PHYSICS of your PROCESSOR. Problem of the nanometer limit

we are used to the idea that in the computer World more means better the more coarse or higher the frequencies the faster it works the more memory it has the more information it can store however soaking chips have one parameter that works exactly the opposite the smoother it is the better it is usually referred to as process node but how does it affect the production of modern chips why does smaller mean better in this case this is MK today we will talk about the physics of your processor foreign [Music] to begin with let's try to understand what process node is to answer this question we need to talk about photolithography photolithography is the process of obtaining a certain pattern on the surface of a material in order for the pattern to have a certain shape the light passes through a special filter mask The Perfect Analogy here is x-ray bones block this radiation better which is why they are clearly visible in the final image unlike flesh through which x-ray radiation passes more freely the back to photolithography in fact this process is somewhat similar to the old principle of taking a photo only with post-processing the lights passing through the filter mask falls on the purest silicon wafer pre-coated with a photosensitive film the so-called filter resist the areas that got light are exposed and then removed in the developer and thus the key pattern remains on the wafer this pattern is in fact a photo of all transistors of the 2D processor and then comes etching Electro deposition and vacuum spraying so that the pattern turns into full-fledged transistors that conduct current when they are asked to it seems like nothing too complicated but then why is every new process node is so difficult to achieve what is the limiter the answer may surprise you the wavelength of the radiation it's simple it's the light passing through the mask that leaves the real outlines of transistors on the filter resist and if we use visible light the minimum wavelength of which is about 380 nanometers in the Violet Spectrum you will not be able to use it to create a transistor the smallest part of which is significantly smaller than this wavelength even if you use lenses so it's simple the smallest part of the transistor is the width of its gate and as you can see it is directly related to the wavelength of the light used therefore it was decided to outline such an important characteristic which directly speaks about how fine the chip is and call it process node however this continued somewhere until the mid-2000s that is until marketing Statin and gave way to various tricks which led to the fact that the fake 10 nanometer Intel process node is now called even more fake Intel 7. other companies are doing the same but why are manufacturers so eager to reduce the size of transistors why not just make micrometer chaps like in the 80s and hear her majesty physics bursts in first the smaller the size of the transistor all other things being equal the less heat it generates since the current physically has to travel a shorter path and there is no such thing as 100 efficiency on the other hand the most effective way to increase performance is to increase the number of transistors that are responsible for computation magic but at the same time there is a third Factor heat dissipation 100 200 300 watts for the removal of which a water cooling system will be required which ordinary users can hardly afford thus in order to increase performance you need to increase the number of transistors and in order to keep tdps within limits you need to constantly reduce the transistor size at the same time increasing the number of transistors is a pretty simple task in itself which is why we went from 3 million switches in Pentium from 1995 to more than 10 billion in the top tier ryzen 5000. that is over 25 years the number of transistors has increased by three times while the process node even from a marketing standpoint has shrunk only a few dozen times from 350 to 7 nanometers and in reality even less and as a result if the Pentium consumed 10 to 15 watts at most the top end ryzen reaches 150 and even 200 Watts which is close to the cooling limits in desktops and this perfectly demonstrates why all manufacturers are going out of their way to continue to shrink process node and reduce power consumption but this is not the only problem that this stupid science throws at us modern processors have already reached such high frequencies that this speed of light does not look like something infinite anymore for example take a chip with a frequency of 3 gigahertz that is it performs 3 billion clock cycles per second taking into account the speed of light which is roughly 300 000 kilometers per second we get that in one clock cycle the electromagnetic wave will have enough time to pass the length of only 10 centimeters it would seem that processor dies are smaller than that but remember that there are winding labyrinths of billions of transistors inside them which is why the paths inside the chip can be long enough so that we need to take into account the speed of light and clock lag in different parts of the die that is why we do not see huge silicon chips appear it is extremely difficult to calculate all possible delays and clock cycles and of course this will affect the final performance in a negative way the second limitation of the size is defects the larger and more complex the die the more expensive it is and the higher the chance of getting defected dies this is another trap as effect process nodes plus the size limitations of the die mean that fewer transistors will fit into this die and it will be less performant the way out here is obvious reducing transistor size will solve all the problems but there exists a workaround of another kind and it is called a chiplet the idea is as simple as possible if it is not possible to create a huge chip then why not create several small ones and combine them with a common bus with a known identical delay thus two birds can be killed with one stone Small chips are easier to produce they are cheaper and a defect ratio is relatively low in addition the issue of delays inside the die itself is removed this is the path that AMD is following for example its resin 3000 