The global supply-chain crisis affects all kind of sectors – there are currently neither wet-wipes, rice crispies nor blood transfusion tubes available at the hospital where some of us work, due to supply-side problems. The major shortage, though, relates to microchips. We translated this article written by comrades from Wildcat. It will surprise many of us to read how labour intensive nano-chip production actually is…

In Germany alone in the first quarter of 2021, one million cars were not built because the necessary microchips were missing. Worldwide, the figure is expected to amount to more than four million vehicles. The automotive corporations – or rather their suppliers – had cut their orders a year earlier because of the Covid pandemic – and are now left with nothing. Given the global production structure, it is not possible to ramp up the production of microchips within a short period of time. 70% of all car microchips come from a single manufacturer in Taiwan. There is a local water shortage, chip production consumes an extremely large amount of water, and in general, they are very environmentally damaging to produce. On top of that, there are transport problems. And on top of this, there was a Covid outbreak in the chip industry in Taiwan, caused by the shortening of pilots’ quarantine to three days so that the chips could de delivered to customers worldwide as quickly as possible…

Pretty much all the problems of current capitalism are concentrated in what (not just) the automotive industry calls, the “chip crisis”.

It is no coincidence that the media is full of news proclaiming “ways out of the chip crisis”. Under German leadership, the EU has announced billions in investment into microchip production. At the beginning of June, Bosch celebrated the commissioning of the “most modern chip factory in Europe” in Dresden – “only” three years after the foundation stone was laid there! – and “as early as” the beginning of 2022, the first saleable microchips from this factory are expected to hit the market. They will primarily be used for the automotive industry and factory automation (the chips produced in the next few months will be installed in Bosch power tools). Bosch CEO, Volkmar Denner, boasted that production would be “fully digitised and networked”. “With the help of artificial intelligence, we’re taking semiconductor production to a new level.”

In the following interview with a comrade who has been working in this field for decades,it becomes clear that high levels of automation can only be achieved where it can be based on relatively conventional and stable processes. In Dresden, they use 300 millimetre wafers (silicon wafers on which the chips are created), which is the current modern standard, but the chips are made in 90 nanometre processes. This is relatively outdated, (partly in order to facilitate a more automated production), compared to the new factories in Taiwan, which produce in 5-nm processes, requiring much more manual labour. 

Where are chips produced today?

Siemens and Philips have been producing semiconductor devices since the 1950s. The first integrated circuits came out of the Valvo plant in Hamburg, which belongs to Philips, in 1966. In 1985, U.S. manufacturer LSI began producing microprocessors in Braunschweig; they closed down in 1992: “too high wages” in the face of a dwindling dollar made production unprofitable. [1]

The semiconductor industry was gradually outsourced to East Asia (especially Japan) in the 1990s. Infineon [2] still manufactures in Dresden itself and is currently expanding in Villach (Austria) – as well as in Malaysia and China. In 2005, the then Infineon CEO declared that nothing would be produced below 90 nanometers – and this has just been confirmed again by the Bosch factory in Dresden. 

In the U.S. too, outsourcing began in the early 1990s. In the early 2000s, the U.S. still accounted for 37% of global semiconductor production capacity. Today, the U.S. accounts for only about 12% The official justification was always labor costs, although that was not the only problem. For example, toxic manufacturing in the U.S. had also come under fire. They could no longer afford to do these chemical processes in the U.S. itself.

Companies used to manufacture the entire microchip in their own factories. There are two production models: IDM (Integrated Device Manufacturing) and the foundry concept (“silicon melting”). In the early 1990s, the foundry concept, i.e. the separation of design and manufacturing, gained more and more ground. The founder and longtime head of AMD, Jerry Sanders, resisted this development for a long time and became famous in 1992 with his saying: “Real men have fabs!” (short for in-house fabrication plants) [3] Later, AMD also ended up outsourcing production. Today, only Intel still manufactures microchips in its own factories. The foundry model gave rise to factories in Southeast Asia. Today, only South Korea (Samsung), Taiwan (TSMC and UMC) and mainland China (SMIC) have large foundries that manufacture for their clients as sub-contractors. These semiconductor fabrication plants (“fab” for short) are purely process-oriented; the process is defined by the smallest structure size that a fab can produce, which has been specified in nanometers since 1999. [4]

These factories have customers in different industries. Demand has always been cyclical. Sometimes everything depends on the manufacturing of chips for server infrastructure, the so-called data centers. Then it’s production for the infrastructure of cell phones, then chips for the cell phones themselves, and so on. There is no one company that covers everything. Intel essentially builds central processing units (CPUs) for PCs and servers. Samsung essentially for cell phones.

