🌙 👧🏽 🎇 Who is who in world microelectronics 👩🏿‍🎓 🤘🏿 🍄

Typical news about electronics in early 2020: "Intel probably won’t place a production order with TSMC, but is considering a partnership with GlobalFoundries." Who are Intel - everyone understands, but what are GlobalFoundries and TSMC? When the trees were large, each microelectronic company independently produced its own microcircuits, or even equipment based on them, like some Toshiba or IBM. Since then, a lot of water has flowed, production has risen in price, the complexity of devices has increased, and dozens of high-tech companies from three continents participate in the creation of such a mundane and widespread device as an iPhone. The size of the global market for semiconductor chips and devices is estimated at more than $ 400 billion,but not all the giants of this market deal with end users and often appear on the news. But when they appear - they can be confusing. To prevent this from happening, I will try to briefly describe who is who.

Important disclaimer: there will be a lot of figures in the article, mainly financial, but please keep in mind that I am not a real financial analyst, but only an engineer armed with Google.

Market

Let's start with the market in which semiconductor companies are trying to make money. Its annual volume as of 2019 is estimated at 400-500 billion dollars in annual revenue. Such a wide spread is caused by the fact that due to the US-Chinese trade war, instead of the expected growth, there was a fall, so that, on the one hand, the lower estimate is closer to the actual state, but the upper one shows where the market could go right now if the geopolitical situation calms down . For comparison, the global market for electronic-based end devices is more than one and a half trillion dollars, including smartphones in 2018 exceeded 480 billion. Compared to the not (quite) electronic market, carmakers around the world receive 450 billion in one quarter.

Figure 1. The global microelectronics market (revenue) by end consumer.

The segmentation in this figure is very vague, but we will assume that it gives a general idea. Servers are probably included in computer, and communications are mainly smartphones. I would also like to note 1% of the state order (Russian electronics engineers laugh nervously) and 12% of the automotive industry. These 12% are important in that, unlike the more or less established and understandable other markets, the number of electronics in cars is growing rapidly - these are driver assistance systems, an electric drive, and just displays instead of dial gauges on the panel.

Figure 2. World microelectronics market (revenue) by type of product. Source

In this figure, we see which chips are sold. Please note that these are not pieces, but money. 5% penny discrete components in pieces would give much more than a hundred-dollar microprocessors. The largest share is in memory, it is mainly DDR and flash. Market volumes are such that key manufacturers are occupied only with memory or almost only with memory. This is also facilitated by the fact that both DDR and multi-level flash require special technology options that are usually not found in processes for other applications. Logic is all digital circuits that are not memory or processors: modems, interfaces, microcontrollers, and so on and so forth. Optoelectronic components - mainly LEDs (actively replacing incandescent lamps) and sensitive elements of photo and video cameras. Now let's seewho is making all this crazy money.

The biggest

Table 1. Top 10 semiconductor companies by revenue in 2019.

What do we see here? With Intel, Samsung, Toshiba and Nvidia, everything is more or less clear, these companies are constantly at the hearing. SK Hynix (formerly Hyundai's semiconductor division) and Micron specialize in memory. TSMC is the world's largest contract manufacturer of microchips (more on this later). Broadcom and Qualcomm are leaders in communications chips, while Texas Instruments are leaders in analog chips. It is worth noting that these ten companies are two-thirds of the entire market, and all the rest basically collect what the giants did not get their hands on. However, the remaining third of the market is still huge, and many great and small companies are working on it, many of which make unique, albeit niche, decisions.

An important addition to the table is that it shows only the revenue from sales of microchips. For example, Toshiba’s total revenue is about three times that of their semiconductor division. Samsung's total annual revenue is more than two hundred billion dollars. As benchmarks for comparison, you can use Apple (about 260, of which its own chip design corresponds to about 7 billion), Microsoft (125), Gazprom (120). The scale of Russian electronics can be illustrated by the sizes of Rostec ($ 28 billion) and Micron ($ 0.16 billion).

One more thing - three companies from the list (Broadcom, Qualcomm, Nvidia) do not have their own chip production at all. By a funny coincidence, all three, being the grandees of the so-called fabless model, outsource the production of their chips at the Taiwanese TSMC factory (although not only on it). Many companies with their own production, but placing orders to the foundry at the smallest and, accordingly, the most expensive design standards, do the same.