and 5000 have three dies inside two with cores and one with controllers United by a common bus called Infinity fabric but does it work those who use them will clearly answer yes and the increased intercourt delay due to a pair of dies is successfully solved by a large size L3 cache but I will repeat myself chiplets are still a workaround in case with ryzen tiny pieces of silicon the size of a fingernail are easily heated to 100 degrees and as a result we again come to the same conclusion we need to shrink the transistor size and many of you are probably asking what is the problem with shrinking it even further okay then it depends on the wavelength of the light used in photolithography and visible radiation is not suitable but what's the problem with using ultraviolet light that extends up to 10 nanometers and that is exactly what's being used for example UV radiation with a wavelength of 248 nanometers allows using a lens system to create transistors with a minimum size of about 100 nanometers but if we look at modern chips their linear Dimensions could reach several tens of nanometers how do they manage to achieve this that's right further reducing the wavelength and going into extreme ultraviolet and what's the problem this time you may ask the problem is in the equipment itself when we are talking about tens of nanometers we have to work literally with separate atoms because the crystal lattice spacing of silicon is only 0.5 nanometers therefore extreme ultraviolet lithography is using mirrors made by a world famous company Zeiss these mirrors are polished so much that even individual molecules are knocked off their surface for the sake of making it perfect the filter masks also became drastically more complicated a single layer mask made of chromium is often enough for ultraviolet lithography whereas for euv templates with 40 layers of silicon and molybdenum are used the physical effects that happen there are almost at the quantum level when even the same wavelength due to diffraction yields chips with varying characteristics on the same wafer for those who are interested Google app brag condition and yes in the case with extreme ultraviolet The Masks lenses and mirrors absorb 95 percent of the radiation that they catch that means that in order for the magic of photolithography to happen the lasers used must be another more powerful than with conventional ultraviolet for a better understanding one modern asml extreme ultraviolet lithography machine can consume under one and a half megawatts only 10 kilowatts of which will reach the Silicon Wafers themselves yes the efficiency here is about one percent as Steam Locomotive is a lot more efficient than the machine that makes processors by the way such a machine allows you to process about two Wafers per minute in short such photolithography is at the age of modern science and it is truly an engineering Wonder and all of this is just in order to achieve the 10 nanometer process named in Silicon semiconductors and allow to further increase the number of transistors without getting sky-high tdps and this partly explains why processors are getting more and more expensive each year for example in the case of Intel they are 10 nanometer chaps such as the 12th generation core CPU cost the company twice as much as the 28 nanometer core CPU of the fourth generation and every new process node although it would be in marketing requires huge costs for the most complex machines and scientists who dig into the depths of the universe but what's next sooner or later we will reach the limits of extreme ultraviolet now the best asml machines have a resolution of 13 nanometers which is only 3 nanometers more than the theoretical boundary that separates ultraviolet and x-ray thus sooner or later we will have to dive even deeper into x-ray lithography which will allow to create conductive structures the size of several nanometers or even tenths of an nanometer sounds cool the only problem is that in comparison with x-ray lithography extreme ultraviolet lights will not look that extreme anymore firstly the equipment will become even more expensive in order to effectively block x-rays masks made of gold 10th Lumber tungsten will be required at the same time the substrate for the mask transparent for x-rays will have to be made perhaps of diamond and this is just the tip of the iceberg in the case of X-ray we are dealing with such high energy radiation that is carriers when they get into the material are able to generate powerful secondary radiation with a mileage of tens of nanometers which can seriously change the key pattern on the Silicon wafer and the effect of such a phenomenon as ionization which generates free electrons is also becoming more noticeable here in general there's a ton of problems with x-ray lithography and scientists from all over the world are already trying to solve them one thing is already clear this will be the end for silicon x-ray lithography will allow to reach the sizes of the crystal lattice of solokin which will make it impossible to shrink the process node further however the end of silicon is not the end for chips as we discussed in our previous video creating modern ships is a complex art that requires the most precise equipment and the latest knowledge therefore we should not be surprised when we see slow downs and hiccups in the development of new process nodes Humanity has already entered the Realms where physics does not work the way we think it should but in any case there is still time before the end of silicon which means that companies will continue to please us with new processors and graphics cards more important is that we have enough time as a developed civilization I hope you enjoyed the video while we were preparing this material we sifted through a lot of information and got a lot of interest in knowledge I thank you too this was MK I'll see you again goodbye and yes yes I forgot I forgot to say if you liked it leave a like subscribe to us if you haven't subscribed yet write comments we will read it all

Channel: My Computer

Views: 63,202

Keywords: silicon, my computer, mikhail kroshin, semiconductor, microprocessor, Photolithography, Extreme ultraviolet, nanometer

Id: a_cb3mx7V-Q

Length: 13min 6sec (786 seconds)

Published: Thu Nov 10 2022