption on that everything is going with the oil okay. The text of this course is found in 3 books. I normally do not teach from any book in general but these the first book in particular plump deal and reference book on silicon VLSI technology, fundamentals, practice and modeling. I had only second edition please check it if it is higher editions are available they will be better. Okay so this book is available or not check it but hopefully available maybe costlier. So the other book which is not in the print now. So this book is one of the most or rather one of the best possible book in 80s or 70s rather. He is the pioneer of VLSI technology or semiconductor technology and his name is Saurabh Ke Gandhi. He is Indian but settled maybe born brought up in US I thought so. He is actually from Mumbai okay. There is another person's name I say Chang and Zee. They have a book on VLSI technology. There are many other books in the library check for them but as I say mostly I will teach from the first book mostly does not mean I will teach from that book okay. But possibly as most material is available in this first book. So we are talking about what kind of things I am going to talk about I may talk about crystal glues, diffusion, implants, oxidation, CVDs, PCVDs all kinds of deposition techniques and all models for them okay that is most important part. Ethography which is the major crux of all the process going on. We will talk about that. We will also look into some failure mechanisms and as I say some silicon solar sale also we will talk. At the end of the day or maybe the important output of this is to show you how silicon chip is particular IC is fabricated in the process. There may be 24 or 30 process steps or mask as they called. Total steps may be around 500 when the chip is made but the major step which I say 25 or so may go to 30, 35 these days and these are something again how an IC is made okay full IC. Processes in here all these processes are used to generate an IC. So how an IC is actually made typically we will talk about NMOS and CMOS okay. When we start with something today first lecture is micro to nano a journey into ICs, IC technology. Anything in VLSI or anything in electronics for that matter everything in the world these days is related to economics okay that is why all these people rule us you know banks and every one of you also want to join that because that is the bare money is. So actually if you see the this is slightly old slide how many things are actually electronics things are in this there is a equipment material semiconductor electronics products and customers demand are all kinds of equipments you can see from there mobile to whatever it is okay since there are around 300,000, 30,000 billion dollar businesses going on so very important for us to be part of it just to for your own sake and to say hopefully so if some of you take projects in technology hopefully many of you should then there is an I hope in near with this there is a place north of this Palanpur first India's fab will come already land has been acquired some work has started this is Hindustan semiconductor Michael and central HSMC like TSMC of Taiwan via India's now and it is starting with a group which whose head is professor Dr. Devain Burma who was my classmate in Masters. So I am also consulting them so I assure you that maybe 3 years down you will have first India's fab and will be working on 32 nanometer process. So history if you see from 1900 to 2000 I did not slide so I did not change but I can show you higher up later we started with a vacuum tubes diode then we went to 30s we talk of concept of transistors concept no transistors first transistors came in 47 this is essentially from Bellab effort first MOSFET came in 60 and then we started connecting those circuit components into one the most important part of integrated circuit is that all components are made out of on the silicon itself or on the wafer itself and that is why it was called word integration the metal connection everything there is no wire there is a connecting line which itself is a connector inside on the top on the bottom also there are many kinds now then if you look at the circuits what is our effort right now we are looking for first reliability then we are looking for low power high speed high integration all kinds of requirement scale and we started putting larger number of devices on a chip typically now we can put 1 billion devices on a single chip okay that is the kind of technology we have and sooner we may have 3 to 4 billion devices on chip but I will give you some sheet like this slide later which getting too many is also not profitable first vacuum tube leading for rest can see you I do not know any one of your your maybe your parents also may not have seen but hopefully yes there is to be glass tube roughly 2.5 centimeters to 5 centimeters and used to be a rectification as well as transistor equivalence amplification there are 3 terminals there one here one here and one here so one is called cathode one is called anode and in between is called grid grid is biased so that the electron motion from cathode to anode can be modulated and that sort of amplification can be obtained so this is the first transistor version which appeared in 1906 and leading for rest and Wilson got the Nobel Prize so it is a very bulky in 28 some army officer air force officer in England actually suggested that you can have a solid state equivalent device which can control the current from source to drain he gave you these names were not given by him this was little added later by himself but initially his first paper did not have these names so what he has he has a some aluminum over which he deposited aluminum oxide by oxidation and put 2 contacts which he calls source and drain and he figured out that this device can actually connect between source and drain and gate can control the current this is his first but he has not made them he has not made any mass the first transistor which was made was in 47 the 3 Nobel Prize winners are shown on the right Barden, Bretton and Shockley and the first transistor looks it is a point contact transistor may be a better figure is here you can see how bad it was in those days just to connect things but today the IC looks to be fantastic the first integrated circuit was due to the effort of Jack Kilby at Texas Instruments and what he thought that why actually get different components from separately so why not use silicon areas itself to create resistor capacitors and those who have never thought of inductors but and he could put some device bipolar transistors created there itself and connected on silicon chip itself okay and that was the first IC just to give an idea in 58 Jack Kilby at Texas made simple oscillator which has 5 IC component resistor