New Disruptive Microchip Technology Explained. ASML is out?

New Disruptive Microchip Technology Explained. ASML is out?

Show Video

modern life runs on semiconductors and TSMC is  arguably the most important fab in the world it   makes around 90% of the world's most advanced  chips now TSMC has announced a new technology   that will take us beyond 1nm as a chip designer  I can tell you this is a big deal for the entire   industry so in this video I will break down the  genius behind this new technology what is the   next big thing in semiconductors what's going  on with the TSMC stock price and why we will   rely less on ASML lithography machines in the  future when TSMC was starting out in 1987 their   first chip was manufactured in 3 micrometers  technology that's 3,000nm just think about it   and recently they announced that they are ramping  up the production of 1.6nm technology this means   that over the last 40 years or so they managed  to make their chips 99.99% smaller and the next   big breakthrough is just around the corner and  it involves new materials and a cutting-edge  transistor architecture CFET but to truly grasp  the beauty the genius behind this we must start   with transistors first if we go back up until  around 2012 the basis for all silicon chips was   so-called Planar transistor and we love Planar  transistor because this one is very easy to   understand it's just like a tiny electronic switch  which controls the flow of the current imagine a   pipe with water flowing through it if you place  a valve in the middle you can turn the water flow   on and off in a similar way transistor controls  the flow of electric current it has three main   parts source where electric current starts drain  where the current exits and gate which is used to   control the transistor when you apply voltage to  the gate it creates an electric field that either   opens or blocks the path between the source and  the drain so this Planar transistor remained the   state-of-the-art technology for about 50 years but  when we were approaching the 22nm process scaling   its dimensions further were no longer possible  what's interesting for a long time we were   reducing the gate length but further reduction  were causing the channel becoming too thin which   resulted in electron tunneling from the source to  the drain and that's bad imagine in the example   of the pipe imagine you close the valve but the  water is still leaking so basically we were not   able to control the transistor anymore and as  you will see today since then it has become all   about the gate the first step in this evolution  was to stretch the channel in a kind of Fin and   wrap the gate on three sides of it this new  architecture was called FinFET and this was a   very elegant solution because the gate wrapped  around the channel allowed for better control   it which means it could be faster switched on and  off quickly improving the speed of transistor and   allowing us to pack more of them per silicon area  these new transistors leaked less power when they   were switched off means devices can run longer  on a single charge in fact Intel was the first to   manufacture FinFET devices which hit the market in  2012 at the 22nm process node and one year later   TSMC also moved to FinFET devices in fact today  almost all high performance electronics including   Apple Silicon NVIDIA and AMD GPUs are manufactured  in FinFET technology now we are coming to the most   interesting part as of today the limits of FinFET  architecture has already been reached and now the   industry has to take yet another bold leap in  transistor architecture now we are entering   the truly 3D era of transistors and remember it's  all about the gate if you think about it the only   way to shrink the footprint further and get even  better control of the gate is to control it from   more sides here the genius is in turning the  channel on its side stacking horizontal nano   sheets instead of standing vertical Fins now  instead of growing horizontally the nano sheets   are multiplied vertically and this design allows  the gate to fully wrap around the channel so a   true gate-all-around structure eliminating leakage  and improving electrostatic control and good news   the first gate-all-around transistors are coming  in N2 technology is already entering the mass   production and first devices coming to iPhones  now let's talk tools even though gate-all-around   architecture was an evolutionary step from  FinFET where TSMC was a decade old leader still   many processes had to be readjusted and here the  general trend is that from now on EUV lithography machines will be less on the critical path and  other tools and noble materials will become   more and more increasingly important EUV tools  essentially create a controlled tiny explosion   inside the machine and use the high energy emitted  photons to print the finest features on the wafer   and this process is beautiful it's often compared  to tiny star explosions because there is a similar   process behind it it involves extremely high  temperature plasma generation similar to what   happens in the Sun if you look at gate-all-around  architecture the main challenge is to create the   channels and wrap the gate around it especially  when you can't directly see the underside of the   nano sheets that's where atomic layer deposition  and epitaxial growth become critical it's crucial   here to precisely deposit gate dielectrics and  metal gates around these intricate shapes and for   these steps Applied Materials and ASM not ASML  but ASM own the secret sauce what's more after   the deposition we have to edge away the material  in between the channels and here lateral etching   tools from LAM research are critical and finally  we need one more deposition step where the gate   around the channels is formed and here again the  tools from LAM research and ASM are critical so   this is the elegant solution behind today's  state-of-the-art gate-all-around transistor   architecture which is mostly about going vertical  and this will enable shrinking beyond 3nm what's   interesting we had Planar transistor for 50 years  and then we had FinFET as the state-of-the-art for   a decade how long gate-all-around will last after  roughly 30 years in the making probably five years   or so just one of many examples where we see  everything accelerating technological progress   is accelerating as Ray Kurzweil said we