MIT.nano Seminar Series: Ahmad Bahai

it's great to be here and it's great to be part of MIT community I've worked with many of you different capacities and I look forward to collaborating with more of the faculty members and students and programs in MIT so today's the I guess is the first talk of the nanofab in 2023 so it's my honor and pleasure to be here and I'm glad we have some attendance in the room not everybody online because this presentations online is really doesn't give the speaker at least the full satisfaction of getting feedback from the audience it's just you're talking to your screen most of the time so today I'm gonna talk about not knowing exactly what background of audience and this sequence and area of our interest is I wanted to talk more about what's happening uh in the semiconductor Innovation and how it enables a massive amount of innovation across all disciplines in electronics and electrical engineering across the board these days a few years ago I had to explain to my family what I'm doing for for living but now you don't have to do that chips is becoming such a popular thing in Washington when President holds a wafer in his hand and all that stuff so there is no need to talk about that anymore but also it shows the amount of value and strategic importance of innovation in semiconductor as a geopolitical Advantage as well as the foundation or the Bedrock of the Innovation for many fields so my first few slides I'm going to talk about how before getting to deep technical side I'm going to talk about how the innovation in semiconductor is important and how enabling it is and how exciting it is so let me start with a couple of charts it's amazing that people thought that they explained is there pointer or should I use maybe I use just mechanical pointer it's not there's no Semiconductor in this one but uh because this one yeah actually this one works too and this screen maybe it's not but anyway if we thought that a semiconductor or or chips in general has seen their Peak before think again because we see that in the past the big disruptions in chips was like by one application when Mainframe was started or PC and internet started and then Mobility handsets started but the next one it's not the one thing that is driving the Next Generation everything around you I mean I always use example of your doorbell used to be just two metals touching each other now there is 35 worth of semiconductor from Wi-Fi and Ai and radar and everything around that look at the uh everything around you from Automotive just give you an example a typical Automotive had about 300 worth of semiconductor today I was talking to CTO at one of the high-end car manufacturers in Europe just a couple of days ago 2500 of the Semiconductor in the car and 200 clock references including a cheaper scale atomic clock in a car so you can see the level of electronics that is going in everything around you and that's what is driving and of course data the fact that you have massive amount of data available and how to process them and how to take advantage of them just to give you an idea the expectation is that this is going to be 1.3 trillion dollars Market I mean this is number three or four U.S export after aircrafts and oil I guess and it took us 50 years to get to 500 billion in next 10 years we are going to go to 1.3 billion so you can see the Dynamics of what's going on in semiconductor and just to give an idea as a Vladimir said I'm part of this government activities on on chips act here is how many organizations the government are involved in semiconductor one way or the other from National Labs then DARPA and so on and two areas that are driving Innovation today in addition to consumer electronics consumer electronics always tries Technologies first because they tend to be more uh risk-taker because the lifetime of the Consumer Electronics is shorter but also automotive and data centers are pushing semiconductor to the edge in terms of performance and what we need next and the fact that without more disruptive Innovations we cannot offer enough machine learning and AI capabilities to to handle the data so here is another chart that I mentioned an internal combustion is about 300 that was the electric vehicle without Adas with the Adas and machine and and autonomous driving we're talking about about two thousand dollars and interestingly by 2030 as you know no new internal combustion car can be produced in California in Europe and in China and many other places so you can imagine annually about 100 million cars are manufactured built if 30 percent of them or 40 percent of them or all of them become electric is 100 million 100 million times two thousand that's another couple of tens of billions of dollars of semiconductor coming to the picture so it's exciting uh I think the next breakthrough in Automotive in drones in everything is when the Battery Technology goes beyond below 100 per kilowatt hour that is the Breakthrough because then the cost of electric stuff is going to be same as internal combustion it only subsidies and everything is going to be battery driven from flying objects to driving to robots on the roads to deliveries and everything around that [Music] here is the chart that is very interesting to show people who are talking about sustainability and environmental issues without semiconductor presence in this grid our energy usage would have been something like that and with use of this smart grid we think that the energy usage we are going to say about 3 000 billion kilowatt hours by making the grid more intelligent and more semiconductors in the grid and that is talking about high voltage semiconductors like I'm going to talk about gallium nitrite silicon carbide all the way to controls