capacitor distributed capacitor and transistor and in 2000 the importance of IC was recognized please look at it it took 50 years for the Nobel community to think that this was a great thing so they awarded him in 2000 and which was also shared by him by others so in 2000 Kilby got his Nobel prize the other person who would have got is the next figure and show was nice but somehow he expired and therefore he was not the participant on that so this is the first IC and just to give an interesting anecdote Kilby was hired after his graduation by Texas Instrument and he had no work to do so he was just sitting reading so in those days it was not high-pressured industry so he was allowed to sit and do something watch what's going training as they called in those days so he has to get bored so in his off time what to do so he did this okay so think of it even if you are doing idling do something constructive this is rather nice he made the you know earlier on the stage we figured out for example there were the contacts on the bottom side on the top and it is difficult to put a wire from top to bottom okay so only one discrete device could be made so make an IC how to connect different devices you know some top to bottom bottom to top connections that is wires will come and that whole integration will lose will be lost so what he did that he could bring all contacts on one surface it is called planar so that was his invention he said say planar technology was his choice Robert Noyes is also very famous for he was one of the founder members among the three of the most famous company of worldwide now in electronics Intel so he is the founder of Intel okay in 60 at Bell Labs the first most transistor was made 26 people suggested it 28 it took almost 32 years before it was first device was made it is a mass capacitor shown here which was created by Kang and Attala at Bell Labs and one can say semiconductor history is Bell Lab in history you are the Korean okay but unfortunately if you see there is no Indian anywhere okay we always follow we do think ahead okay so let us start changing this concept sooner the first vacuum to base computer was made in 46 is called Eniac it is electronic calculator okay and it has a big size huge power short life because the filament which actually generates electron by heating the ingest used to spoil you know an equivalent vacuum to probably know of course these days even that has gone earlier TV tubes used to be and now of course there are solid states but earlier there were TV tube at least you could say vacuum tube being used now most TV's do not have that the LCD or LED displays so they don't even have tubes now so the filament was to heat and it will generate electrons in vacuum and they will travel by the voltage of the field and they will be connected and if you moderate them in between so I can amplify something from the modulating angle so this was the first such computer this is your pocket PC and that is your computer this had it of course a smaller part has been shown it was having around one acre of land internally used to create a small calculator which does addition subtraction multiplication not even division 70 71 first large scale integration devices came DRAM 4K 4000 K or 4000 bits or 4096 bits this was 1103 from Intel and also there were first microprocessor which was a P-channel device is 4004 4-bit device 4-bit microprocessor this is the first microprocessor made of course there is a fight between many other company than Intel which is still not solved just to give an idea the recent SD that is your card which you see on your memories which is typically 1 in 128 GB okay this SD card is available you put everywhere in every system if it is a bite then 128 G into 8 bits which is terabits 10 to 412 is Tera compared world population 7 billion brain cells 10 to 100 billions this number is varying different people are different brain cells okay some may are 100 I may not not even 10 now many of them might have burnt stars in galaxy is 100 billion so we are talking of numbers which is even beyond the what one can really perceive and therefore this size is hardly one inch by one to one and a half inch thickness may be half half a centimeter even lower maybe two millimeters and you just plug in and you have 128 GB memory availability so the kind of thing technology could do is enormous just take it if I want to make the same whether you must have seen many of the slide have some borrowed from Iwai Hiroshi Iwai Hiroshi is a very famous PLSI technology person in Tokyo Institute of Technology and for good sake for me I am his friend last 12 years 15 years or maybe 98 so maybe 16 years I visited the lab 8 10 times so many of his PPD that could just marode to show you he visits every December those who wish to wish we have been here as permanent faculty for one month I had a Bombay Department so let us say you have on this each this is the 2.4 centimeter by 3.2 0.21 centimeters at the volume is around 1.6 centimeter cube CC and 2 gram weight and typically apply 2.5 to 3.5 volts now if you look at the vacuum tube which is a 5 centimeter by 5 centimeter let's say equivalent square if I create and 10 centimeter high 100 gram weight and it consumes 550 watts of power typical voltage you have to apply 300 volts so now equivalently if I create one terabit something in the vacuum tubes what will happen here is the number one terabit is 10 to power 12 so I made 444 bits for each of them if I put into a vacuum tube given then you can see it has a 500 meters this 500 meters this and 1 kilometer down that kind of system I will have to create to must make as a ticket the present day other towers of course is the old one which is Shanghai is which is possibly will come in 2016 is 700 meters India is going to build in Mumbai I don't know when now things are changing it may be seven and seven thousand meter also I don't know but the present day which has already been is the highest tower in the world is right now 828 meters in Dubai so look at it even the tower we they made is 800 meters and we are asking 1 kilometer thousand meters so if I had to make one small memory bit they scarred that this many tower this much is also not possible okay so the progress what we are done over the years these are something interesting data which you all should understand how much we are progressed if you look at the power 1 terabyte and about 12 bits as I said 50 watts of power it consumes and tubes so it is 50 terawatts we need 50,000 nuclear power plants for just 128 GB chip okay an example he gave me is in Japan they are only 54 nuclear power plants one of course is down now last summer Tokyo electric power company can supply only 550 55 billion watts okay now you are terawatts or one is memories one memory chip so you need thousand such companies