On a web resource habr.com interesting material of the General director of LLC "MAPPER" (semiconductor factory for the contract production of MEMS devices) has been publishedDenis Shamiryan (@CorneliusAgrippa) "Microelectronics in Russia before and after 02/24/2022" about the current state of microelectronics production in Russia and about the possibilities in the conditions of large-scale Western sanctions to create a fully localized production of microelectronic components.
Installation of lithography at JSC "Micron" in Zelenograd (Moscow) (c) "Made with us"
In light of recent events (for posterity: Google Russia, Ukraine, February 24, 2022), which led to the imposition of sanctions against Russia in the field of high technology and, in particular, microelectronics, I often hear the question: what next? What is the current state of Russian microelectronic production? Will Russia be able to create a completely local chip production?
I will make a reservation right away that this article does not pretend to be a comprehensive independent analysis of the situation, but reflects my personal point of view, based largely not on open sources, but on experience: more than 20 years in the industry, 15 years abroad, both in R&D (IMEC) and mass production (Global Foundries) plus 8 years in Russia (starting from scratch a plant for the production of MEMS), personal communication, the opinion of other specialists; in general, everything that cannot be found or is very difficult to prove. Therefore, I will not provide proof - everyone has their own point of view and the right to express it (at least for now).
I will only talk about production technologies, since I myself am a former technologist, I have never had anything to do with design, and phrases like "licensing of processor cores" are dark and incomprehensible to me.
I will also note that I will only talk about CMOS production, firstly because this topic is most interesting to consumers (this is consumer electronics - processors, memory, etc.), secondly, I worked abroad in CMOS (aka CMOS) production and I have a good idea of it from the inside, thirdly, I myself now I work in the MEMS industry and I will not write about it, since I am an interested party.
The article consists of three parts:
Analysis of current manufacturers
Reflections on the topic of fully local production of microelectronics
An attempt to look into the future
Analysis of the current situation
To begin with, let's look at the current manufacturers of microelectronics. I will only talk about more or less modern factories capable of producing chips according to the 180 nm process technology and below. To make it clear, I will give examples of processors produced using a certain technology, the data is taken from Wikipedia there is a column with all the technical processes on the right in the article, you can click and see what was produced for this technical process (and when). So, 180 nm is the beginning of the 2000s, Intel Celeron and PlayStation 2 processors. We will not consider any old Soviet factories (such as NCPP) that work with technologies larger than a micron (for example, Intel 80286 was made using 1.5 microns technology).
A small caveat about the size of the plates. Modern production works either on 200 mm (up to 90 nm) or 300 mm (65 nm and below) silicon wafers. The most advanced equipment for technologies less than 65 nm exists only in the 300 mm variant. Therefore, it will not be possible to make high technologies on 200 mm plates. And equipment for 300 mm plates is significantly (at times) more expensive than equipment for 200 mm plates.
So, what we have at the moment.
Micron
Micron is the most vibrant microelectronic production in Russia. They work on 200 mm plates, have 180 nm technology (in mass production), 90 nm (I'm not sure if it's very mass, but I can be wrong; 90 nm is Intel Celeron M/D, AMD Athlon 64), 65 nm (here I have big doubts that there is mass production; 65 nm is AMD Turion 64 X2, Microsoft Xbox 360 "Falcon"). I once participated in attempts to develop 65 nm technology on 200 mm plates (IMEC, Belgium), but the equipment did not pull, so the 65 nm process technology was transferred to 300 mm equipment.
Produces Micron in large volumes, mainly chips for bank cards, passports, subway tickets, etc. In small volumes, they produce what they were sanctioned for. They have been under sanctions for quite a long time, so they have somehow learned to cope with it. The turnover is more than 6 billion rubles, of which they earn about half by themselves, the rest is paid by the state (for example, in the form of subsidies under the 109 resolution of the Ministry of Industry and Trade - Micron is always among the recipients of subsidies there).
Angstrom-T
Do not confuse it with just Angstrom (without T) - Angstrom is just an old Soviet production, they made chips for Soviet calculators and the game "Well, wait a minute" - if anyone is old enough to remember it, there was a wolf catching eggs from under chickens. Angstrom is still alive and produces products (of course, not for calculators).