and data and machine learning actually space also is going through a big change I call it new space or democratizing space remember satellites used to be like at least a refrigerator size maybe multiple refrigerators flying over and a huge amount of weight and propulsion the new ones are mostly are running into Leo low orbit satellites and low orbit satellites uh of course the advantage is about 500 kilometers 300 miles up there so you need a much lower link budget to get a signal up and down the beauty of that is it's closer the difficulty is that it's not the same point so you have to track it so you have many of these Leos running around so you have to track it handy off to the next guy and so you need a system that can track these devices I remember for my PhD I did one of these systems that cost about a million dollars today is about hundred dollars and the reason for that is we can put 1024 phase array beam and forming antenna and a single dish I bought 20 inches and have these massive processing power to do this phase array and track these satellites and offer very high data rate so it's a fascinating opportunity also for for those who are working on machine learning and RF and by the way another beauty of Leo is that the radiation hardened requirements is much more relaxed than Geo and that makes it easier because once you wanted to make a hard radiation hard semiconductor you spend a lot of time in TI I should have a department doing that but it takes cost and complexity so so that was the background to get you excited about the fact that Chip's Innovation is driving almost everything in every discipline not just Electronics but every aspect of our life but clearly we need to talk about Beyond traditional scaling because traditional scaling is getting to the point that just by doing the better lithography you are not going to achieve all the performance that you need because you're almost at the limits of you know the add-on per transistor and the fact that you are limited by other factors as I showed that you know the number of transistors doesn't mean necessarily higher performance so the future is even more exciting because in the past we always hope that the next node of the CMOS is going to solve our problem if you don't have enough speed or enough transistors or you can't afford so many transistors wait for two years next node is coming it's going to solve most of these problems then we got to the point that interface became the bottleneck no matter how many transistors now you cannot get any smaller with the same approach of just relying on lithography put aside the fact that the typical lithography at five nanometer is about two to four hundred million dollars and then cost of the designing the system a typical AI GPU cost about uh billion a half a billion dollars to to from the design to manufacturing so these are massive amount of energy design investment and the fact that what's next and we cannot rely on that one so what is interesting and what is next is no longer one vector defined in the future it's multiple vectors and these multiple vectors need to work in harmony to create that kind of performance Improvement that we expect in the future so let me give you a couple of examples on the materials I think one of the most breakthrough opportunities we are facing is how to come up with materials that in addition to Silicon not in replacement to Silicon can offer exponential performances and I have many examples of that I briefly talk about some compound materials 2D and 1D and some metal oxides and other materials that is not new people have been looked at looking at it some of them are but not all the point is that most of them are don't have the same manufacturability that we are so good at in Silicon and that's where we see [Music] it is and we expect that if it is going to solve all the problems so we can see gate all around a complementary effect bi-directional fetch for power applications and so on all of these are becoming very interesting device approaches beyond what we are doing traditially I went fast okay so how do we go back okay and I had some animation too packaging is becoming one of the most exciting packaging by the way was one of those areas that universities didn't bother having a course on it because it was not cool for research you cannot you cannot it's considered industrial back in the line but reality is that once we get to the limits of Monolithic packaging and it's not just putting two chips next to each other call it MCM today in most manufacturings in semiconductor big companies uh there is a penalty for MCM because of a lot of pecan place involvement you need to do manual kind of like uh almost manual it's not like way for level processing the future is how to do packaging in a way that is as efficient and it's not limited by the parasitics being it or memory or RF and source and that's where of course there is a massive amount of investment from government DARPA has these two and half and 3D chiplet is becoming now a name of the game and the beauty of that is there is no some standards being developed for chips to talk to each other so different manufacturers of different silicons or different semiconductors can integrate it and one of the most exciting Parts I'm going to spend a little bit of time on it is back end of the line processing technology that really are compatible with silicon in the sense that you can do all the processing in 450 degrees below so you don't mess with this you don't mess with the transistor structure on your on your chip you can build new structures back end of the line post metal and that is something that is very exciting and also on top of all of that how to go beyond just one Neumann architectural processing