to make 18 GB memory and imagine how many memories are being used in camera than everywhere in electronics so how much power you would have required just to create that small memory okay so think of it vacuum tubes great in those days 1900 it was real great but things where are we now okay so the progress of integrate circuit is extremely important for power saying if nothing else okay and otherwise if so much nuclear plants would have to put more just to create one memory if you look at an equivalent in your human this you can see your brain is like ICs there are many ICs there which are mostly processors neurons are essentially processors okay so there is an information which you receive through sensors your sensors are ear and eye your mouth which is like an RF up to device it actually takes it monitors it vibrates it and at very high frequency and that's how digestion takes place that's how the vocal cord works so yes let's say these are the devices which we make and this is what humans are your stomach is something like a photovoltaic device huge power is generated out of it okay and your hands and legs are like power devices they can move okay so he already humans are doing whatever is what we are now trying to do and therefore most of the biomedical people keep trying to put equivalence of that you know oh neuron this processor this equivalent okay but humans are right now the only thing which so far we cannot imagine emulate all robots are maybe at least 0.1% of the human capacities best of Robo since you are all becoming too smart not just smart everyone wants to do this okay and you also demand all kinds of applications so you need a very high performance extremely low power you don't want to plug in every now and then okay so the CMOS probably the only answer which is relatively low power earlier we used to say very zero power it's not true it also consumes a cell of a power comparatively please take from me that one of the thing which will maybe in this course I may not discuss is this current technologies of 30 nanometer down maybe even 45 nanometer down actually had devices in your cell of cells or mobiles which are all mobiles are stand by because they are not always on but had to be on because otherwise when the message comes or this it has to be turned on so power mode is also there power on mode is also there and when you can actually shut or shut it off so the problem is in this stand by mode okay because stand by mode power is consumed even if it's not really working the problem started in 30 nanometer down technologies or devices made that the off power is higher than the on power 66% off power 33% on power which means if you just keep your mobile it will leak without you doing actually that's why most of you keep talking 24 hours okay okay so that's some fun part in that semiconductor device market grow five times in last years it's though it is very matured market 2011 to 2025 300 billion USD was a 2011 and we expect in next 14 years I mean from 11 1500 billion US dollar business okay so if that much is the business then you can think how many people will be involved just to do this kind of business and therefore the research therefore the effort how do you want calculates cost of the chip let's say we make 1000 wafers per month that are parts in fact 600 dollar is a 6 inch wafer of course we will show you now it is not 6 inch 12 inch 16 inch quarter micron thickness quarter micron so it costs you roughly 3.5 dollar per centimeter square at packaging cost of 0.25 cents or pin or card or any kinds of package you have this is really costly part package is the costliest part in the chip okay chip is very cheap okay so typically from the numbers you get you find that the cost is calculated how many chips per wafer you'll get how much well this is the how much package cost and then there is a profit at the end of the day there is a profit otherwise who will sell okay so this profit makes industry go and whenever there is no profit which you see many of you are now good economists than me I never gone to so called the famous insects market this here in Mumbai except I used to see when I was in TI for the tower I have no money to really invest so I couldn't go there okay so I figured out that those people are deciding what we should do is very funny but that's how the money matters okay today silicon device is indispensable most important devices for a one human society everything has to be controlled by silicon silicon relies extremely high frequency high speed operation with extremely low cost low power small size reliable today's IT industry or today's IT products if you see such as internet i mode cellular GPS game machine entertainment robos they could not have been realized if there would not have been silicon ICs coming to entertainment it's very interesting that the major market for an integrated circuit is in this game machines if you go to luckily in India so far it has not been profitable there are games this playstations across most of the cities in the world in Tokyo at least there are thousand I know where are streets I have been there are four to five on every street and I don't know what whole night hack-hack-hack they keep doing and enjoy fine they are money so they enjoy so what you are doing these days it seems so what is that is driving us last hundred years if you see at we started a vacuum to be 900 let's say up to 2000 we went from 10 centimeter kind of size to 100 nanometers below now so typically in hundred years we have one million times reduction in sizes which was remarkable late suddenly in 2000 people started forgetting the word micro and they invented word nano now this word micro to nano is not very fair and this is the graph which I always show for whom I find the left side this side is a microns and this side is in nanos so if you see up to point one micron which is 100 nanometers we then started from few microns 10 micron structures down to 90 nanometers by 2000 and below 90 nanometer we say it is nano only one of the dimension was less than 100 nanometers and we suddenly jumped into because it looked the all politicians bureaucrats they have fanciful new words you know so if they only pay you project wise if I say I want a micro I will make a micro chip is a no money I want nano chip how much so that if I say I am going 350 student that salary is enough what else you want if I say I am going to teach 10,000 colleges 5000 percent from here oh how much money no problem there is huge money with us so there is some number game going on okay so in nano CMOS of course 90 is not bad 90 and we are now reached a state we will show you later is around 16 nanometer we are already working at but the game is nano were started when we crossed 100 nanometers and only channel length was less than a nanometer so