Figure 3. One of the iphone 3G (2008) boards.

Figure 4. Two boards with iphone XS (2018).

In these two figures, you can see how integration in commercial electronics has progressed over a decade. Instead of a large set of small microcircuits and discrete passive components, everything that is possible is now packaged in complex functional systems on a chip - for the sake of reducing the size and area of the board, reducing energy consumption and cheapening production. At the same time, one can observe how Apple has stepped up its own chip development in ten years. And since the release of the pictured iphone XS, Yabloko has been bought by teams that made most of the chips labeled as Intel products (modems, Qualcomm's main competitors) and Dialog (power management chips) in the figures. But even with this in mind, one can see that a trivial cell phone is a cooperation product of a dozen microelectronic companies and who else knows how many companiestheir suppliers and subcontractors.

Production

As you can see, in the top of table 1 there are companies from four countries: the USA, Japan, Korea and Taiwan. Does this mean that the entire global production of microcircuits is concentrated in these countries? Yes and no. About the fact that at least three of the ten companies produce their chips in Taiwan, I have already said. There are Intel factories, besides the USA, in Israel and Ireland, and the American Micron factories are also in Taiwan, Singapore and Japan. In general, the picture of world production looks like this:

Figure 5. World production of semiconductors by region, in thousands of equivalent two hundred millimeter wafers per month.

As you can see, Taiwan, Korea and Japan account for more than half of world production, and if you add China and Singapore and Malaysia, which occupy a large part of the “rest of the world”, then Southeast Asia will occupy three quarters of world production. Please note that the figure does not show money, but the number of plates. At the same time, plates made according to more subtle design standards are more expensive and usually bring more profits. If you take into account design standards, the picture is even more interesting.

Figure 6. Distribution of global chip production by design standards.

What do we see in this figure? The first is that half of world production is done according to the standards of 28 nm and higher (and this half in volume gives more than two-thirds of money). The second - the most subtle standards are disproportionately developed in Japan and Korea (this is a consequence of the placement of specialized factories for memory there), while the norms for mostly contracted Taiwanese production are more evenly distributed. Third is the absence of India. Paradoxically, the fact is that in India there is still no semiconductor production at all, and in general, microelectronics in it has until recently been extremely backward. However, in recent years, the famous Indian outsourcing programmers are also mastering the design of microcircuits, too, and development in India is growing by leaps and bounds due to the excellent combination of quality and labor cost. And the fourth - among the regions,where there is still chip manufacturing, Europe is clearly lagging behind. In the Old World, in fact, there are only four 300 mm factories: the Intel factory in Ireland, the STM factory in French Croll, the GlobalFoundries factory in Dresden (this is the same AMD factory whose old equipment was bought by the long-suffering Angstrom-T) and the Infineon factory too in Dresden. Another STM factory just recently began to be built in the vicinity of Milan, and that’s all.

However, the old design standards are also a multi-billion dollar market, and with skillful use, such factories can bring significant profits. Outdated equipment is being redesigned for the production of power electronics, MEMS, LEDs, and even discrete components. A fresh squeak of fashion in equipment designed to work with 100 and 150 mm wafers is the transition to silicon carbide, which is extremely popular in fast-growing power electronics. For example, STM did with its factory in Sicilian Catania, and now instead of vegetating, it is the main supplier of power transistors for the Tesla model 3. Factories with 200 mm wafers are already mastering silicon discrete elements with might and main - and the same carbide will be the next step for them.

Figure 7. Comparison of production on 200 mm plates in 2006 and 2018. Please note that the total production volume almost did not fall, despite the obvious obsolescence of design standards (200 mm is 130 nm or more). The “evidence” of obsolescence is clearly visible on the change in the share of memory in the total volume of production: there is no reason for it to remain at thicker standards if thin ones are available.

Equipment

One of the main problems of microelectronic production is its rapid rise in price with each new step of design standards. Modern factories cost billions, or even tens of billions of dollars, and the prices of technological plants are tens of millions apiece. The total annual market for manufacturing equipment is about $ 50 billion.

2. .