just memory processor iOS traditional and there are some really interesting approaches on accelerators that can improve the performance especially within memory coming that you can do some of the processing within the memory so you are not limited by the bottleneck sub-memory and processor perfect I remember the first book I I read about this very interesting book was in 1981 doing this analog signal processing to model the brain uh I think today we are in a much better position to we don't know still how brain works so I was wondering how we can model the brain when we don't know how it works but the point is that can we get the efficiency of the brain processing I think our brain depending on the size of the brain Burns about 8 to 20 watts uh it's about two percent of the body's weight again whereas for some people are smarter than others but uh but it burns about 20 percent of the energy so extremely efficient billions of neurons talking to each other at that uh 8 to 20 watt power how can we model the same kind of circuits in a mixed signal environment between analog and digital and of course last but not least how to use uh randomness of quantums to model random things in life you're dealing with a random word I mean random everything around us from the weather forecast for tomorrow all the way to drug Discovery and everything around that so if the randomness is built into the world around us how can we take advantage of randomness of quantums to model this without going through zeros and ones and that is the beauty of that except that you need cryogenic situations so that's the area that you see a lot of research going forward I love this term this quote from Heisenberg one of the uh founders of quantum mechanics for us we spend a lot of time looking at periodic tables which element is going to help us to do the next big thing in transistors and switches and power and he says that don't worry about it because they are not real they form a world of potentialities that's what Quantum language is and the possibilities rather than the facts so that is interesting I mean but I take it in this talk more metaphorically I'm not going to use that as a non-real stuff because we have to build them the Vladimir can tell you you have nanofab for that so but the reality is that the amount of technology and materials that you need for these Technologies are amazingly moving fast so let me give you an idea when in 1985 that was about 30 years or so after transistor was invented these are the elements of periodic table we use for Semiconductor here is what we're doing today almost half a periodic table is being used one way or the other in semiconductor design and it's not just silicon I mean silicon is there but not just silicon and all of these by the way need to be manufacturable because if you're dealing with one trillion Parts they have to be reliable and you need to make sure yield and manufacturability is high so we already are occupying more than half of the periodic table and I wanted to emphasize a couple of them for the sake of discussions we have today so what are the Silicon plus opportunities that in the future near future I'm not talking about 10 plus I'm talking about uh next couple of years silicon germanium has been around for a while the compound material we see tremendous opportunities on SOI for RF and for photonics you know in photonics Silicon can go so far about 900 nanometer wavelengths after that becomes transparent so you need to go to Germanium and then the question is how to deposit germanium and silicon to optimize it for photonics at the same time relax the alignment of the fiber optics to the device so these are IO plus performance and plus the fact that you can integrate multiple devices by the way for RF also silicon germanium today can offer performances that in the past was only for three five materials like can and yes of course again and guests have also made a lot of progress three five materials I missed one I just want to make sure if you notice uh it's a three five materials uh these materials have been around on different substrates in fact the small wafer size four inch can substation substrate has been around in using defense quite a bit what we see is a huge difference is that now Ganon Celica takes advantage of the productivity and cost advantage of silicon processing but using gap on the as a device on top of that and of course algan and the Epi that you need to grow to to uh to buffer the device from the substrate to different materials also a silicon carbide gallium arsenide indium phosphate they have their opportunities to impact a lot of Technologies especially when it goes to terahertz technology high power uh I don't know if you have seen the some of the updates on this massive now because when you go to terahertz hundreds of terahertz the path loss is very high and you may complain why you have so much path to us but the beauty of that is when you have so much path laws then by Nature it's secure you cannot tap into that because attenuation would be very high so now there is some Advanced projects to have these drones in the air massive number of them that can talk to each other in in hundreds of terahertz and these are by Nature more secure but then you need to have some power and you these devices especially again for power amplifier as well as for power switches is a very exciting so and also on power management we again uh special again is uh is a game changer and the reason for that is power devices don't benefit from scaling necessarily in fact sometimes they don't need a scaling or they're not they are they can't scale because you need a voltage and and power density so obviously when you don't scale then your switching frequency will