we suddenly became new name new money okay that's how they see and appeared now if I was a center for micro electronics no one would have paid nano 200 colors we got we fooled them well okay not that we are not doing anything I mean that this money they would not have given if you would have now put the word nano okay so luckily for us this nano itself the this is some graph which I initially showed you this direction below nano it's somehow not going in that slope but to do the gate length the gate length is not scaling down okay in the same proportion and therefore some respite is there for a technology man you scale exactly same way then technology has a tough time to match okay however things are for example a nano 90 nanometer node is actually gate length is 50 nanometers and if you have a polygate device its thickness is 1.2 nanometers 12 amps nonce okay now this scaling down is very important why are we scaling we went from large microns point 10 micron 5 my first of India along with TIFR other colleagues few of them are here professor Dinesh Sharma was my colleague after I just resigned and left he was my colleague and professor Pinto who actually manages our lab is also from TIFR and few more and many TIFR faculty after retirement joined here let's see me okay so because of us we were first to make first IC in 79 80 time though I must honestly say did not work fully but it was made partly it was work it was shown to Indira Gandhi so all of us where it kept in behind only my professor was ahead okay probably didn't know how did we make okay that's how professors are okay why professors actually do this because if you know it you will do it if you don't know it you will teach it that's the way however I must tell you that first 25 years I actually worked in the lab so I am not one of those faculty who have never been to have I worked many devices including thyristors and all photo I have been many many devices in my career micro devices so I am not saying that but generally this is what most people believe teacher okay I don't know what I am I don't know so in 19 electronic started applications were amplified radio TV wireless 70 micro electronic started silicon IC scheme device features were 10 microns or lower slightly applications were still digital computer PCs they were the technology revolution then and in 2000 electronic started devices still silicon CMOS ICs features are less than 100 nanometers major applications are still micro processors cell phones and there is a technology ration maybe just evolution and innovation but great evolution and innovations and so many innovations if you think keep going this you can understand where we probably may reach 2014 nano-electron is continued still silicon CMOS around 10 nanometers still digital okay still evolution innovation is continued so think of it since I may not survive 2005 I hope so I mean I wish I don't want to survive beyond that itself I mean even now I am enough old but I can at least say 2025 silicon CMOS will not go beyond that I may not so I may say may leave up to 2050 also it's okay but I may not be answerable but 2025 I can assure you silicon technology cannot be defeated by any other material technologies whatever they may say whatever they may claim for taking money otherwise silicon ICs cannot be replaced okay take from me this was in 2002 a 7 nanometer gate first transistor are made 7 nanometers they of course it was not an IC it was an isolated device so even as early as 2012 years ago the first device with 7 nanometer was made okay so think of it technology going from lab to a actual five-house it takes 10 years now even now 7 we are not reached so research is a part of the game keep doing keep doing someday someone will pick it last few slides all of us are looking for tera instructions per second mega flops and this all a beta we are looking for terrors more than tera in fact now so we are looking for processors which can give one tips instructions by 2012 actually we are not reached so far we may reach soon we start with 8084 to 4004 I did not write but 8085 86 8286 386 Pentium Pro ethylon many years names appeared but still we have to reach normal case paralleling is paralleling is okay quad may do but quad has another problem for at a time working synchronous is very difficult sharing memories is also difficult it is the major way people are increasing the speed so a number of instructions per second but that is single microprocessor with capability of one tera instruction per second is still to be achieved though we start that 2012 it will but 2014 it has not reached so what is that game in this technology is and why we are learning over the years new technologies because even if this is a mass device I trust many of you are aware now in 1980s people did not know a mass transistor the mass transistor theory is to be two three pages in milkman halkia's book in those days that was micro electronics God knows why they call micro electronics and whenever we will interview a student for his empty entrance he will exactly draw the figure which is vertical mass device which never exists actually our silicon planar isn't it will go down but he will show you source drain gate because that's the figure given in the millman halkia then I will say carriers are generated because the minority carriers I keep telling minority carriers so small how much current no no that's what millman gives because millman was a circuit man he didn't know anything of physics so he keep telling nonsense and everyone come here and told me the same so I have to keep telling them sir this is wrong sir please listen from me so the mass device was so much odd for most of the engineering institutions that they came only at bipolar's so the question was asked that why bipolar was left and why suddenly moss came in two reasons one can see one is bipolar require larger power supply voltages compared to moss then they were not scalable like moss as I say we keep scaling down by some ratio performance still can be achieved a better performance is achieved but in bipolar scaling was just not possible base with could not be made zero so you can't scale base with too much so bipolar technology being costlier more power consumer suddenly could not match the number of devices required on a chip would be smaller in bipolar compared to moss so in every sphere it was not able to and costly so then they say why continue but 2000 again we are still continuing with some bipolar processes and simply because they are very high speed but they are not in silicon okay we are working on three five materials where the mobility is the important part and their bipolar circuits are coming back at least part of the bipolar and part of the moss is can be together called by CMOS but otherwise bipolar is out for all practical