In the production of equipment, by and large, there is only the USA and Japan. Only the Dutch company ASML, which is engaged in a relatively narrow sphere - lithography - but has more than 80% of the world market (Canon and Nikon fight for the rest), falls out of the general picture. Moreover, at the advanced design standards, ASML is a monopolist, which, to my taste, is very surprising and that from time to time leads to interesting consequences. For example, most recently, the Chinese microelectronics update program came up against the fact that the Dutch government, at the request of the American, banned ASML from supplying the latest lithographic steppers to China. In general, there are ITAR limitations for these purposes, but to the surprise of the Americans, ASML products turned out to have less than the required 25% of American technologies, and instead of the usual householder ban, I had to bow to the Dutch.

Foundry

As I said above, when production equipment was cheap, all microelectronic companies themselves developed and manufactured their own microcircuits, and many manufacturers of electronic equipment had their own microcircuit production for domestic needs. But already in the seventies, the entry threshold began to grow, and it became very difficult for promising garage startups to enter the market. To solve this problem, the MOSIS company was created at ISI University of South California, collecting small orders from everyone and placing them in groups at the factories of large companies that were underloaded with their own production. Although the experience turned out to be successful and contributed significantly to the development of microelectronics, including the first experiments with RISC architectures, third-party customers always had lower priority in factories than internal ones, which didn’t help business.

A major turning point in the industry came from a Taiwanese engineer named Morris Chang. Born in mainland China in 1931, he left college in 1948 to study in the United States, where he made a successful career at the peak of which were vice president of Texas Instruments and CEO of General Instrument. In 1985, when Chang was about to retire, the adviser to the president of Taiwan and former prime minister Sun Yunxuan (who is called the architect of turning the country into a powerful industrial power) invited him to the post of director of the Industrial Technology Research Institute (ITRI). So Chang returned to his homeland and got the opportunity to implement his long-standing idea - to create a factory that will not develop and sell anything,and concentrates solely on the development of technology for the needs of third-party customers. This factory was founded in 1987 under the name Taiwan Semiconductor Manufacturing Company or TSMC.

The TSMC business model turned out to be extremely successful and allowed to turn the fabless model into a huge stable sector of the global economy, give small companies from all over the world access to advanced technologies, and make Taiwan one of the most important world economies. Of course, TSMC had a lot of followers, but it was this factory that remained the most successful, having a little more than half of the global contract manufacturing of microchips and actively developing the most advanced design standards. TSMC customers range from giants such as Apple and AMD to very small companies around the world.

Table 3. Largest contract manufacturing of microchips as of 2019. For Samsung, indicative figures are given only for contract manufacturing for external customers.

As you can see, in order to make good money, it is not necessary to have the most modern technology - although it is still worth having something good. Good, however, does not necessarily mean design standards. For example, GlobalFoundries, abandoning the fight against the TSMC-Samsung-Intel trio in adopting the 7 nm standards necessary for consumer segment processors, relied on 12-28 nm FDSOIs, which are promising for the Internet of things and microwave processes, and a number of dozens left behind factories offers integration on a single chip of logic and high-voltage power transistors, elements for microwave circuits and other "chips" in demand in specialized industrial and automotive chips.

8. FDSOI . FDSOI MOSFET, ( FinFET), , .

Figure 9. TowerJazz technologies.

And here is the offer of the Israeli-American factory TowerJazz (Tower is the Israeli part, and Jazz is the American one). The processor-digital logic has stopped at the level of 45 nm, but the offer has actively grown in breadth, covering many other applications. Three lines of radio-frequency technologies, two lines of analog and power processes, optical chips and MEMS, as well as the availability of automobile and military certification can satisfy the needs of a wide variety of customers, forming a portfolio of not a few large orders, but of a large number of small ones.

Fabless

Table 4. The largest fabless chip developers.

Of all the companies mentioned in the article, the participants in this table are the most understandable to the average person. Qualcomm Snapdragon is the de facto standard of processors for mobile phones on Android; Broadcom - a variety of network equipment (and a chip in the Raspberry Pi), Mediatek and HiSilicon - are also known as manufacturers of accessories for phones, with apple trees and AMD everything is clear without me. Xilinx is a little more complicated, but everyone who once heard the term “FPGA” is aware of them, and Marvell and Novatek are already so small that only specialists and exchange players need to know about them.

