End of the silicon era. Processors of the future

[Music] we can admire the sea sitting on the sand endlessly this silicon wafer has lots in common with the sand the purest poly silicon of electronic quality is made of it pure silicon was first obtained by the Russian chemist Nikolai decadev in 1865. [Music] I bet he couldn't even assume that in 150 years All Electronics around the world would be centered around this semi-metal but the Silicon era is coming to an end each new marketing nanometer value is becoming more and more difficult to achieve the manufacturers resort to using various tricks like developing multi-chiplet Solutions and increasing TDP while violating their own heat dissipation standards but it is already abundantly clear that this is a dead end for silicon so what's next is it possible that the progress in technology will stop in a couple of years and we will keep adding more and more pluses to the Twisted names of process nodes of course not scientists from all over the world are looking for and already find in other approaches to the creation of computational electronics it is okay here and today we will talk about what life will be like after silicon foreign I guess I went too far with the epicness in the introduction new process nodes do still appear in 2026 tsmc promises us two nanometer chips and maybe a little earlier a weird Intel 28 process node will appear but I'm sure you know is just marketing we should not be expecting a qualitative reduction in a transistor size and the reason here is physics the minimum gate size of a silicon transistor is 5 nanometers 5 silicon nanometers are the limit with a smaller size it is simply impossible to make a transistor it will not work as a switch electrons will freely tunnel through its Channel ignoring the band Gap in other words such a transistor will always be on so the computation magic will not happen and the smaller the marketing nanometers are the closer we get to this physical limit and the more significant the tunnel in effect becomes which is bad for computing of course chip makers are trying their best to solve this problem thus you can reduce other parts of transistor or give the gate a clever shape for example shape it like a thin like in the finfet technology where transistors are essentially three-dimensional however all these tricks have led to the fact that the transistor density in chips has seriously increased and now there can be hundreds of billions of tiny switches in a piece of silicon the size of a fingernail actively releasing the heat when working backed close to each other and dissipating heat from this sandwich is a serious issue the owners of the last two generations of ryzen CPUs have already encountered this the small 7 nanometer chips even without overclocking easily reach the temperature of 90 degrees Celsius you've probably also heard about the latest Snapdragon NXT socs which thermal throttle under load due to overheating and with each new reduction in a transistor size the problem will only get worse [Music] looking at all these problems caused by silicon at the end of its life cycle the question arises why not just replace the scanical element with something else after all it is hardly unique in the periodic table which already contains more than 100 elements that's right it's not there is this metal called germanium from which it is also possible to make semiconductors what's more the first transistors in the late 40s were made of this metal it has three times higher electrical conductivity less voltage and hence heat loss at the PN transition and less resistance of the open channel in general germanium seems to be better than silicon in semiconductors but by the 60s it had been almost completely abandoned there were three reasons for that first this metal is much more expensive and less common than silicon the latter is almost 30 percent of the Earth's crust silicon is second to oxygen there second germanium has much less thermal stability that is when heated it loses its characteristics faster besides it has problems with oxidation and third it has worse thermal conductivity that is it is more difficult to remove heat from it than from a silicon chip all of this has led to the fact that the frequencies of the best germanium chips made couldn't get higher than hundreds of kilohertz for further increase it was necessary to switch to Silicon the era of pure germanium and its 60 years ago but we live in a time of advanced Alloys and chemical compounds remember the T-1000 from Terminator 2 is it really impossible to modify germanium in such a way that it becomes better than silicon for semiconductors it turns out it is possible and such a compound is called germanane in fact it's like graphene only from germanium a thin monatomic film its production is a form of art first a layered pie is made of graphene and calcium after which the latter is washed out with water which in the process gives off its hydrogen making germanium bonds stronger and allowing the separation of single layer films of this metal as it turned out such films conduct current 10 times better than silicon and the cooling issues are not of big concern here but of course it is still quite far from commercial production they have learned to create germanaid only in Laboratories and so far there is not a single ready-made chip based on it however German 9 is not everything there is another Compound on the basis of which they actually made a semiconductor chip it is called molybdenum disulfide AKA molybdenite now it is mainly used to make various Alloys but it has excellent semiconductor properties that persist at such tiny sizes where silicon is oxidized to glass thus scientists managed to