be limited because because now you're dealing with uh you know the nodes that are not switching your high speed as high speed as digital nodes that's where again comes with the picture because again you can have higher electron Mobility switch at higher frequencies and when you switch at higher frequency what's the first implication your passive devices are going to shrink and the power density is going to increase significantly so that's why we see that Yan is now is already there in fact I was in Best Buy I see the section says that Gan inside I mean that's very interesting so I think so it's no longer just a research part but we need to do more to improve the power density and possibly voltage and integrated with drivers and the rest of the system silicon carbide is now the main power device for anything kilowatt kilowatt and above and we see that Automotive is a Big Driver of that I don't know if the topic is hot or the temperature so do you want to I mean is there a problem there on the temperature I think because it's a little bit too hot in yeah yeah and also photonics photonics on Silicon bringing more Photon photonics closer to Silicon is becoming a big deal because of the fact that uh almost the data rate that you get out of the chip chip to chip or work to board or rack to Rack almost we are running out of steam when it comes to Copper and the fact that now you're talking about 400 gig and Beyond per photonics needs to get closer to silicon and that's where we see some technologies like uh Indian phosphate and others are very important and metal oxides in another area that people aluminum acts right and many other oxides that people are looking at for power devices and higher density power devices so you see that massive number of Technologies and and devices that can offer performance Beyond silicon also 2D and 1D material uh uh TM TMD materials uh molybidium sulfide new salinide and tungsten and platinum uh they showed some amazing performances all about that later still it's a very researchy topic but we think the beauty of that is you can build it on top of Sima so you can have your own uh chip doesn't matter what technology and on top of that you build the structure for RF or other sensing modalities that you wanted to address and we see the performance is quite promising or even technology like CNT CNT has been around for quite a while I think we need to look at CNT in a slightly different way CNT if you expect to use CNT the way that you are using silicon today how long has it been that we're looking CNT 10 15 20 years not no not manufacturable not as manufacturable as silicon and other uh Technologies like Silicon germanium but there are new ways to look at CNT I'll briefly talk about that to plant the seed but the fact that Purity and reliability and performance you need to CNT should be looked at differently from just traditional silicon and have to look at it more statistically so a statistical reliability not just kind of perfect reliability and that lends itself to some applications that can really benefit from high electron mobility of these devices also for sensing and passive devices building mems materials on top of silicon I'll show example of that adaptive materials these are metamaterials that you can really engineer them in a way that they can have a special properties for sensing applications and also for RF applications so you can see that the domain of innovation is much bigger than just the one material and one approach to device the challenge is that when you have so many vectors obviously your dilute your resources to support all of these how do we do a different approach to device design Ed so that it's manufacturable in less than 10 years because history shows that a lot of these new devices you can show one sample in the lab publish a great paper it takes another 10 to 15 years before it becomes manufacturable so that's a topic that would be very interesting to use machine learning to model some of the manufacturing problems early in the game and see how we can we can benefit from new approach to to material discovery so here is the chart this uh transition metal let me see if I can spell it one more time but die call gogenite materials and these are like kind of nutrition is this column look at the performance this is bulk and this is the 2D some of these are better than silicon and some of them even want the carbonara tube is almost a performance of s and indium phosphate so you can see that electron Mobility carrier Mobility is quite promising and it could be a back end of the line process in many cases the question is can we make it reliable enough to offer as a kind of and that is the beauty of this what I call it uh uh Next Generation heterogeneous devices that doesn't mean just that is beyond way for of course wafer bonding diet bonding Nano tsvs and all that stuff are important but here is more like uh One Step Beyond that looks even very promising and exciting to combine this on the integration and heterogeneous integration chiplet is now the name of the game because you have to bring different Technologies together for performance for applications that you're looking at and there are some advances on wafer bonding die bonding monolithic 3D inner posers I mean you're familiar with all of them and I think it's gonna be more of that but more interesting research part is related to how to do Post CMOS fabrication so that you can benefit from scaling at the same time add more functionality on top of that in a low temperature below 450. so let me give you a couple of examples of video oil that was very exciting and is exciting this one we did it in TI many years ago it took us much much longer than expected but it's amazing that is like a DLP or a