But how among companies without their own factories was such an old and respected company as AMD? The answer to this question is quite simple: in 2009, immediately after the devastating financial crisis, AMD, as part of a business recovery campaign, separated its production into a separate company called GlobalFoundries and sold it to investors from the United Arab Emirates, becoming an anchor client for the new company ( in fact, just leaving everything as it was, but taking off infrastructure costs). After that, the same investors bought and joined GloFo a large Singapore factory Chartered Semiconductor (also partnered with AMD) and IBM's semiconductor division (including, inter alia, factories in the vicinity of New York that were admitted to the US defense industry). The resulting conglomerate became the second largest contract factory in the world (however,five times smaller than TSMC) with factories in the USA, Germany, Singapore. Fun fact: GloFo is the largest private employer in the US state of Vermont.

The subsequent arms race with the leaders - TSMC and Samsung - did not end very well for GloFo, with the loss of its largest customer (AMD) as a result of the inability to launch 7 nm standards. However, this shock was beneficial for the company and allowed to start the process of disposal of inconvenient assets and reorientation (albeit forced) to other markets, as a result of which the financial position of GloFo has significantly improved. AMD partially returned (for I / O chips, which are suitable for 14 nm standards), orders appeared in the areas of the Internet of things and 5G. In recent weeks, there are also rumors that the factory has the prospect of getting a tasty contract from Intel, which lacks its own capacities at 14 nm. If the deal goes through, and the former AMD factories will produce blue chips, it’s pretty ironic.

IP and CAD

In general, IP is an intellectual property, but in the semiconductor industry the concept of “IP core” is actively used. Such cores are complete parts of designs that developers do not produce, but sell to other developers. Such an approach allows companies selling microcircuits to buy ready-made standard blocks (for example, controllers of popular interfaces), independently creating only that usually small part of the chip that makes up the know-how and creates the main added value. Read more about the IP ecosystem here.and now I’ll say that this market is relatively small (about $ 3.5 billion), but growing steadily (doubling is expected in five years) and important for the industry as a whole because its existence greatly reduces time to market and improves the quality of standard spare parts, which are deeply involved in the topic of professionals (well, in theory), and not the auxiliary teams of chip manufacturers.

You are probably familiar with the key player in the semiconductor IP market: it is an ARM company that develops processor cores of the same name and licenses them to everyone, for example, to almost all manufacturers of processors for mobile phones. ARM's share of the semiconductor IP market is 45-50% ($ 1.6 billion in revenue in 2018), the other two big players are the IP subdivisions of CAD manufacturers Synopsys (629 million out of 3300) and Cadence (189 million out of 2100). The remaining third of the market is almost evenly spread between hundreds of small teams specializing in some narrow tasks.

The electronic CAD market, of which the IP market is sometimes considered part, is approximately $ 10 billion. In addition to the above-mentioned Synopsys and Cadence, the “Big Three”, which controls almost the entire market, includesMentor, a Siemens Business (formerly known as Mentor Graphics) with revenue of $ 1.3 billion. An interesting feature of the microelectronic CAD market is that the products of competing companies are not only compatible with each other, but can even be integrated into each other, allowing developers to assemble the design route according to the specifics of the task.

Figure 10. The Cadence Virtuoso window and one of the tabs, which is actually the Caliber software developed by Mentor Graphics. As you can see, Stirlitz did not give out anything, except for a warmer gray color. And yes, modern microchip design software looks exactly like that. This is still a graphical interface, not a command line.

If not silicon, then who?

Despite the enormous amount of money invested in the development of silicon alternatives, which is obviously not even about the economic and technological, but the physical ceiling, nothing substantial can be expected in the coming years. Candidates who appear to be the “material of the future” come and go (gallium arsenide, carbon nanotubes, graphene, and so on and so forth), and silicon MOS transistors do not go anywhere. However, there are quite a few market niches in which other materials successfully compete with silicon, or even completely displace it. These niches, on the one hand, make up at best a few percent of the total market, and on the other hand, we are talking about billions of dollars.

Almost a billion dollars is estimated the market of emitters for semiconductor lasers, which are complex heterostructures of GaAs, AlGaAs, InGaAs and so on and so forth. By the way, Zhores Alferov received the Nobel Prize in Physics for the discovery of these heterostructures. Or for example, LEDs, also made of heterostructures based on materials A 3 B 5 , rapidly penetrating into our everyday lives; while the LED market is worth a few billion, but the global diode lighting market has already exceeded fifty, and it is expected that it will continue to grow at least until incandescent lamps disappear completely.