Global automobile production peaked in mid-2018. Since then, sales have fallen sharply. In 2019, the auto industry suffered from sales problems. Nevertheless, complaints about long delivery times for chips (from order to delivery, in some cases a little less than a year) were already coming in at the end of 2019 – before the Covid pandemic. Counterintuitively, orders for chips were reduced or canceled with the onset of the pandemic  – except for Toyota, which had learned from the 2013 tsunami that supply chains can’t be easily switched on and off, and acted differently this time. In the Spring of 2021, familiar problems arose: entire plants were shut down and workers were put on short-time work because cars could not be built due to chip shortages. 

Can you briefly explain what specific chips the auto industry needs?

The essential component is the so-called MCUs (or “microcontrollers”), they are in every engine control unit and every sensor. Cars have a great need for ‘specialty chips’ that are very robust and can withstand very high temperatures and voltage peaks. While computer chips are qualified for a temperature range of 0 to 85 degrees Celsius, the simplest requirement for microchips in cars is -45 to +125 degrees. Internal combustion engines need ignition, which works in the range of kilovolts. Such a high voltage is not found in other sectors at all. These chips are produced on old production lines, which are only maintained for this purpose. These are mostly still 200 mm fabs (production with silicon wafers of 200 mm diameter started in 1992). Since 2002, the standard has been 300 mm. TSMC also still has 200-mm fabs, and 70% of all chips built into cars worldwide come from there. These factories cannot expand production so easily, because the machines for this type of chip production can, if you’re lucky, mostly be found on the second-hand or scrap market (and prices are exploding there right now).

Tesla has less problems with this issue. For their E-cars, they primarily use chips, which are also used in cell phones. These chips are similar to normal computer chips. Once a car no longer needs an ignition, this source of interference is eliminated. However, they sometimes have quality problems with these standard chips when they are used in automotive electronic control systems.

When the car industry cut back on orders, TSMC used its manufacturing capacity for other customers. Shutdown is fast! And during the Covid pandemic there was an unpredictable demand from people who were now working from home or attending virtual conferences, now needing new computer devices. TSMC could have sold twice as many chips as they did! Graphics cards have become much more expensive.

The automobile industry’s just-in-time strategy hit a wall. Firstly, because the car industry has nowhere near as much power in relation to the chip-makers as they do with their other suppliers -they’re just not a major customer! Secondly, they have shifted far too many functions to their direct suppliers: warehousing, technology development, but also the entrepreneurial risk – and smaller companies can no longer shoulder that. VW is now suing its suppliers, saying they should have ordered more chips in Spring 2020 – and paid for them in advance – with no guarantee that VW, as the main buyer, would actually need them. And third, it takes at least 13 weeks, (realistically more likely 26 weeks), for chips to go through the production process. Just-in-time can’t work like that! By the way, it’s not about “stockpiling” in the physical sense. The chips ordered are stored at TSMC, in a so-called die bank. This means that the customer pays in advance, but does not receive the goods until they are needed.

How does the manufacturing process work in practice?

It is a photochemical process. Memory chips have about nine exposure steps, logic chips 13-15. The wafers come from the supplier. In the factory, they are first treated to make them conductive (doping), then fine etching structures are applied, which are actually metal layers. Nowadays these are three-dimensionally etched structures. Then the wafers are exposed, then the coating / etching layer is washed off again. Then comes the next process of etching, and so on. This is the central line with the automatic exposure machines. The whole process takes at least twelve weeks, and with each step the wafer becomes more special and more expensive. At the end, the chips are tested, and then it is decided whether they will go into a car or just into a washing machine.

Does TSMC manufacture the complete microchips or just the electrically complete wafers?

TSMC is primarily a wafer factory. Further processing (cutting the wafers into individual dies, assembling and packaging) is done by others or they do it in a separate factory they own. AMD, for example, buys the complete wafer. TSMC also does packaging for other customers.

How far ahead of the competition is TSMC, and who buys which chips?

At the moment, only Apple is using 5 nm chips, Nvidia 7 nm. The processes are one to two years apart. TSMC is offering 3nm to its customers in the second half of 2022. The first fab for that is in the process of being built right now. And Apple has booked these 3nm chips for a long time!

How many people work in a chip fab?