digital light processor which is really totally uh post-sima so we take any CMOS bass CMOS and we build the structure on top of that just to give an idea two million meters on single chip each mirror independently moving with billions of Cycles reliability well if there is a projector here most likely is doing that on a stereo so so these are like a lot of issues and with distinctions and all that but everything can handle it with pretty much the most compatible processes and these two billion 2 million mirrors can interface with the processor under that to achieve a combination of advanced mems processing and AI machine learning to give an example we challenged some of the researchers in other universities that can be used this to create some unique capability of the headlights in the car so headlight today or in the past was one beam one light one Beam by using dlps in a headlight you can create 64 000 independent beams each beam can go in a direction and the sign and the road and The Pedestrian whoever but one challenge is that can we do something even more fancy can I improve the vision of the driver in a rainy condition we call it technology for rainy day and it's Unique is that you have to detect the drop of the water and steal the beam in a different direction and you think it's not possible so I have a video that we did it let's see if I can run this video so here's a DLP on each headlights we detect rain that is coming and steer the beam around that so you don't see the rain as much so you can drive safely yeah excuse me can I stop it somehow so here is the projection of the arrows on the road so you don't need to see the arrows on your GPS it's going to be on the road but here is the situation it's rainy situation so we detect each drop of rain and cancel it so there's no rain now so so that shows how much real-time signal processing and machine learning you have to do and by the way these are 3D stuff that we do with the same technology so you don't need a screen so many screens in the car it's like the holographic screens that you can have any any any area in the car so so the number of I don't remember exactly how much GPU processing power we needed for that I can dig it out in my slides but you can imagine for that kind of processing you need a massive massive uh GPU or uh uh processing engine and you need a massive number of mems and they need to be highly coupled each other because of the interface limitations you cannot have it on two chips and here back in the line kind of processing of this technology two million mirrors on top of this managed to do that and that is the complexity of real-time processing because there is nothing more real time than rain right I mean or kind of the way that is going around and randomness of each drop so another example of is that what's happening in the what we call it a medical field we can have no massive array of medical sensors all built on top of CMOS with some variation of the ipswet ion sensing fat that can detect or it's like open gate transistor and the ions I mean it's like pH detection so you have a polymer a pathogen sits on top of the polymer changes the ph and then you translate it to voltage so it's pH per volt period variation per volt now you can have massive number of these transistors a single chip can detect all kind of pathogens by back end of the line polymer printing on top of that so you can imagine I don't use lab on a chip because it has a bad connotation these days because of what happened in Silicon Valley but it's going to be type of like chemical sensing chemical sensing is a massive number of chemical sensing could be hazardous materials could be so you can see all of those big machines or functionality of big machines can be in a small device here is what DARPA published as far as the target performance for powered fets for RF and you can see that we have a long way to go beyond what today we can offer of course silicon germanium is a big player we can see that indium phosphate and gallium nitride are going to be a big technology of choices on this one and RFC mass now is limited by power and of course limited by the fact that you need to have a enough voltage on top of the device to get the power out of it so I just put by CMOS and CMOS but you can see the three five materials are playing big deal on this to get to the performance that defense and error space need another one is metamaterials as I said metamaterials today uh there is a project that you know in the smoke detectors today just detect smoke right I mean with this LEDs and you know reflections of LEDs but now in the future you want to take the particles that you have in the chamber and that's where the metal materials you can process the surface of the Silicon or create these resonators with different resonance frequencies they react to different chemicals we also have a spectroscopy in every smoke detection you can detect the material and the gas that is coming to the to the chamber and these are very affordable and very compact so in order to have this kind of material research going we need new tools that's why we have Nano files you know and these new tools are amazingly powerful today I mean the way that uh semiconductor industry for last 50 60 years has driven innovation in Tools in edas in materials in devices it's whole ecosystem is fascinating I don't know any other fields of the science in humans history to have this kind of ecosystem Innovation across the board to create a situation that you can have so much performance at so much so affordable cost so the machines that we are very familiar with of course lithography litography is I don't know if you have seen these machines the latest one it's the almost size of this room maybe it's