Another interesting story is the market for high-power high-voltage devices, where silicon MOS transistors and IGBTs began to rapidly yield to silicon carbide (SiC) MOS transistors and high electron mobility transistors (HEMTs) based on gallium nitride (GaN). And those and other beginners due to the best properties of the material can achieve better parameters than silicon can provide even theoretically. The same newer devices take away part of the microwave market from GaAs - thanks, for example, to a better temperature range. The main markets where new products are introduced are electric cars, base stations of 5G networks and, for example, chargers for mobile phones.

And in the news they regularly write that some scientists have achieved a breakthrough in the use of a new semiconductor. All these messages, of course, must be divided by ten, but even with this in mind, a lot of new interesting materials are really on the way: gallium oxide (Ga 2 O 3 ), diamond, new variants of compounds A 3 B 5 and A 2 B 6 , perovskites - this is just what I saw in the news over the last couple of months. None of them will replace silicon, but they will make our solar panels more efficient, the Internet faster, and lasers more suitable for installation on huge humanoid robots.

A little different

The microelectronics market, of course, is not limited to the above. Around it there are many more related markets, for example, measuring equipment with giants such as Keysight (4.3 billion), Rohde & Schwarz (2.04 billion) and National Instruments (1.02 billion). Or the production of silicon wafers proper, on which chips are made, is also a ten-billion-dollar market, with its own whales, which an ordinary person never hears. Yes, there’s an ordinary person, even I, as a chip designer, knew only one of the five top manufacturers of substrates before starting this article, and only because the French Soitec is the leader in silicon substrates on an insulator, to which I have a professional interest. By the way, the popularity of SOI substrates is growing rapidly, and, funny thing is,the leaders in their use are European factories GlobalFoundries and STM, working with FDSOI technical processes for the Internet of things and connected radio frequency circuits. Oh yes, just in case: the world's largest manufacturer of silicon substrates is the German company Siltronic, whose headquarters is located in the Bavarian city of Burghausen with a population of eighteen thousand people.

Figure 11. Panorama of Burghausen.

Microchips, in turn, serve as raw materials for the trillion-dollar electronic industry, which today covers generally all spheres of human life. Moreover, the production of final products almost always turns out to be more profitable than the creation of universal spare parts. On the other hand, if we are not talking about spare parts, but about something specific, then having become the best in our niche, it is quite possible to build a billion-dollar company - as we can see in the examples of Xilinx, Qualcomm or, for example, Skyworks - which are almost the size of AMD, whose products are in a huge number of smartphones, but about which you probably do not know anything.

On the other hand, competition in the market leads to the fact that manufacturers of final products try to pick up the key core competencies for themselves, without outsourcing them - a kind of partial return to vertical integration. Of course, IT companies are not talking about building their own factories, but Google, Amazon and Facebook have started creating server microprocessors, Huawei is expanding and expanding its microelectronic “daughter” HiSilicon, and in the case of Apple we can see (for example, in Fig. 3 and 4) how, in the design of microchips, they went from completely external components to some of the best processors on the market, after which they began to actively buy the relevant departments of their suppliers - Intel and Dialog, but did not bother with graphics and lured Imagi employees nation in a nearby building. However,in the latter case, the prank was not entirely successful, and already in 2020 the parties, after numerous mutual threats by the courts, announced the signing of a new licensing agreement. Nevertheless, the market is becoming more competitive and competitive, so the desire of IT giants to ensure their superiority at the hardware level is quite understandable.

This article could be completed, if not for one “but”.

And what about Russia?

I wanted to do without this section because it is too sad, but it would be dishonest to the readers. So, the volume of the Russian microelectronics market is 0.7% of the world market, and defense industry enterprises provide most of this volume. According to the results of 2019, the largest chip manufacturer (Mikron) had revenue of about 10 billion rubles ($ 160 million or 0.5% of TSMC) and for the first time in ten years (!) Did not turn out to be unprofitable. For comparison, the revenue of Yandex for 2018 is 126 billion rubles.

