PHYSICS of your PROCESSOR. Problem of the nanometer limit
HTML-код
- Опубликовано: 5 авг 2024
- We all know that modern processors are works of art with billions of transistors inside, but why is it so difficult to create them? Why have the technical processes of the last 10 years been pure marketing? Let's dive into the physics of our processor, what photolithography is and what difficulties its application has.
00:00 - Entry
00:47 - Photolithography and process node - how are they related?
07:23 - Extreme ultraviolet
10:33 - X-ray
#mycomputer #processor #silicon 
Наука
In the early 1980's I was using near UV light for making discrete bipolar transistors with 1 micron wide surface features. A 1 micron line is a fuzzy blur at those wavelengths. We were using alignment marks on opposite sides of a 50 to 75 mm wide Silicon wafer which made use of that blurriness. When you have two concentric rectangles, an interference pattern becomes visible between their respective edges. So the wafers could be aligned manually in 4 degrees of freedom (left-right, up-down, rotationally, focal depth) to an accuracy of about 0.1 micron at an alignment mark spacing of more than 70 mm. Mask alignment across that distance is therefore incredibly precise. Modern wafers are maybe bigger than 300 mm in diameter, and similar mask alignment techniques are still used at far UV wavelengths. So what this means is that chip surface features somewhat smaller than the wavelength of the light used can be produced.
Chip quality always varies across the wafer, no matter how precisely all parameters are controlled, even ignoring Silicon lattice faults and the inevitable residual contamination which occurs at all stages of production Even with simple transistors, one small wafer can yield a range of devices from very close to the theoretical limits to totally useless. And as you say, the higher the device complexity and the smaller the elements of that device, the more faults accumulate, and the lower good yields result.
Even with ion deposition, there will always be anomalies.
Wafer fabrication engineers have been known to refer to their skills as being very much like cooking; no two sessions result in precisely the same outcome. There is as much art as there is science, mathematics, and a wide range of other skills, along with years of experience in the production of high quality semiconductor devices. Their retail prices really are justified.
This is the most comprehensive explanation I have heard yet about modern processors. Great work -- Thank you :)
P
Lll
I second that.
Most underrated channel. Should have millions views
We're reaching a limit with current technologies in hardware. New technologies that meet the thermal, cost and conductivity requirements, and can be mass produced to take over from the current generation is going to bring with it some growing pains. Really demonstrates the importance of software to take up the slack in the meantime. New algorithms, better use of parallelization, etc. are going to be vital. Low level programming is far from dead and a lot of the open source, low level C dependencies, that has been the backbone of modern software for decades will need some rethinking or be entirely re-written. Poorly optimized software has been too reliant on increasingly faster hardware. The lessons learned by the old guard of software engineering are dying off, with preference of merely learning and using high level frameworks. In the meantime, our digital infrastructure is built on a 40-50-year-old foundation.
Yeah software nowadays is incredibly bloated
I just wait for the day that people figure out Windows NT isn't efficient enough because of its age and will switch to Linux
Many channels explain this matter only about quantum channeling
Glad you explain other things
you're the best sir
This is so fascinating! Thanks for all the work you dis putting this all together and explaining it so well!
This is information I didn’t know I wanted to know but I’m glad I found this channel. Great video
Nice info, excellent, hope videos rare like these and detailed come in future for us to understand simply
simply explained and with enough details for someone that has had few physics courses but continued on another path
impressive work mate
It is crazy to think that in our world something unpredictable in it's true nature yields predictable patterns just further outside its scope
Another amazing video, keep em coming!
please make podcasts this was cool and educational , your voice will will make you a lot of money.
Cool. Thanks for sharing.
very informative... love this channel
i worked in a wafer fab for 20 years...I learned more about it in 13 mins...nice stuff there!
Great video MK!
Спасибо, было интересно! )
Nice video
You may have also mentioned technologies such as multi patterning
And the shift to more of task specific silicon instead of general compute,
I think from the viewpoint of electronics we have to live with the structural limits of solid state physics. If we like it or not, the lateral limit is several atomic distances. If you go below it the structures become thermally unstable and you get a lot of throuble with electro-migration. The current densities in the metal layer structures are allready now in the order of macroscopic superconducting structures (wires and tapes). I think the limit will be around 1 nm. This is also the order of magnitude where we come to atomic clusters and the materials loose their bulk properties. The material behavior becomes more and more domiated by surface physics.
I am also impressed that the phototoresist itself is capable to depict such small structures.