not bigger but what I was shocked about is that the mirror can only afford one picometer variation in the height one picometer so if I put a matchbox stick on the face of the Earth that's too much variation you know it cannot be like that so you have to polish the surface of mirror to get that kind of picometer smoothness to achieve below two nanometer lithography capability but not everything can rely on this couple of hundred million dollars machines plus matchings becoming a big deal for some of these high aspect ratio post CMOS processing and they have come a long way very exciting and not as expensive mbes mbes for growth of Novel materials now I read the report on last year there is a new MB with the laser that pyramid as you can play with the angstrom of the depositing new materials and that is slightly expensive than than class managing but not that much so these are tools that really powerful ALDS ultra thin a few nanometers of oxide or nitrites depositing and AFM which used to be just observation tool now you can manipulate atoms one at a time pretty much so that kind of resolution that kind of capabilities enables us to look at these materials and some exotic uh proof of council like massless lithography or even printing printing now you can get a resolution of less than one micron that is with 1024 nozzles so you can do a lot of stuff on this thing so you can see that all of these are opening new doors but uh we also need to add to that some uh Manufacturing prediction or capabilities in early in the material Discovery because as you said when it comes to yield and throughput and manufacturability of these materials a lot of surprises show up when it comes to volume manufacturing and that's where we wanted to use some of these machine learning algorithms to see if we can we can preempt some of those manufacturability issues early in the game so my message is that more innovation in chips will enable innovation in every other discipline not just Electronics not just eecs department but everything from sustainability to Health Care crisis to everything around us and I think it's just the beginning thank you for your attention [Music] Ava thank you so much for amazing talk um uh the floor is open uh those of you who are online are more than welcome to send your questions via the Q a button on the bottom of your Zoom screen um I'll make sure I go and read them I will uh to start us off I will open the floor uh to the audience in the room please raise your hands uh AMA this year he can ask you any question that uh you pose I'm very much sure yes let's start first um uh thanks um this is a kind of a journalist question I guess um there were three areas where I found myself wondering so um how are you going to be able to do this one was materials where the material is going to come from the second was where is all the energy going to come from I mean you're talking about producing power density but like there seems to be a lot of power and energy and implicated in this and then the other thing when you talked about the drones I was thinking what's the FAA going to have to say about that you know it seems like you know we're going to fill you know space with drones yeah yeah so I need to give you a journalist answer or research answer I mean that so either way yeah on the on the materials actually the downside of it is the most expensive part of innovation is materials because it takes longer and as I said taking it from Lab all the way to Fab is a relatively long journey just to give you an idea for us to get so good at steam us it took us about three decades for the entire industry to make it so high yield High uh reliability manufacturable device for some of these power devices like Yan and others almost two decades so we can not afford to have this kind of lawn because that means a lot of a lot of waiting stuff so the funding side I was talking to Vladimir about that that's where universities and government funding because in the past a lot of material research was in places like Bell labs unlimited budget when I was there it was like you can work on anything and once in a while you hit germanium and make a transistor out of there but but uh now we need to have a situation that under more realistic way with government with universities and with pre-competitive that's the key pre-competitive research behind the material part of it but also on technical side you cannot do the same way of material Discovery in the past that we get super excited by one of material because at the end of the day you're talking about 1.3 trillion Parts

a year type of semiconductor industry so you need to look at manufacturability early in the game how that's a interesting open question and I I mean there are some ideas and I have some ideas but uh needs a lot more work so yes material is the big part foundational most expensive and most critical and I think we need to do that in a different way as well especially with all these machine learning capabilities and forecasting chemistry and physics of materials and all that stuff on the energy side obviously if I have 100 million cars electric each car is gonna have about 100 kilowatt hour battery multiplied by that that is going to be a few power plants that needs to be dealer factories for the batteries so um I'm quite optimist about energy because the consumption we have made a huge progress as far as efficiency you saw the smart grid chart we are saving a few thousand billion of the kilowatts of the power but on the generation also other than solar and wind that always is going to be a part of the game there is a massive amount of research today on Fusion as you heard and fusion has been always the technology of tomorrow but now we see that startups with