bring the thickness of molybdenite to 0.65 nanometers with full preservation of semiconductor properties and most importantly it was possible to create a semiconductor photodiode based on it which is five times more sensitive than silicon this will allow to create matrices for cameras with even higher photo sensitivity in the future but is there a perfect chemical element that can replace silicon yes it's carbon it's ironic the basis of our life can become the basis of our future Terminators although it must be admitted that in an extremely unusual form this unusual compound is called carbon nanotubes it represents sheets of graphene that are rolled up and yes there are also semiconductors and of atomic thickness besides their electrical conductivity is three times higher than that of silicon it is noteworthy that on the basis of such carbon nanotubes it has already been possible to make the first chip with 14 000 transistors although its process node is not striking only about a micrometer that is the level of silicon processes of the 80s still it is a full-fledged chip on which it has already been possible to run a hello world program in the future scientists are planning to reduce the size of nanotubes and thereby create faster and more efficient chips but this is still a relatively distant future all of this arouses the question is there already such a replacement for silicon that anyone can get their hands on and not just a couple of scientists and large Laboratories yes there is charging devices with the gallium nitride organ this semiconductor became popular in the 90s the first White LEDs and some types of blue lasers were made on its bases its peculiarity is that it's possible to produce Electronics based on it at the same factories where silicon semiconductors are made but at the same time gallium nitride has a wider band gap which allows it to work at higher voltages or less heating than silicon and this property is very necessary for compact charges where it's actively used well the first approach has been sorted out you can replace silicon with another element which is still far from reaching its physical limitations but there is also a second way to abandon the CMOS transistors that we're used to and switch to something else CMOS stands for complementary metal oxide semiconductor structure [Music] and the vast majority of modern micro circuits are based on this principle which was invented back in the 60s hence a logical question what if we go a completely different way and not give up silicon but change the very principle of the Chip's operation this approach is similar to the change of processor architectures for example Apple has shown that its M1 arm chips can be both noticeably more powerful and run noticeably cooler than many modern x86 chaps from AMD and Intel perhaps the rejection of CMOS transistors will do the same well such ideas do exist indeed not only on paper there exists the so-called tunnel Field Effect transistors they work completely differently compared to Field Effect transistors if for the latter electron tunneling is a failure a transistor cannot get closed and turns into a conductor the tfet's whole operation principle is based on this effect the thing is the tunneling looks like cheating physics it makes sense that if an electron does not have enough energy to overcome a potential barrier then it will remain behind it unless it's given the missing energy however the tunnel in effect allows electrons even with insufficient energy to leak through the barrier in other words this allows tfts to operate at lower voltages than conventional CMOS on top of that reducing the size of the gate does not hinder anything here in order to avoid excessive tunneling you just need to lower the voltage even more which also leads to less heat generation so is it a victory at last not everything is so simple first graphene that's needed for the production of tunnel effect transistors is the only element that has the necessary properties second such transistors require ultra low temperatures for operation and what a cooling is not enough here liquid nitrogen is a prerequisite thus scientists still need to do a lot of research before we can see first devices based on tfetes and since we are moving away from the classic CMOS transistors let us also mention memristers memristers were developed on paper back in the 70s and their name comes from the words memory and resistor and this perfectly describes their main feature a resistor is basically an electrical resistance that does not change in any way whereas a mem rester has a memory effect in other words it changes its conductivity according to the amount of electric charge flowing through it this property allows A Perfect Memory store to be both volatile and non-volatile memory at once that is combined RAM and SSD and it can turn our ideas of our data storage upside down booting the system will no longer be necessary after all all the information from the RAM and the drive will be stored in one place a power outage will no longer lead to loss of information the memory store being non-volatile will retain the last state downloading any data will speed up significantly after all you no longer need to move information between RAM and the drive sounds like fiction huh yeah but this is already a reality the Israeli company we bet Nano announced a successful completion of testing of a risk-5 modular architecture SOC with 128 kilobytes of built-in resistive memory re-ram such memory is less susceptible to temperature fluctuations radiation and other negative factors which makes it attractive for the industrial and military application and if we look into the more distant future memory stars are ideally