The history of Angstrom-T began in 2007, when Global Foundries (then it was still the AMD - Fab36 plant, Dresden), began the transition to 300 mm plates and sold all equipment and technologies for 200 mm Angstrom-T: 130 nm (AMD Athlon MP Thoroughbred level) complete documentation for the process with a guarantee of output suitable and 90 nm - developed, but not yet in mass production. At that time, these were fairly recent technologies. But then something went wrong. The equipment got stuck in a warehouse in Rotterdam, and when I came to work for Global Foundries in 2011, it was already a byword - how they sold the equipment to a Russian company, but instead of being used, it has been rotting in the warehouse for 4 years. It was rotting somewhere else until 2014, after which it still came to Russia. A plant was built in Zelenograd, almost an exact copy of Dresden, they even built their own power plant to buy gas instead of electricity and generate electricity on their own, so as not to depend on power outages. The same was done in Dresden, however, the Germans managed to turn off the electricity at the factory themselves (just in my watch) - but that's another story.
So, the plant was built, the equipment was delivered, I was there and experienced deja vu after Dresden - everything is exactly the same, installations in the same places, with the same code names.
That is, everything looked more or less normal there, but there is some strange story with the leadership there. I have a whole collection of business cards of the general directors of Angstrom -T of the same design, only the surnames are different - they were constantly changing there (along with the whole team). Once I talked to one of the deputies, he asked me how we have quality control, I told him, he started laughing and said that I don't understand anything about quality control. Well, our quality control is arranged according to the same principles that I used in Germany to make modem chips for Qualcomm for fifth iPhones, Apple didn't seem to complain about the quality. So I shrugged, but didn't argue. Another time I talked to VP sales ASML, he was interested in how Angstrom-T was doing and said that since their scanners had been in stock for 7 years, it would be very difficult to launch them and offered to hand over the old scanners to trade-in, and put newer ones in Angstrom-T with additional payment. I recounted this conversation to the management of Angstrom-T and said that in my opinion this is a good option - they will get a quick result of the best quality, even for extra money. The management of Angstrom-T said that it does not know anything about this proposal. Strange, I thought, some left-wing dude like me knows, but those to whom it was offered and for whom it should be important - no.
The result is that 15 years have passed since the purchase of the line, production still does not work. Whether it will ever work, I do not know. At the moment, the company has gone bankrupt.
Crocus Nanoelectronics
The original idea of Crocus is the production of MRAM - magnetoresistive memory. I will not go into details, in short - you get non-volatile memory (as on flash drives) that works at the speed of RAM (as DRAM). This combination makes many people salivate, so many have tried to make it (I know for sure about Sony and Infineon). The problem turned out to be that theoretically everything is beautiful, but in reality it didn't work out very well, more precisely, it turned out, but the performance turned out to be at the level of ordinary flash memory, and flash memory is already there, why bother with another technology for something that already works fine?
But, before it became clear, Rusnano decided to invest in a 300 mm factory for a 65 nm process technology in Russia. You can make fun of Rusnano as much as you like, but at the moment it is the only factory in Russia with 300 mm plates with 65 nm technology. However, there is a nuance.
In the original model, it was assumed that MRAM cells would be manufactured at metallization levels (the so-called back end). Since the transistors themselves (front end) can be manufactured at any factory, this is an easily accessible product, it was decided not to spend money on a full-cycle factory, but to build a part of the factory that will contain only know-how to manufacture MRAM. I note, by the way, that the equipment for the front end is much more expensive (there is simply more of it there, and there is not much for the back end, in principle, it is not necessary). So the original model looked like this:
We are building a semi-factory (back-end only) for reasonable money
We buy c front end plates for little money on the world market
Adding MRAM back-end
We sell for a lot of money on the world market
PROFIT!
If MRAM technology worked, it would be a very beautiful solution. But it didn't work (and not only at Crocus), and Crocus turned into a kind of suitcase without a handle.
On the one hand, it is not a full-fledged factory, since it does not make transistors (front end), and it is pointless to order a front end at a foreign factory and then finish it at home, it is easier to immediately order a full cycle at a foreign factory. If you are denied a full cycle, then you will be denied half of the cycle.