Fusion technology working so I don't think it's around the corner next year but with announcement will just be for the holidays right in Lawrence Livermore is doable the problem is they spend 300 megawatts to create 300 kilowatts of power but that was because of the laser inefficiency and all that but but the the fusion is doable and there are some fascinating ideas I've been talking to some of those and it's quite impressed uh and of course vision is always there Vision uh I know that environmentally people are questioning that but now there is a wave that they're saying that's one of the cleanest in fact if you find a way to to handle the Wasteland there are a lot of discussions there so I think [Music] on the energy side you're getting much better semiconductor on the consumption and on the production there's some promising Technologies on the horizon right uh okay other questions yeah mark so you talk about the angstrom Asia you talk about very short Channel links or device sizes right and then you also talk about manufacturability and the problem of path to manufacturer goalie so how do you see things working out when in the lab we can do stuff with EB lithography direct right with throughput that's hundreds of times hundreds of thousands of times slower the manufacturer and most of which is not happening in the U.S in terms of Link scales below what 20 nanometers um actually they I mean there's factors in Arizona and others played for three five nanometer and three nanometers so that's exactly my point you know you can prove the concept with MBE or with some you know maskless techniques that you wanted to play with this only material and that's a big step to show that it's doable you can get the performance the big question is that and that's a separate research topic that what's the manufacturing capability because in the past it's like you put it in the flow and then you see what's not working and if you have full flow here and then we give feedback here that model served as well if you have infinite time and budget today just to give an idea if your data usage is you buying 20 in one week you need a lot more processing power and power density and high-speed interfaces in a matter of years not decades so um how to make it manufacturable here I'm really glad the awareness of chips [Music] strategic value is here now and so we're going to have capabilities to do that but even that requires a machine that you go to a couple of hundred million dollars of lithography and math exactly Mass cost is very big Mass cost is staggering high in fact so uh that's why we need to have a way that you can share some resources to explore these new ideas but those days of okay I proved the device under mbes and now you'd make it manufacturable it's not going to work it has to be a tighter look around that what's the answer I don't know exact answer but there has to be different thank you for the talk uh so I have a quick question uh and a more sciencey one the quick one was I noticed gold and silver were part of the uh elements that are used in CMOS now it's kind of surprising yeah gold or silver is very important for medical applicants in art materials so they want wall so that you know these ions and chemicals don't interact with so it's like and by the way some of the your tactics pay for bonding also using gold because of the fact that they can create a very special bonding between two Wafers I see so we don't use gold on this you have to okay yeah uh the the more uh technical question with video that you showed uh of where you have a car and it's right so is the correction happening uh by the dmds or it's happening on the screen what you're seeing is being corrected we have two versions so version is that on the screen we regenerate the we regenerate the image but also we're working on kind of steering and these are all experimental it's not like you know it's steering the beam around the exactly it's like Face Diary all right question in the back thank you so much for the talk um one question I have is um how much work has been done and where do you think work will go in terms of using biology to generate products and uh and chemicals and other things like that because we've been using machine we've been using animals and plants forever as factories so perhaps there's something we can do on a nano scale but more um with more um shall we say practicality in in terms of getting things to do exactly exactly what we want this one but we don't really know how the brain works actually we know a fair amount about how the brain works we know very little about how the mind works so you still have a lot of luck over there yeah absolutely yeah actually on the first question uh on biology there are two aspects of that it's one of the most exciting aspects of uh semiconductor applications there are two angles to that one of them I'm more familiar with and the second one less familiar but I have some exposure to that first is the biological sensing modalities that is becoming very uh uh practical with some of these new polymers and metamaterials that you can deposit on Silicon to offer a massive array of sensors for biological pathogens or any if let's see if I can go back to one of my slides single molecule detection that you are working on with some uh oh sorry maybe I'm going wrong way here let's see uh okay this one here yeah so really here is that like uh with this ald deposition we can have this monolayer surface functionalization to detect a single molecule Spike and some of these receptors so in other words the semiconductor sensitive enough can detect the variation of the pH with single molecule it's like a single Photon detectors in the APD is another single molecule biological so that part of it is