Half of the employees in a fab are in marketing and process preparation. In production, a little more than half are highly skilled process engineers, and they also work twelve-hour shifts at TSMC. There’s nothing but round-the-clock production there. Starting salaries are around $100,000, high for Taiwan. A quarter of the workforce is female, so it’s almost all women in the direct production department. They operate machines and transport the products, intervene in case of malfunction. These are all processes that are very prone to technical malfunctions. In Taiwan, there are 5428 “managers,” 24,809 “professionals” (engineers), 17,414 “technicians” (also called “operators” – these are the female workers) and 4,394 “assistants” (office and cleaning staff) working in ten TSMC fabs. [5]

These are all highly automated processes – but despite this, about 1,700 workers work directly in the manufacturing department in each of the factories?

At least theoretically, stable processes can be automated excellently. But just think of all the problems with robotics! Engineers are as happy as children when a robot picks up an egg without breaking it – which is no problem at all for a human being. The products are highly sensitive. The new processes are so short-lived that they can’t get out of the optimisation phase and into general serial production phase at all. Everything has to be highly flexible, they’re constantly developing.

The exposure processes are automated, of course, because that’s not possible manually at all. The machines for this are extremely expensive. By the time a mechanised process is ready to go, two generations of Apple phones will have hit the market. To use human labour instead of machines is simply faster and much more flexible. Actually, something goes wrong in production all the time. People literally climb into the machines to fix faults.

Even the raw wafers are highly sensitive. The wafers are carried by hand, which works better than robots, because people know how to move them like a raw egg, from A to B. And besides, with robots, you always have the problem of abrasion – that doesn’t work at all in the production department, which has to be absolutely dust free! So if possible, no moving parts should be used. It is a huge effort to even get a device into such a ‘clean room’, it can take ages. You can only enter a clean room where mass production takes place by passing through a dozen airlocks, which takes an hour, until you have the necessary degree of cleanliness. You can’t just get out, you have to go through a twelve-hour shift without eating or drinking – I honestly can’t imagine how that works!

The process engineers rarely have to go into the clean room, do they monitor the processes from the outside?

It can be roughly summarised as follows: The task of the workers “on the front line,” i.e., in the clean room, is to ensure the output of chips under all adversities; acute problems must be remedied directly as far as possible, which is what they are trained to do. The task of the process engineers is to prevent incidents in the future. This means, for example, reprogramming machines in such a way that less intervention is required in the future. The work of the process engineers and the workers in the clean room goes hand in hand, and the experience and knowledge gained is poured into continuous process improvements, readjustment and reprogramming of the machines, etc.

Do you have an overview of wages in Taiwan?

The average wage at TSMC in Taiwan in 2019 was about $59,000 – the average household income in Taiwan in general was $14,331 per year. (Female) production workers earn about three times the average Taiwanese wage, but their pay is variable; in 2019, they were paid 26 months’ wages! In return though, they have to work twelve-hour shifts. Most of them are young women who start there after passing their A-levels and then quit after four or five years and get married. While very young women work in production for the most part, the process engineers are mostly old farts like me.

Where does this insane pressure to make chips even smaller and even more integrated come from? Who is exerting this pressure? Not the competition.

They are contract manufacturers. The pressure comes from the customer. 51% of TSMC’s production goes into smartphones. Samsung builds its own chips.

If Apple says, “We want smaller chips,” might TSMC say, “it’s not profitable?”

Why would a company management say they can’t do something? They all get subsidies, like the nuclear industry used to, it’s part of the business model. Semiconductor suppliers are trying to fit more and more functionality onto the same chip area at the same cost, and that means fewer nanometers every year. This is also more lucrative for the foundries. Apple, for its part, is under pressure to get its customers to throw away their cell phones every two years and buy a new one.

So an insane amount of money “from the taxpayer” is being used to push a technological development beyond a tipping point?

There is additional pressure from another side: Amazon is already outsourcing its data centers to Alaska because of cooling issues. Smaller chip structures require much less power and generate less heat while providing higher “performance!” And the capitalist hunger for more is unchecked: cryptocurrencies, more and more pixels on bigger and bigger TV screens, virtual reality and gaming in real time for the home-schooled, self-driving giant trucks on Swedish and other roads, etc. etc. Therefore 3 nm instead of 5 nm instead of 7 nm – and the 2 nm are already on their way.

However, such chips with increasingly finer structures are also much more sensitive – even the operators of large data centers notice this, where “moody” processors that calculate incorrectly are now a more frequent occurrence.