On the other hand, it is the only production in Russia operating on 300 mm plates according to 65 nm process technology, with the possibility of further modernization up to 45 nm and maybe up to 32 nm.
That is, it's a pity to kill, and what to do next is unclear. Build it to a full factory? But this is a huge investment, and there is not much space there physically for a full factory. That is, it is necessary to transfer. And if you transfer - isn't it easier to build from scratch then? (usually easier). And to bankrupt - the hand does not rise.
The annual turnover of Crocus is about a billion rubles, they themselves earned ten percent (mostly one-time orders for spraying magnetic materials for foreign customers - there are no Russian ones, since there are no 300 mm factories in Russia).
As a result, after a long ordeal, Rusnano sold Crocus to one large state corporation. They will make quantum computers there. Don't ask me what it means.
New plant in Zelenograd
Little is known about him. The size of the plates is 300 mm, the process technology is 65 nm - 45 nm (First generation Intel Core i3, i5 and i7). It has been planned to build it for a long time, for example, the news (of unknown dating) that it should be built by 2014. The Sitronics company was going to build, but it is not possible to Google anything intelligible. A few years ago, the government sent me a technical task for the plant for examination, I read it - it was written correctly, clearly written by people who knew what they were doing. According to rumors, construction is underway, with the involvement of Chinese contractors (like UMC - true, this is Taiwan). I can't say anything more. What will come of this is also not clear.
Result
On a Micron, it is theoretically possible to produce something of the Intel Celeron/AMD Athlon 64 level (90 nm process technology, mid-2000s). To move on, you need a 300 mm plant, but it is not in a fully functional state.
Is it possible to fully localize the production of microelectronics according to the modern technological process?
Short answer: no.
A more detailed answer: No country in the world will be able to localize the production of microelectronics by a process less than 90 nm. It is still possible to set up something like micron technology (contact lithography, liquid etching, manual operations) on the knee somehow, but it will be the 8086/80286 or ZX Spectrum level.
Detailed answer. The following factors are necessary for successful microelectronic production:
Availability of a sales market
Availability of production equipment
Availability of competent personnel
Availability of raw materials, supplies and consumables
Let's look at each aspect in more detail.
Sales markets
It would seem, what sales markets - if it is necessary to do, then it is necessary, regardless of costs. The problem is that the semiconductor plant itself is just the tip of the iceberg. And regardless of the costs, you will have to saw the whole iceberg, and this is a lot of money.
Everyone is used to the fact that semiconductor chips are very cheap. Why they turn out to be cheap, I wrote in another article. Many mistakenly believe that it is enough to put a factory on the territory of Russia and we will get the same cheap chips, only produced at home. Unfortunately, this is not the case. A semiconductor factory consumes a huge amount of money, regardless of whether it produces something or not. That is, in order for one chip to be cheap, you need to divide this huge amount of money into a huge number of chips (tens of millions for a medium-sized factory). And they need to be sold somewhere. If there is nowhere to sell them (the Russian market is not that big), then the plant will incur losses that either the state must cover with subsidies (then the chips will be cheap for the consumer), or the consumers themselves (then the chips will be very expensive). That is, if you want to make really cheap chips, you need to sell them to the whole world.
The next layer of the iceberg is the equipment. The plant needs about a dozen installations of the same type (for example, lithography, or etching), and there are dozens (if not hundreds) of such types. A manufacturer of one type of equipment is not interested in a market of ten pieces - again, either the equipment will be gold for the plant, or the equipment manufacturer should be subsidized by the state. Or there should be a lot of factories, then the equipment manufacturer has a sales market and its products become cheaper. But we don't need many factories - one of us doesn't know where to put the chips. That is, if you want to make relatively inexpensive equipment (relatively inexpensive - this means that, for example, installing photolithography costs about as much as Boeing), it needs to be sold worldwide.
The next layer of the iceberg is components for equipment - electronics, pumps, robots, etc. It's the same story here - for tens/hundreds of pieces of equipment, many pumps are not needed, and we are again stuck either in high cost or in the need to sell on the world market.