going really fast and exciting and also there is tremendous amount of work how semiconductor Works inside the body and that is another area that is picking up quite a bit so the sensors or therapeutic aspects of semiconductor we see tremendous opportunity within next 10 years to the point that you can have a lot of therapeutic and medical work on sensors on the body or minimally invasive like under the skin with micro needles I think there is a very successful part today in the market that we worked on it for quite a few years it's a glucose meter that can every 10 minutes gives your glucose reading it's micro needles you don't even feel it on a Band-Aid it's Libre Libra Libre that you can buy it in Pharmacy and be the prescription and you can imagine tomorrow you can give potassium which is important for some of these kidney and dialysis we can have other chemicals detection because once you get under the skin literally that you can have a lot of capabilities offer this biological sensing so that part I think you see a huge variations there is another aspect of convergence of biology and semiconductor designing transistor in a Petri dish that part still I'm holding my breath there I mean I don't know I think if I because you're trying to figure out how to use biological interactions with uh to create a semiconductor like an amplification or switching uh it's way out there I mean that's my my assumption I think maybe I'll be happy to be surprised by by progress out there I've seen a lot of work that one but uh understanding conduct we don't know some of the quantum go to the biological material that's a very complexing so that part of now on the brain side yeah I mean I need to be mindful about the brain you said mind is so so mind yeah mind or brain or cancer or whatever we call it uh we had a project with some really good neurobiologists and the way that signal processing happens in the brain seems to me it's still an enema how these spikes creates without any a to d and they need to air you know I mean one of the fascinating I'm an engineer so I look at the biology from engineering these hair cells in the ear that you know the stiffness so when the air hits the ear they do fft based on the stiffness and then send the fft signal to brain with the reduced complexity so a lot of data compression and stuff is happening uh I need to read more about biology to understand what how it goes but and study more that but uh these are two different aspects and two different timelines from my viewpoint so um I'm gonna just commandeer that questions because time is running short I want to pass a few questions from online sure first one is uh the future of transition but uh metal oxide materials yeah uh in in the future of electronics similarly topological materials or materials having a spin um what is the likelihood of seeing those um in short term available in Technologies we're going to be using yeah as you know there is a lot of good work there I have a slide and it's not my presentation but the number of papers published on this transitional material over the last few years has exponentially gone up uh there's a lot of centers spending time on characterization which is very exciting the part that is missing really as I said is manufacturability aspects of it and so I think a really cool university project would be how to preempt or how to predict some of those manufacturers early in the game while we are doing it because as far as maturity of research and the device itself we see that it's very close but as far as manufacturability I think we have a lot to do and I assume the similar answer is for topological materials the one opening for that is an example CNT 10 years 15 years ago we've seen massive number of papers on CNT people used it for logic and memory then the question was is it manufacturable is it worth it and all that stuff maybe we should look at those materials in a different way when it comes to yield and reliability I'll stop there because there's some really interesting stuff all right then allow me to ask the very last question for the afternoon um you have presented many many different technological advancements uh they are leapfrogging over the possibilities of what silicon can do today what's the likelihood that with what the chips Act is offering to the United States read to reimagine ways of using such technology to truly do a leapfrog what's the likelihood the US will actually be in a position five ten years from now to again reassert dominance in the entirely new way of thinking what microelectronics is that's a very good question I think uh let's let me answer it this way if we don't make that LeapFrog progress we are gonna hit the wall as far as performance we need to run the economy that is based on data and intelligence and autonomous system literally we're getting to a point that you cannot handle the power the processing the interface so it's not um luxury it's a necessity we have to do it and usually we are very good at when it comes to Sputnik moments you know that you know okay if we don't do that we are going to fall behind we are going to lose it's going to hurt the entire economy I mean if you can imagine you cannot offer a semiconductor for data centers that can hundred thousand times have more data processing power Automotive that can have hatch higher power density so and as you know uh it's one thing that we do research for the sake of research another thing that if we don't achieve that performance we are going to hurt economy and to some extent the Strategic political Advantage so I think the level of awareness is there and because of the criticality of this thing we'll see on that note let's think about one more time foreign

2023-02-18 20:54

Other news