In 2007, Mikron licensed 180 and 90 nm manufacturing processes from the French-Italian company ST Microelectronics. Micron processes with 180 nm standards are designed for the production of radio frequency tags (RFID), which you can already find on public transport tickets, and in the future, in everything that can be marked - fur coats, passports and even smart glasses. At 90 nm there is only pilot and small-scale production. At the same time, Micron operates as a foundation (according to the standards of 180 and 240 nm) for many Russian fabless companies, being one of the main centers for import substitution for dual-use microcircuits sanctioned. At the same time, the range of manufactured dual-purpose chips is very wide, and the circulation is small, therefore, to maintain the factory, it definitely needs large-scale civilian orders - which provide metro tickets.

12. 90 “”, “”.

The second largest microelectronic production in Russia is ... no, not Zelenograd Angstrom, as many might think, but the Bryansk Silicon El Group, which has production with design standards of 700 nm (and recently announced the development of 500 nm and plans for 350 nm). The company's revenue in 2017 is 3 billion rubles, in 2018 - 2.6 billion rubles, products - discrete components (transistors and diodes), power modules based on them, as well as analog and power chips of little complexity, almost completely oriented to the needs of the military-industrial complex. They also have the latest issues of the factory newspaper “Crystal” posted on their website, “weeks without turnstiles” are regularly held at the production site and active cooperation is being conducted with the local university. In general, well done from all sides.

Figure 13. Silicon El plate under a microscope.

Angstrom, an eternal rival of Mikron, which is in third place of honor, showed production revenue of 2.17 billion rubles (and another 600 million gave R&D) in 2018 - and, for the first time in several years, hanging around zero, losses. Angstrom does not have modern design standards, therefore, interesting news from the enterprise recently mainly concerns discrete powerful devices - a line of radiation-resistant power transistors has been developed, diversification into civilian markets is being carried out, an agreement was signed with the Japanese on the development of silicon carbide for extremely promising direction of high-voltage power devices.

Figure 14. Powerful IGBT module manufactured by Angstrem. Please note that the company creates not only the devices themselves, but also the final products based on them.

Angstrem-T, which is not part of Angstrem, and which promised to start production at 90 and 130 nm on equipment purchased from the Dresden AMD factory, has been mostly involved in bankruptcy and loan news throughout its history. And about the fact that Mikron proposed to build a factory with 28 nm standards in its building. Of the relatively good news, there was only a scanty contract for such an enterprise with the Chinese for production at 250 nm standards, which apparently were transistors cut off from 130 nm technology for input-output circuits.

Another serial production is Voronezh (VZPP-Mikron, taken into account in the results of the parent company), and the rest are experimental and small-scale factories focused on the military-industrial complex, many of which have not actually been updated since Soviet times. Separately, it is worth mentioning the Belarusian “Integral”, which works almost entirely on the Russian market, with its revenue of about six billion Russian rubles, the minimum design standards of 800 nm and the ability to make devices with an operating voltage of up to 600 V.

With Fabless development, the situation is somewhat better than with production. There are dozens of design centers with billions (in rubles, of course) of revenue in Russia, they have successfully established cooperation with both foreign silicon factories at rates up to 28-16 nm (mainly with the same TSMC) and with suppliers of popular IP blocks and cores. There are also several branches of large foreign companies, for example Intel in Nizhny Novgorod, where more than a thousand people work. The problem of Russian microelectronics as a whole is that, as in production, the MIC also rules the development, and most of the chips being developed, although they have important advantages such as radiation resistance or an extended temperature range, are in principle not intended for large-scale production. A number of companies have commercial ambitions (e.g., MCST,“Baikal”, “Milandra” and “Modula”), but they are all in the early stages and are still oriented towards the state order, albeit not military.

The industry has high hopes for the just adopted state “Strategy for the Development of the Electronic Industry of the Russian Federation for the period until 2030”, which implies a significant leap forward, overcoming the long-term lag behind the rest of the world and the reorientation of Russian electronics on commercial tracks. Own factories will have to be built at 5 nm standards, technological equipment for them will be developed, Russian enterprises will enter the world market - and all this very soon, literally five years after the Russian cosmonauts landed on the moon.

On this infinitely optimistic note, I, perhaps, will round off.

Who is who in world microelectronics