And the same story will be with everything else: with silicon wafers, chemicals, a water treatment system. Everything that will be unique for our production will be wildly expensive, since we will not sell it to anyone else (well, or we trade with the whole world).
One more thing. One plant cannot produce the entire microelectronic nomenclature. That is, processors, RAM, flash memory, microcontrollers, radio modems, etc., etc. cannot be squeezed into one factory. The production of RAM is generally a separate branch of microelectronics with separate factories, technical processes and players. At one time, the Germans tried to play this game, Infineon spun off the company Qimonda, which was supposed to be engaged in the production of RAM. It didn't work out. The cost of the memory chip produced at Qimonda was equal to the cost of the Samsung memory chip on the counter in the store. Qimonda went bankrupt.
That is, to have a fully localized production, you need to have several factories. And to sell the products of these factories somewhere. Or keep these factories running with minimal load. In fairness, I note that many factories will create at least some demand for equipment and raw materials.
Let's roughly estimate how much it costs. For example, Intel is building a new plant in Germany for $17 billion. We need several factories, let's say it will be $ 50-60 billion. For comparison, this is defense spending in Russia in 2020. The entire ecosystem, I think, will cost at least an order of magnitude more, that is, $ 500-600 billion. This is already a third of Russia's GDP. But such an ecosystem can cost more than an order of magnitude.
As a result, it is VERY expensive to create and maintain a fully localized production.
Production equipment
Let's say we found quadrilliards of money somewhere and can afford everything. The first thing you need is equipment. I note that at the moment there is not a single country in the world that would produce all the equipment necessary for microelectronic production using technologies of 45 nm and below. Even the USA, which produces the lion's share of semiconductor equipment, does not produce photolithography machines. They are produced either by the Netherlands (ASML) or Japan (Nikon, Canon). Applied Materials (USA), one of the largest (and maybe the largest) equipment manufacturer, usually boasts that it can supply a full line of equipment only from its machines, but always adds: except photolithography.
It is very difficult to make equipment for modern semiconductor production, and it is impossible to start from scratch. There are two points here.
Firstly, modern equipment manufacturers have come a long way in decades improving and improving their equipment. For example, the Dutch manufacturer of photolithographic equipment, ASML company spent about 15 years to bring the EUV installation to mind. The first prototype was delivered to IMEC (where I worked then) at the beginning of the two thousandth, and it entered the market a few years ago (I still don't know how long it took them to make the first prototype). This is despite the fact that ASML has vast experience in the development and production of photolithography machines and their R&D budget is about a billion euros per year (I think the lion's share of this budget has been going to EUV).
Secondly, modern equipment is actually a lego set, in which 90% of the standard blocks (robots, vacuum pumps, gas flow controllers, etc., etc.) and 10% is the know-how of the company, which spends most of the time and money during development. As far as I know, components of semiconductor equipment of the required quality are not produced in Russia.
You can, of course, try to do everything yourself - but this is just one of the reasons why our parent company Mapper Lithography failed: they tried to do everything themselves: power supplies, RF generators, write their own software, etc. As a result, the car worked for an hour, then broke down and was repaired for a week.
You also need to remember that in addition to production equipment, auxiliary equipment is needed: water treatment systems (and it's not in the kitchen to put a filter), compressed air compressors, nitrogen generators, etc., etc. All this also needs to be taken somewhere, now this equipment is all imported.
Conclusion: you can try to do something if you have access to standard high-quality components, if you also make components yourself, then in my opinion, it is impossible. Plus, what I wrote in the section about sales markets, even if you make equipment, then to whom to sell, to one factory? But, although you can try to sell to China - there are a lot of factories there.
Competent staff
This seems to be the least of the problems, but there is a nuance. In principle, Russian universities produce a sufficient number of specialists who, after several years of training, are quite capable of working in modern production. This is confirmed by the experience of our company and the fact that many specialists of Russian origin work in foreign semiconductor industries (I worked there myself, and I know many Russians who work).