purpose even the opiants 741 initially we started with bipolar open 741 but nowadays you can't get one all moss and moss or CMOS by opiants are available so whole and everyone looking for money everyone looking for better performance has shifted to moss so here is a moss device I hope many of you know this is the silicon this is source this is one diffused area this is drain this is the gate and in between gate and substrate there are thin insulator layer which is earlier used to be silicon dioxide and even now many devices are silicon dioxide so if I this L effective is the link between source and then double is the width of the gate or transistor so if you just use the dimension by 30% transistor density doubles if our thickness of oxide is scaled down we can have faster because electrons taking time to go from source to drain will be reduced speed will be higher better performance and if I scale bars apply and threshold I may get low power so the whole game now is to reduce everything so that scale down everything so that you have better performance and much more density of devices on a chip what's the time like 12 good so what is in future this is the same the clock last only you can see the figure of graph figure this frequency versus year we are expecting around 2010 a CPU should reach 10 gigahertz or even earlier but we are not okay the problem in that we cannot do probably maybe we can hit it some way it cost would be so high that's not worth okay firstly home how many of you really you want to use a processor which is more than say 6 gigahertz mostly you use 3.4 to 4.2 gigahertz so really processors are a fast process are required only for those gaming interested persons who wants to kill the person on the screen before he actually starts okay so only for video gains probably you need a much faster processors okay or if you are doing a large amount of process arithmetic or large amount of numerical crunching maybe you need a supercomputer which should be very high speed okay one looks for tera if possible there also what we are doing is 40 of them we are paralleling merging then one of them fails all of them fail so all that so-called supercomputers in India which you listen good must congratulate them for putting so many parallel series combination and every now and then filling that they don't tell because anything is in series if one fails the series chain goes okay in parallel if one of them draws more current the other fails hot okay so paralleling is not the best solution okay that's why Intel is still not going beyond four processor quad that's it because paralleling has its own problems okay there it is one of the better problem a better solution to increase speed so I don't think even in 2025 we may cross 10 gigahertz though we believe we should be able to at least in silicon it is not possible other materials maybe we don't know something which is not known I should not say but I think there will not be enough use to go for that okay all industry works on how many people buy how much okay so this is the journey first IC has four transistors few resistors and in early 90s this was the first Pentium first processor chip which was Pentium one okay so what is the law which is actually allowing scaling and I'll show you the graphs a little later what he said a famous person who was the founder member of Intel by the noise phase and Gordon Moore and others they were all members of Shockley lab earlier and Shockley started his own company leaving Bella all of them came and started in California what happened there that Shockley's nature was very bad and at least that's what reported I have not met him so I don't know no Gordon Moore I have met so I can tell you something okay but others I don't know so because of them these people were so annoyed so they joined a company called Fairchild Fairchild was a camera company so camera company has some chips requirement so these people actually joined Fairchild and in clandestine way or otherwise they actually started working for logic chips designing how to have and all and when they are really with some blueprint they separated and started Intel some people from Intel also did same thing and they started AMD unfortunately AMD couldn't stand great competition with there was also another company which started equally well was Motorola which I don't know for God God's sake why it has happened it lost market everywhere wherever it entered the last one is mobile Google purchased the moto market and I sold it out to Lenovo okay so I don't know there is some problem with the word moto I have a moto right now my son gave me this so I am I am still struggling with this smart bird there so this problem which we started is that number of devices to be put in larger number created lot many other problems okay how to connect them so what Moore said every year if you scale down earlier of course he said every year but now every three years he changed recently this called Moore third law said 0.7 times you scale dimensions so 0.7 into 0.7 is 49 0.49 which is half so obviously if your area goes half so the component density will double okay half half double that was the law he gave actually it was an exponential law he took it to binary and declared that as his law Moore's law it was done in way back in sixties and it's surprisingly 2014 we are still talking about Moore's law and all the technology people designers keep watching that Moore's law oh I must reach there okay as if that is Sancro signed or that is gospel but that's what the movies great visionary that time in 64 65 the technology was not even mass it was a bipolar process mass was just PMOS devices just started coming Intel started making first logic on PMOS and this suddenly this lock came so great man I must say accordingly so what's the advantage of scaling you reduce the capacitance so you reduce power you increase speed why speed because if the capacitance goes down charging a capacitor if it is smaller the current require a CDV by dt so see smaller obviously charging current is smaller so time taken will be for a larger current higher speeds so that is how first advantage we see can speed up just by reducing size if you integrate then many functions can be integrated so many not just processor you can put everything on chip okay so there is number of functions where you have micro you have arithmetic we have a shift registers we are all kinds of circuits blocks which you can put on a single chip so lot of functions could be created and of course as I said parallel pressing is possible two at a time or three at a time therefore improve the speed so this is all possible simply because by reducing the dimension of the device which is called scaling law which is Moore's law you scale however that scale as I showed you is not going straight now