suited for the role of artificial synapses to create neural networks as similar as possible to the human brain and it can be designed on standard microchip equipment the thing is a memory store behaves in a very similar way to a synapse the larger is the signal that passes through it the bezel it will pass the signal in the future this property is ideal for teaching Terminators to adapt to us as realistically as possible so we have considered two approaches replacing silicon with other elements and changing the principle of operation of transistors what else can be changed for instance the charge and data carriers themselves electrons in some cases can be perfectly well replaced by photons many have heard or even are using now the pawn or g-pawn technology an optical Network that is already actively replacing copper Twisted pairs in large cities allowing millions of users to get fast internet access a similar technology can also be used in computers it is more efficient to transmit information using light since it allows you to get a wider bandwidth immunity to electrical interference and minimal heating theoretically Optical connections will help reduce latency and increase the speed of interaction between the processor's compute units and the cache or between the processor itself and RAM we can already see the limitations of copper conductors in practice only the first pcie Slots of the new Intel processors support the new 5.0 protocol and the gdr6x chips have to be placed as close to the GPU as possible photonics will remove all these limitations which will allow computers to develop further although it must be said that this does not solve the problem of the physical limitations of silicon the only thing that's left is to tell you about the most popular and the latest possible approach which no longer changes physics itself but the entire logic of computer operation the vast majority of various Electronics is centered around ones and zeros the presence and absence of electric charge it is this basic principle that allows transistors to calculate and the memory cells to store information but the simplest is not always the best this approach imposes serious limitations on many tasks these limitations are easily bypassed by quantum computers the spoon does not exist for them there are no clear ones and zeros they operate with qubits or Quantum bits that have a state of 0 and 1 simultaneously but how is this possible here's a simple example how many numbers are there made up of two zeros or ones obviously there are four these are zero zero zero one one zero and one one you need two bits to store each one which gives us a total of 8 Bits And you need only two qubits for that that is four times less the most important thing is that the more complex task we solve the greater the profit from using Quantum bits imagine that we need to brute force a password an ordinary computer will sort through all possible combinations but a quantum computer does not need that it is enough for it to take as many qubits as necessary to store the largest possible password and the task will be solved such a number of keybets Will already contain all possible passwords including the one that we are looking for yes this approach is mind-blowing in fact everything is even more complicated than that for you can't just simply obtain the results of quantum calculations since the system is in an arbitrary State at every moment in time in an attempt to read the data will turn it into a classical one but we're here now not to talk about the operation principles of quantum computers this is a whole different topic in itself we are talking about whether they are the future of complex Computing and the answer is yes the simplest quantum computers can already be bought and the price can be quite affordable and even lower than that of the top tier Apple Mac Pro a year ago the Chinese startup Shenzhen spring technology demonstrated a supercomputer with two qubits at a price of only five thousand dollars below price is due to both a small number of qubits and a fairly simple principle of operation based on nuclear magnetic resonance which has already been well studied also they are using ordinary permanent magnets instead of superconducting ones although still quite powerful by the way the whole thing is controlled from a regular PC to which the spank can be connected and this of course is just the beginning just a few months after the Chinese the Dutch startup quantware presented its five key bit computer true its price was not announced but the idea here is quite clear quantum computers are becoming more and more popular and appear in the masses so it is quite possible that they are the future of personal electronic devices after all thus no need to worry about it now yes we are getting close to the Silicon limitations but it's not the only thing we've got there are enough various elements that can replace it we can even change the operation principle of transistors or even the very logic of computers thus technological progress cannot be stopped Electronics will continue to develop but in what direction only future will tell that is if it continues to exist for us in this line of events if you want to cross paths more often click on the Subscribe button give us a like if you enjoyed the video dislike if not nobody's gonna see it anyway my name is Mikhail Crusher the communication session is over bye [Music] to see [Music]

Channel: My Computer

Views: 317,728

Keywords: silicon, my computer, quantum computers, processors of the future, mikhail kroshin, intel, amd, semiconductor, microprocessor, silicon valley

Id: D--sSNKiVXg

Length: 19min 26sec (1166 seconds)

Published: Tue Nov 01 2022