Well done
From a new viewer of your channel:
Good info, I was expecting you to cover the Gate-leakage current too, which would become a headache when we reach near 1nm process node..
Very cool just being your average high school graduate made it extremely easy to follow!! I'm coming up on 50 and wished I had access to videos like this in my youth probably would of went farther in my education. Damn MTV been a drummer since I was 6 🤣!
Thanks!
Can the engineers even _control_ x-rays to the precision required for photolithography?
Its wild that chip manufacturing got so far that quantum uncertainty needs to be taken seriously
wonderful
What are the difficulties of stacking the silicon chips to make a 3D structure?
audio not synced
Time for a modern renaissance of efficiency, and programming that actually uses the tech we currently have... Stagnation should breed some creativity...
Agreed
No fluff straight to the core, excellent video as only eastern europeans can make.
What about a substrate that stands up to heat better? Diamond semiconductors might be able to clock much faster.
Bro do you know how to sync audio with video? At least check before upload next time
Crazy that we dont get a simillar effect to the double slit experiment when making these 10nm nodes!
3:33 SPACE RANGERS SOUNDTRACK!!! totally makes sense since this is a russian channel, but holy shit this was a wonderful surprise.
In my enormous (obvious not) knowledge, I think that the faster solution is chip stacking. It could aliviate the problem of the distance that the signals need to travel.
👍👍
There quite creative be plenty of work arounds and fudges for performance till a replacement found
i can't believe you used parts of the soundtrack from Space Rangers. going to sub just for that. excellent taste
Right, the distance a signal can travel inside the die is not relevant because the die is tiny and the layout of transistors on a die ensure that signals that are generated don't need to travel the length of the die. So, what matters is how fast you can switch a transistor on and off. What also matters is how much resistance this creates and how much resistance there is in the circuit in general at very high frequencies, which is what increases power consumption and heat. In general the higher the frequency the more perfect the construction needs to be. Imperfections create resistance.
But that's the hard part, etching a circuit ever smaller, but having to make it more perfect.
Это канал "Мой Компьютер" решил покорять иностранный сегмент? Музыка то я знаю откуда.
what about 4nm mobile chips?
are they really 4nm?
are these videos shot in Russian and later dubbed into English? The first video I watched I thought the audio was out of sync with the video, but it seems like it wasn't just a one-off. Nothing wrong with it, just wondering.
I'm thinking stopping at 3-4nm for awhile and bringing us HMB ram on the CPU die and removing the normal DDR sockets. Extremely high cache on the CPU with 512mb shared cache and HMB ram would be fantastic.
I don't know if in misunderstood something about a video he made before, but he said that it cannot go lower than 5nm.
@@elcebollon7750 In that case he got clowned, RTX 4000 series uses 4nm and 3nm is currently being produced by TSMC and bought mostly by Apple for the future M3.
@@MickenCZProfi it was my bad, i look it up and the silicone atom is 0.2 nanometer and china made a 0.3 nanometer transistor, also there is a chip with 0.5 nanometers transistors that can switch from insulator to conductor, so probably I heard wrong.
Nasa has the power of a 286 processor to sent a man to the moon. We need 3GHz to play candy crush.
👍🏽 🇬🇧 January 2024
A subject as interesting as it is complex - treated brilliantly & edifyingly.
2:15 give credits to the background video, i remember seeing a video about diy lithography thing with micrometer(i forgot video title) and it had this part
Sound sync issues. Otherwise nice work 👍
Ah shit, here we go again... It's not poor sync, it's a dub. Can't be perfect, but we did it as best we could.
@@vadulme8640 that explains it! Sorry for the noise 😬🙂
Would be awesome to see your most recent content as well
It's dubbed from russian
He's speaking in a different language btw
110% 👍👍👊👊
Can't wait for quantum computer
You left out the fact that each layer requires a separate mask and there may be up to 25 layers on each chip so that each wafer will have to go through the photo lithography machine 25 times
_Actually, it’s the closer the transistors are the faster it operates, this is a knowledge by AMD’ˢ R&D department, The problem is overheating and performance, this was AMD’s biggest Challenge to defeat Intel_
_Nevertheless, AMD was able to overcome the heating issue, the performance was still down, so what they did was incorporated, the north bridge processor into the CPU, giving birth to AMD’s APU, this allowed the CPU/APU to have its own direct communication to other processes on the motherboard; although they were not able to increase performance, they were able to increase efficiency_
_This reduce bottlenecking, and decreased motherboard resources, allowing for a smaller motherboard, using less power; with this kind of processor, there would be no need to increase speed of the CPU if you can reduce the bottlenecking through the motherboard; CPU speed is contingent upon other processors on the motherboard_
After reach the minimun size, scientists may find a way to double different status for one transistor
A pause in chip fabrication advances aimed only at processor speed may actually have some positives, it will incentivise more efficient code and delay mass-unemployment due to AI
i agree if intel fakes its process node
but for TSMC? do you really have a proof if TSMC process node is not smaller than 10nm???