it is slightly bent down so now what is it every three years new law maybe someday say five years see what is the demand from future VLSI much higher performance much low power consumption and therefore downsizing or reducing silicon devices is the most important and very critical issue for all technology and designers that's what we are that's why every year we have to change technology because the demand will come from video game side for example I want this then everyone will work for two years somehow to give them when you reach there there is a no no no this is not enough we are going to work for it okay that's how we are saving money to some extent and improving but this will continue 2025 beyond I don't know there are limits coming there are effects called short channel effect there are so in resistance effect direct annealing if these are failure mechanisms okay on the right cause which we said everyone is saying in 1987 or 1978 book by Mead and Conway the first VLSI design book in the world Mead is a very famous he's still surviving so Calor Mead said that you don't need to know technology to design a chip which was his he himself was a technology man earlier but he said so these are called design rules technology will give constraints so this constraints actually are essentially where up to you can go so he said by 1987 no more progress will be done 2014 still progress is being done so sometimes no one is like me always said it will Calor by his physics he said no it cannot but it did so what has happened why Carver thought that fundamental limits will be crossed and why because you know the we all understand physics but silicon doesn't so it behaves the way it wants that's why so it will be a silicon okay so now today we are working on 10 nanometers 7 nanometers maybe some day so there is something is not possible there is a association which I'll show you okay last few parts before I come back it's called ITRS that is the industrial technological roadmap company these people who meet some hundred 500 people in different areas join and predict what is next every next year so the ITR has also said that where there is something you cannot do so they put what we call red brick in that map you can't cross this but Indians are smart so so are worldwide so they made a small hole and got in that's what that is no one jumped okay red brick but I know under so you can't cross thermally tunnel it okay so no tunnel devices appeared okay so as I said by we are expecting 1 billion transistors already on the way this is the most law okay I did not put his face but it's okay this is also all law but still so one can see if you see here and if you see the number of died transistors per die the top most is essentially is memories so yeah we have already reached SRAM memories of 256 megabits of 5 from this then of course we are gone to 64 to 128 to 256 GBG bits processes we are it in your material and this yeah they are billion transistors on chip now so we are still following Moore's law okay to great extent only thing if slope is changing means number of years required to reach the higher value is increasing but that's so it's not saturating the whole game is Moore's law is still valid it's not saturating and that's something this person has to be on this so that that time he said from one year he changed to 18 months in 2013 3 and 2012 he said every 3 year it will be double all that he modified it go Moore's third law as they say so can see typically this is exponential law which was being followed and so he said double 2.7 is a bad number so you made it double so let's say we were at 1 nanometer and we wanted to reach 30 nanometers size hard every 18 months let's say in those days so we Moore has given a formula which is interesting whichever year you are in whatever technology okay so it is 2001 1.5 is 12 by 18 18 by 12 so it is one so you adjust the actual numbers there multiplied by the technology in that year divided by the technology you want to reach and you will get some number of years that's the Moore's law which no one no book will give this is I got from Moore so I can tell you it's the year whichever year your technology is today let's say 2013 or 14 and let's say 3 as he says so 36 divided by 12 so it will be 3 log 2 we are working on let's say 16 nanometer I want to reach 6 or 5 put it there find which year roughly this technology will be reached this is Moore's law okay of course question is arises what then what knows what power that means power and shift essentially what means we will have to change our thinking right now what we are thinking is what Moore says we think okay move the wheel is something you have to reach here running running running running that's what we are trying but let's not think more think something different none of us none of the world people are thought so far yeah there are quantum and everything I'll show you tomorrow day after tomorrow many devices which are come which are good and good on cat tools but no one is making any chip on that because cost and reliability is very poor so unless you really go to silicon equivalency mass no one is going to put money a typical fab lab for a one technology node let's say Intel has just put 11 nanometer technology node and invested 8 billion dollars okay the India Stan semiconductor our company we are invested 6 billion dollars beam is I have not paid any money I don't have that money my classmate has he is a venture capital himself we have a lot of money and he has borrowed a lot of money including 2 billion dollars from government of India so that's the thing which you should limit so more is not the end of it but as I say silicon doesn't understand anything it may still do something wonders now question arises many people ask us over the years initially our wafers are one inch wafers silicon wafers then we went for 3 inch and we went for 6 inch and we went for 8 inch and 12 inch now we are going for 16 inch so why not we start at 16 inch itself okay firstly the growth condition that is why the first part of crystal growth what is the problem to create larger size wafers and what is the problem there so now of course we are looking for 16 inch wafers the latest technology this catch phrases talk of nano technology but if you ask me nano technology existed way back thousands years molecules atoms are of nano sizes chemist people are always nano technologies so why are we now calling ourselves nano they existed alchemy was known some thousand two thousand five thousand years so everyone who works on molecules or atoms probably is a nano technology we only converted to silicon what could be done and therefore we became greater we thought no one thinks otherwise we only okay so now everyone wants to work on nano technology and then suddenly found silicon may not be the