Now about the nuances: firstly, specialists need to be trained, homegrown specialists turn out badly, especially in the field of production culture and quality management. In my experience, quality is a headache for Russian companies. Everyone can rivet analogs in a single copy, but few people can deliver products of sustainable quality. In the presence of management / leading engineers with foreign experience, it is not difficult to put quality management, but purely Russian companies do not do it well. Do you remember how I wrote above that the production management of Angstrom-T laughed at our quality management system? That's just about it. In general, foreign (either expats or Russians with foreign experience, like me) specialists can come and teach, the question is how to lure them now?
The second nuance: as soon as process engineers become more or less experienced specialists (several years of experience in normal production), they immediately begin to look abroad. A process engineer at a semiconductor manufacturing facility in Europe receives 3-4 thousand euros on hand (to understand the level of expenses, I will give an example of Dresden: I will eat a 3-room apartment 700-800 euros, food 200-250 euros per person, clothes are one and a half times cheaper than in Moscow). As a result, there is a constant leakage of personnel, since technology engineers are always needed abroad (although not as acutely as IT specialists), and unfortunately we cannot afford to pay as abroad.
As a result, for our hypothetical plant, we should invite foreign specialists with their management technologies, and then keep our specialists from emigrating.
Raw materials and supplies
To operate the plant, we will need silicon wafers, liquid chemistry (especially photoresist), gases, all sorts of small things (such as gloves, masks, tweezers, etc.). And all this is not of any quality, but a very high degree of purification, small things compatible with clean rooms, etc. With all this, the situation in Russia is not that very rosy. An interesting example with masks. When covid started, our supplier of masks (special for clean rooms, ordinary medical ones are not suitable there) said that they had thrown all the power on medical masks and there would be no special ones now. I had to invent reusable ones and wash them. In Russia, such masks are not produced.
We tried to work with a domestic photoresist. Then bubbles, then garbage, then it doesn't stick to the plate. Each batch is different from the previous one, we had to adjust the process parameters for a new batch each time. It fell into disrepair two months before the expiration date (sometimes, and sometimes even after the expiration date was normal). In general, we played roulette for about a year, switched to American. We set up the process once and forgot about the problems. And it was a micron-sized photoresist. I do not know how things are with a Russian photoresist with a technology of less than 65 nm.
Silicon wafers. There is a wonderful Russian company that produces them. The nomenclature is not very large, but there are the most popular sizes. The quality is good. But, as usual, there is a nuance. The plates are cut from imported silicon ingots, on imported equipment using imported consumables (the stock of which, as we were informed for two months, there are no new supplies yet). That is, if we want a fully localized production, we also need to establish the production of ingots (for this we also need to develop and produce equipment), the production of machines for cutting, grinding and polishing and consumables for them.
Photo templates. In Russia, there is a production of photomasks using older technologies (definitely not at 45 nm and below), and, of course, on imported glasses and imported equipment. The production of modern photomasks is also a whole industry, there are not so many manufacturers in the world (one example is the AMTC company in Dresden). There, too, you need equipment, raw materials, etc., etc.
Conclusions
You can't just build a microelectronics factory. Such a plant needs a huge ecosystem (consumers (many consumers), equipment, raw materials, personnel). Recently there was a translated article about such an ecosystem, And this ecosystem is very fragile, with the disappearance of at least one component, the whole system collapses. In my opinion, it is impossible to create such an ecosystem completely isolated from the outside world.
And what's next?
Short answer: I do not know.
What would it look like? When integrating into the global microelectronic ecosystem (having the opportunity to buy equipment, raw materials and the ability to sell products), choose a niche in which there is no fierce competition (as in the production of memory and processors) and try to take their share there by playing on lower labor costs and unique system solutions of intelligent local engineers. For example, in the field of RF microelectronics. As far as I know, in the already mentioned Angstrom-T there are (were?) good developments in such areas, and they could be in demand in the IoT, which is growing at a fairly rapid pace. Well, or some kind of power integrated electronics. Or integral photonics. Having a couple of high-tech factories built into the global ecosystem, you can already do some things that you don't want others to see.
Behind the iron curtain (meaning full localization from beginning to end), you can only do something like 80286 processors for a lot of money, nothing more. I think globalization is happening because it is impossible to advance alone above a certain technological limit - no country can pull, only the whole world. Whether we will be a part of this world is a separate question.