won't  experience 100 years of progress in the 21st   century it will be more like 20,000 years of  progress at today's rate i totally feel that   do you we all know that AI can help us now to  create things faster than ever and one of the   most exciting shift is happening in the software  development what if I tell you that I managed to   create a custom made web app with just one  click without a single line of code with   Hostinger Horizons you can turn your idea into a  fully functional web app using AI generated code   no coding no complicated deployment no third-party  hosting just couple of clicks and you're live here   is how it works simply type your web app concept  into the AI chat box and watch as the AI instantly   generates the first version of your app and in  case you need any changes you can quickly tweak   it with AI assistant edits when it's ready you  can instantly publish your app domain included   what you can do with it in just few clicks  is mind-blowing so whether you're a founder   an entrepreneur like me or a micro SAS startup  Hostinger Horizons gives you all-in-one solution   to build and deploy your software on top of this  Hostinger offers a 30 days money back guarantee   so why not to give it a try use my link in the  description below and apply my code ANASTASI10   to get 10% off your first month on Hostinger  Horizons thank you Hostinger for sponsoring   this episode now we discussed the current  state-of-the-art gate-all-around transistors   and that controlling the gate is crucial and we  are coming to the most interesting part the next   big thing in semiconductors how can we build even  smaller and faster transistors what's interesting   TSMC is currently working on two parallel tracks  horizontal and vertical let's start with a   vertical one so the genius of the new architecture  lies in details in the new stacking approach   instead of placing transistors side by side the  new architecture consists of a vertically stacked PMOS transistor on the bottom and NMOS transistor  on the top and this cuts transistor footprint by   half and keep Moore's Law going and this new  architecture is called CFET Complimentary FET   Transistor other companies calling it differently  but the idea is the same the next big thing in   semiconductors is a vertical transistor the bottom  layer contains a P-type gate-all-around transistor   called PMOS when a negative voltage applied to  the gate it creates a channel of holes allowing   current flow and the top layer contains  an N-type nano sheet transistor and here   a positive gate voltage creates an electron  channel enabling current flow both NMOS and   PMOS nano sheets are surrounded by gate material  forming a gate-all-around structure the sheets   are typically very thin around few nanometers  and narrow around 10 to 15 nanometers what's   interesting one of the key persons responsible for  the CFET development at TSMC is Szuya Liao she is   a director of device architecture pioneering at  TSMC and she joined TSMC in 2021 after 16 years   working in Intel where she contributed to all the  latest development from 65nm down to 4nm process   now at TSMC she's pioneering new device  architectures such as CFET while at the   same time looking into the integration of novel  materials which we will discuss in a moment her   recent paper describes the first working CFET  device and it's a pretty remarkable achievement   recently at the latest International Device  Manufacturing Conference TSMC presented the   first working CFET and they managed to solve the  interconnect challenge to interconnect the bottom   and upper transistors and it seems that TSMC is  leading CFET development and being ahead of the   rest of Samsung Intel and even IMEC the main  challenge with CFET is interconnect complexity   you know interconnect those tiny fine metal lines  that interconnect all the transistors and other   electronic components on the chip usually it's a  complex network complex metal network of up to 20   layers that enable data transfer and  power distribution across a chip and   in modern chips interconnects run mostly in  horizontal layers across the chip now with   CFET the connections must go vertically which  again increases the resistance capacitors and   add additional delays slowing the signals down  and increasing power consumption adding complex   vertical interconnects means more processing  steps new alignment challenges and potential   defects in wiring and it would probably require  backside power delivery and potentially even   backside signaling for the PMOS transistor not to  mention extreme thermal challenges that are coming   with this new architecture already now the high  performance NVIDIA GPUs generate several hundreds   of W per square cm of silicon and it's projected  to reach 1 kilowatt per square cm of silicon now   imagine what will happen when we pump 1 kilowatt  of power into this tiny multi-layer transistor   piece of silicon which is a part of a 3D package  it's going to be hellishly hot so CFET approach to   scaling is genius but it comes with a whole new  set of manufacturing challenges more processing   steps higher cost at the moment is expect to  reach mass production in roughly 2030 and this   will allow us to scale the transistors beyond  the 1 nanometer node as you saw in the early   days of the transistors innovations were primarily  focused on geometry but now the focus is shifting   now TSMC is pushing on the development of new  materials mostly focusing on the 2D materials   with the goal to use them in the channel of the  transistor you remember early in the video we   discussed the problem which arised when we tried  to shrink the Planar transistor the gate of the   Planar transistor further this is a limitation of  silicon that's why researchers are now focused on   other materials to push the performance further  it turns out that 2D materials are more robust   to the effects causing leakage and much easier to  control in general TSMC is mostly focusing on TMD   materials i will put them on the screen and these  materials have the potential to enable atomic thin   channel transistors but those are still mostly  in the research phase and still quite far from   commercial use and the main reason for that is the  growing process of 2D materials currently they are   grown on the sapphire wafers and then transferred  to silicon wafers and this is not scalable for   mass production to