better material so look for three five compound materials look for quantum mechanical phenomenons cool-downs blockade some charge phenomenons cellular automata many other interesting physics-based devices come spin transistors all these are good fantastic okay but FinFET will work for next many years okay not at least 25 okay I don't know ahead there are a feature size of the order of 10 to power minus 9 meters nano so all of them we are working below this are nano technologies just to give a cost you know how how one cost how if you increase the die size die is the smaller chip area in a wafer 6 inch 12 inch water size you have some rectangles on some squares which forms one die which is the circuit okay so for example to shown here on this there is a larger chip size die size the other one is a smaller die size because questions were asked that why not we started larger die size day one okay so answer for this so let's say if you have this is larger and this is smaller this so there are different cost involved and as I said industry only works on cost nothing else what is good is money so if I can fetch money more then all technologies are good okay so don't say that 19 nanometer is worse yeah for many applications 19 nanometer technology is fantastic is doing all that job what people are asking so why invest money just because you want no the industry you only will do as much as what is demanded okay so the cost of it's called variable cost depend on the die side die test packaging and the yield means how many good chips out of one wafer okay it's called die yield total yield and then each is a wafer has a die yield die itself may not work so it's called die some part may work actually so the cost of die can be given by cost of wafer divided by dies per wafer into die yield now what are the definition die per wafer can be defined from its area and perimeters okay and this is a formula which I got from somewhere pi wafer diameter by 2 square upon die area minus pi wafer diameter into root 2 die area this is perimeter based this is area based now I can calculate die per wafer I know how many good dies are available so I know dies per wafer into die yield and I know how much is the cost of a wafer okay so I know what is the cost of each die and if I know my cost of each die and if I know die if I want to calculate die yield I we have created another formula which says it is 1 plus defects per unit area into die now these defects are the one which is material defects okay which may be created during fab unknowingly by knowingly also they are there even if you know I can't remove them or during actual chip operation also effects can start okay so based on this formula 1 plus defects per unit area into die area so alpha is technology parameter I will give some interesting numbers which I calculated yesterday the wafer size let's say of 12 inch let's say each die is 2.5 centimeter square that is 1.2 by 1.2 centimeters let's say each centimeter square has one defect which is many times larger but as these are technologies improving 252 dies per wafer we can get from the please remember wafer is circular die square so obviously age chips are not possible so typically 252 dies per wafer one can get for 12 inch wafer if your die size is 2.5 centimeter square let's say you have die yield of 16% due to defects which is what the worry is okay so only 40 dies per wafer out of 250 are available so you can think of it if any industry has to work if it says that only one sixth of the chips are I mean devices are dies are working then it's too costly for them okay so the cost is something now related to defects how many defects in process you get that's why we need better processing so one can see from here if I have larger yield then your cost goes down larger yield okay if your die yield is higher obviously you can see your variable cost will be smaller so the gain in the actual processing is to reduce defects okay okay here is the figure let's say there are seven defects localized okay in process if I have a larger die size you can see not a single chip is available so it's randomized but the intentionally shown this way so that you you don't actually get a maybe one chip out of four you got okay maybe I should have put it here but it's okay. However if I reduce the die size and kept same defects you can see that at least there will be 14-15 chips out of 20 odd numbers I am saving actually so if you increase the wafer size if you increase the die size it doesn't really help actually is that clear unless your processes are defect free and these are something which is not so easy to control all people style it their best so there is always a game how big the chip size should be for example Pentium has around 1.75 into 1.4 centimeter first Pentium chip but they are not going beyond 2 centimeter to center they are not looking for 4 centimeter square chip because that is the problem if I put larger chip size here no no chip is possible 8 defect everything gone okay. Other way I will have full wafer as one chip system okay but a one defect whole wafer gone 12 inch wafer cost around 600 dollars so both Pura Hama processing cost you have done so much that also billion dollars 100,000 dollars cost. So the problem starts how good is the defect free processing so one of the area which is called manufacturability which this course may not teach you or may not discuss I just wanted to tell you why processes you must learn because at the end of the day our value is cost okay so as good a process you create at that much cost will reduce and that is where the market is okay so unless so why people are going to go for off-shell products deep memories DRAMS SRAM because they are sold in millions or trillions okay so even if it is little less yield the sale is so high you don't mind actually wasting some money but if you are making a particular circuit for an application how many chips will be bought maybe 10,000 20,000 a lakh then the cost is so high and people may not buy say $50 then who will buy so the question and I must tell you due regards to all of you and myself I should keep but since this whole silicon technology is from us that money also goes in dollars we know it is changing every day with our rupee so I don't know what multiplier should you 60 or 58.5 or 64 so I don't want to keep USD which is standard but that's not fair why should we look for US but that's our life can't much do can't do much so proportion to also the die cost is strong function of die area it's proportion to third or fourth power of die rating actually is very high if larger dies are used before these are the three possible technologies which VLSR people are looking for in the market first is transistor typical MOSFET transistor is shown here this is of course interesting this is the one single fin