I counted the leads of the Intel Core 2 Duo, and the Intel Pentium Dual core. The Core 2 Duo was rated at 2 Ghz, 2 mb cache. The Pentium was rated at 2.2 Ghz, 1 mb cache. The Core 2 Duo laptop cpu had about 5-10 more strings than the other one, but I couldn’t count them exactly.
Music from The Space Rangers II?? WOAH
"if it ionize, it ionize" - Captan Ivan Drago
4:25 did you really mean "increased by three times"?
I think he meant 3 thousand
The sound sync stressed me out, but good explanation
Well then shrink time n space.
Space rangers soundtrack?
Shame about the audio syncing, otherwise a perfect explanation of modern processors
This was a great video Thank You 🇺🇦
How do they connect 1 billion transistors with a few hundred pins. Seems like and explanation of how they landed on the moon.
You should post your video on Reddit
www.reddit.com/r/videos/comments/yrr503/physics_of_your_processor_problem_of_the/
R/technology maybe?
I wouldn't call technical processes of the last 10 years "pure marketing". It's still engineering. Not a huge breakthrough, but improving the tech in various smart ways, to overcome the limits.
Maybe they should create led cpu next time 😆🤣
Intel began gluing (according to AMD) since P4D and early version of C2D and C2Q, and knows that there's a huge bottlenecking.
How did IBM make the world's fastest chip
Well. What is up with the upbeat music? I thought we were fucking conquering Mother Russia ;))
XD
3 milions x 3 = 10 bilions.
USA magic calc.
Hmmmm .... ;
*I have been asking fancy-pants geeks these simple-damn questions for decades, the least-worst answers I got was endless damn circular tech jargon. Thks for anticipating my simple questions (aka crux system design factors) & clearly/succinctly/etc answering them w/o all the endless damn circular tech jargon. Pls don’t fall prey to the religion of Geekism & start talking in its religious geek tongues just like the rest of'm. I’m not anti-science, just a scientism heretic tortured for decades. Oh ?Any succinct-books explaining this stuff clearly like you do?
*Oh ?Hows-abouts doing FPGAs? Supposedly a FPGA turns a theoretical hardware processor into a software development project. Perfect for low-volume projects where an ASIC processor is far too expensive, takes too long, etc. Since you can update the FPGA just like you updated its software, an FPGA unit potentially has a longer life than an ASIC implementation. If unit volumes ever gets high-enough, an FPGA is a perfect tool to implement/develop the specs for an ASIC. I'm working myself to death trying to learn the secret/black art of FPGAs (ain’t much help out-there).
*Thks again
Is it 10 billions divided by 3 million equals 3 times??
Why do your hands perfectly match what you're saying but your mouth is speaking another language lol
Привет инглишмэны
The next big leap will be quantum computers
Unfortunately, practical quantum computers for mainstream use, ie. desktop quantum computers, are decades away - assuming that'll be possible at all. In the interim, I foresee a major change in how we implement chiplet and core design. AI will help design better and better masks, thus making processing in the current nodes much more efficient. Moore's law died with him. We have to concede to that, and attack the problem from other angles.
@@Harvey_Pekar I agree, we're decades away.. Controlling thermals would be good leap forward. Maybe one day we'll have nanotubes that act like vapor chamber pipes built right into the die. We're going to need something more substantial than what we have now for dissipating heat in order to maximize the potential computing power.
Hier Darcho Jandreoski Pegasus Galaktica 7 - Now intel Pentium 4 Chip for Home PC Computer dann habe ich ein Intel Pentium 76 Chip Prozessor 76 X Sterkere Chip als Intel Pentium 4 Chip
Why are your lips not in sync with the video?
Zelensky? That you bro?
I think your video is out of sync.
Why is your video quality so crappy
Disliked... disturbing music and audio error
My thoughts exactly, then I looked closely. It's dubbed from russian
Anyone know what the fuck he’s talking about 😂 Alien tech to me
Stupid musical background, errors and uncertainties