enable the mass production we  have to learn how to grow them directly on silicon   wafers in fact I have about three to four episodes  on this channel about semiconductor material   innovations covering 2D materials indium selenide  graphene and other materials so subscribe to the   channel now so you can enjoy them later on another  big shift worth mentioning in this episode is   towards material innovations and if we look at the  N2 process node and also all the innovations we   just discussed we see that the trend is that the  focus is shifting from lithography tools towards   other processes and mostly material innovations  to be clear EUV lithography tools will be still   important what I mean is they will be less on a  critical path towards future transistor scaling   over the past few years we talked a lot about EUV  lithography machines and that's because they were   crucial for transistor scaling and as transistors  were becoming smaller and smaller the interconnect   which connect all these transistors into  logic gates all of this had to also scale down   in fact transistor scaling was always outpassing  the interconnect scaling let's say at 7nm down to   3nm EUV machines were the most critical tool  for scaling because older DUV machines Deep   Ultraviolet Lithography machines which use longer  wavelength light compared to EUV machines were   not enough to pattern the finest metal lines  for the interconnect layers but looking at the   latest process nodes like N2 from TSMC its role is  decreasing it's decreasing because we are shifting   to CFET architecture which has its own challenges  and because we are shifting to backside power   delivery which is actually removing the congestion  of interconnect on the top of the wafer this shift   to backside power delivery is a big change for  the entire industry because till now all the metal   so signaling and power layers all were placed on  the top of the wafer now the idea of the backside   power delivery is to offload the power to the  bottom of the wafer so those metal lines which are   thicker and wider which are used to deliver power  they will have to go on the other side which will   free up the space on the top of the wafer and  this means with backside power delivery we can   reduce the number of fine-pitch layers on the top  of the wafer means fewer exposure steps fewer EUV   masks and lower cost and this is a big advantage  for going beyond 3nm while EUV tools from ASML   are less on the critical path TSMC remains the  most critical fab that the entire world depends   on so then what's going on with TSMC stock why it  has gone about 15% down over the last few months   one of the contributing factors is concerns over  potential US tariffs on computer chips that could   reach as high as 100% well I think it's a part of  Trump's strategy to encourage TSMC to build fabs   in the US and this is driving the stock down which  means shopping time because despite all these   concerns they show strong revenue growth about 43%  year to year which shows strong robust demand in   silicon chips now consider the high costs required  to building the cutting edge fabs and also the   costs of constantly upgrading equipment to staying  on the leading nodes no surprise that more and   more fabs are dropping out from this race you  know in 2007 we had 12 fabs manufacturing chips   at 45 nanometers and now we have come down to just  two fabs fabricating silicon in 3 nanometers and   with Samsung and Intel recently struggling with  yields it's really just TSMC leading the charge   so the role of this company in the modern economy  is undeniable and a few days ago TSMC announced   their plans to invest another $100 billion dollars  to expand its semiconductor manufacturing in the   US for once billions are coming in instead  of other way around that's a change and this   investment will be used to build three additional  fabs in advanced nodes two packaging facilities   and we know that packaging is so important for the  advanced GPUs and in general a part of the success   of keeping the Moore's Law going and they will  also build one R&D lab as you may know they're   already building two advanced chip plants in  Arizona with the first set to begin production   already this year and the second planned for 2027  and this is a huge fap it's hard to imagine how   huge it is it's around 1100x in size which is  roughly equivalent to 650 football fields and   this making this manufacturing facility one of  the largest semiconductor production sites in   the world anyway I think the costs of TSMC chips  will go up whether as a results of high tariffs or   higher manufacturing costs in the US let me know  what you're thinking in the comments as Stephen   Hawking said we are standing on the threshold of  a brave new world what we do with our discoveries   will shape the future of humanity clearly the next  decade in technology will be defined by AI and   high performance computing which are demanding  exponentially more computing power and the key   bottleneck is energy we are currently hitting  the limits of what our power grids can deliver   and what our cooling systems can handle without  power efficient transistor technologies our data   centers will hit the power ceiling and clearly  we need new materials and a breakthrough or two   in semiconductor technology to keep up with AI  power demands i think it's rather going to be a   complex solution which means noble materials new  architecture of transistors integrated cooling   technology including also backside power delivery  chiplet approach with photonic interconnect and   advanced packaging it all has to come together  let me know what you think in the comments   now check out our sponsors in the description  below to support the channel and watch this   video where I explain the backside power delivery  how it works and the secret plan of Intel or this   video where I explain the new photonic chip  which is on the way to data centers right now   and connect with me on LinkedIn all the links are  below thank you for watching love you guys ciao

2025-03-21 12:41

Show Video

Other news

The First Probe Zoom lens – Test & Review of DZOFILM X-Tract 18 – 28 T8 Probe Zooms 2025-03-30 11:47
ASUS WRX80E Sage SE WiFi II & Threadripper Pro 5955WX Review: PCIe Lane Heaven or Proxmox Nightmare? 2025-03-29 07:07
Bumpy Week for Nvidia, Tesla, Billion Dollar Space Bet | Bloomberg Technology 2025-03-28 17:01