FET a single fin FET this is inter process this is silicon germanium this is silicon noitride okay there are three different types of insulators used and this is a typical transistor can be used in logic the second technology is called logic technology the second is DRAM which is the highest sellable memories where do you think DRAM is put on a desktop or any system where is it you have a main processor which has how many what is the what memory it has cash so at best L1 L2 L3 L4 maybe few kilobytes whereas we need giga so the first memory just outside the processor which is connected is the first DRAM maybe data RAM but it's DRAM then there is a next DRAM and then hard disk so DRAM is the highest sellable component as of now okay camera every such come will require some temporary memories of high value okay and that's why this is a second technology one is processor technology or logic based the second is memory based technology both for DRAM as well as SRAM but also DRAM and third technology which is taking over now because most of the problem started with SRAM and DRAM was DRAM is a dynamic so it actually loses data after some time of use or even otherwise and SRAM of course is a power device all-time so it loses data and the power is gone gone away so we came with e-square PROM or EPROMs which can be stored out without power but the problem there is you can't put large amount of data on those EPROMs you can probably increase the memory but then they access time time taken to take data out is very high in ePROMs or e-square PROMs so the effort now SRAM is the best because it's very high speed memory access times can be few nanoseconds whereas in the case of flash it may be milliseconds in earlier ones now of course we are going to micros so we are looking for even faster time memories and we want to erase them in one go it's called flash so effort right now is to make flash memories which are closer to DRAM speeds if not to SRAM speeds okay what is the advantage then I don't have to power them all the time okay and that's something great it may happen if it happens SRAM market will go away or DRAM market may go away but this is still far off 16 years from now 10 years from now I'm not there typically this is the processing sequence maybe we'll come back again and discuss but this is what it is I'll come back it later so just to give before we quit is the process steps which are the main processes which we use in this whole course are the following that you may note down okay normally we talk about short forms a chemical vapor deposition is called CVD okay so there will be some processes called using CVDs then there will be processes for oxidant oxidation of silicon or any other material they are created used for insulating dielectrics okay CVD and oxidation insulating die they create dielectrics that is silicon dioxide now I tried many other things half name oxide, zirconium oxide, germanium oxide all oxides can be deposited or by oxidize the material itself okay on gallium arsenide I cannot create silicon dioxide I will have to deposit because gallium arsenide is a different material so you need depositions and you also need growths okay if I heat silicon at around 850 and above or any other process maybe low temperature I get silicon plus oxygen Si plus O2 is Si O2 and to a great surprise sand Si O2 is sand and it is third largest abundant material on the earth that's why silicon will stand because material infinite water and sunlight these are the other two third largest material available on the crust of earth is sand okay and you create silicon out of the sand process is damn costly okay but that's what it is the second process of interest which is the most important process right now which allows you reduction in sizes some 90 nanometers 65 45 32 now these are called nodes okay these numbers they came from simple idea of more you divide it or multiplied by 0.7 19 to 0.7 is 63 so we say 6 next node is 65 65 into 0.7 is roughly 45 45 into 0.7 is 32 so these numbers essentially are 0.7 down okay however this is all humbug now because what was thought as a technology limits has nothing to do with this number now 16 nanometer process is using 9 nanometers of gate oxides okay so there is no scaling in that order but nodes are still called that way okay so 11 nanometer technology will have actually get length of 7 nanometers okay 7 nanometer technology may have 3 nanometers of course 3 nanometers is worrying some because one monolayer of silicon is 5 and strong okay now that is 0.5 nanometers one knows how do you create less than a one monolayer at least one atom so we don't know what so we will do something else I cannot create silicon dioxide then what else so circuit people give some hints so we use that technology so lithography is separating the two lines okay generally actually shine light so light wavelength decide how much separation because the optics has its own laws so the minimum feature which I can separate or create is essentially the limit of lithography and we are gone from optical lithography even now we are doing it to some extent electron beam lithography external lithography and finally ion beam lithography so right now we are still working on optical lithography even for 16 nanometer process which has a wavelength of light used is 190 nanometers and still able to actually get your features of 16 nanometers which is great we will tell you this is what lithography is all about then there is a PVDs physical vapor deposition you can heat the material in the vacuum either evaporate the material or sputter the metal we will see thing goes then you have to remove certain areas to create different parts P channel P area N area oxide non oxide so you need to etch there can be solution based wet etching or dry etching by plasmats so to pattern the area then there will be impurity incorporation which may be due to diffusion or implants and they need they will create defects you need some thermal processes called annealing which actually modified modifies the material itself P channel P device N device or others and finally we will require many instruments for both optical as well as electrical measurements which is called characterization in fact IIT Bombay has the best characterization lab in the most Indian universities IITs IIC no one has as good a characterization lab as we have and we believe that that is good enough because we can get devices from somewhere and can still characterize and still publish great paper making small diode is not easy but characterize anyone else's device is very relatively because you know the process okay we will come back to it later the first question was asked is transistor is a good switch and if it is not technologically what do I do that it makes a good switch okay thank you very much for the day