By the time they come onto the market, server capacity will have doubled or tripled…

Yes, from the standpoint of capital accumulation in general, it’s not a stable situation. But semiconductors make it possible to switch more quickly – for example, from the internal combustion engine to the e-motor – so “only” software is needed. That’s the promise. And the rapid growth of companies like Apple or Amazon was not conceivable with a mechanical process.

Geopolitically, this has led to the situation where 70% of all chips used in cars come from a single manufacturer in Taiwan, and that currently they are the only ones that are capable of manufacturing with a 3-nm process…

The entire semiconductor industry is concentrated in three countries. There is, for example, only one (Dutch) company worldwide that builds image setters for these new processes! This company in turn has 5,000 suppliers. This creates extreme dependencies. If anything goes wrong in the whole chain, parts are missing. Pandemic, ship in the Suez Canal, water shortage … not to mention a major accident or war. The fragility of globalisation, which they had simply ignored until now, is impressive! That’s why strategists are now cheering Biden’s aggressive China policy. Because what will happen if Taiwan becomes a part of China? Ten years ago, the Taiwanese government said: TSMC is our life insurance. Now TSMC wants to build six factories in the USA! With subsidies, of course.

In Arizona, they are already hiring people for the new 5-nm production and 3-nm is projected for 2022. It will be interesting to see what kind of people they hire there! Is it a matter of low labor costs or people with good school degrees who will do the job for a few years? And where will they find such workers? Women in the U.S. who are well educated need a career path. They need to offer them something. A lot can be regulated by the wage level, but not everything! And TSMC can’t pay three times the average wage paid in the U.S. if they want to bring competitive chips to market! Maybe they just bring the people with them? The other problem is the high toxicity of chip production….

Where does the water consumption occur? Can the water be cleaned again?

Water is consumed during the entire process. A fab in Taiwan uses 15 million litres of water per day. In Taichung, the government drilled 88 new wells for TSMC so that the fab there would have enough water. The Taiwanese have had to conserve water in their households in recent weeks, and two days a week their water is cut off so the fabs can run.Meanwhile, chip production has also been curtailed due to the water shortage. Only now has TSMC begun to build a pure water treatment plant in Taiwan, which will be operational by the end of the year and will provide half of its water needs from waste water in the medium term (the plant will be exclusive to TSMC, of course – after all, people can tolerate more impure water than chips can!) In addition, the chip factories need 3% of Taiwan’s total electricity demand. In Germany, the aluminium industry is the industry with the highest electricity consumption: 8.1 terawatt hours are needed by the entire aluminium industry in Germany per year. But 6.3 terawatt hours are consumed by a single 3-nm fab from TSMC. The new image setters consume 20 times more power than the old machines.

We learned during the Covid crisis that neither masks nor vaccines nor semiconductors can be produced in Germany. How realistic is it to assume that there will be new chip factories in the low nanometer range in Germany?

I don’t think the EU will put ten billion Euro on the table – this is the figure Intel gave. TSMC has not yet commented, they won’t build a factory without subsidies in the double-digit billions. And aside from the question of subsidies, can these production processes still be set up and productively controlled in Germany today?

The 89-year-old founder and former head of TSMC, Morris Chang, celebrated the “workhorse mentality” of Taiwanese workers in an interview with the Asia Times in the Spring – and boasted that the U.S. might not have a shortage of “cheap land and electricity”, but rather a shortage of “competent technicians and workers”, “because industrial jobs have not been popular with U.S. Americans for decades”. Managers could be sent to Arizona – but they alone would not be able to achieve similar productivity there. [6]

In the interview, Chang also made clear that the current global technological race in chip production is exclusively between South Korea (Samsung) and TSMC; China is at least five years behind, despite billions in subsidies – and will fall further behind if they try to develop independent technology.

Given these problems, I doubt that the European capitalists are even seeking ‘to bring back jobs’. Infineon has been quite clear on this; they would invest in more 300mm fabs (with subsidies flowing freely), but explicitly not in the single-digit nanometer range.

* 36 years ago, we translated an article from Processed World for our magazine Wildcat (then still called Karlsruher Stadtzeitung), issue 34, January 1985: “Chips of our lives”, which dealt with the electronics industry and the work in Silicon Valley – from the system analyst to the programmer to the PCB assembler in the clean room. It picked apart the myths about the microelectronics industry as clean and pointed to the drastic poisoning of workers, environmental degradation and, above all, working for the Pentagon. The magazine Processed World was based primarily amongst ‘tech workers’ and hoped that these would become more politically radical. In subsequent issues, we published reports on our own attempts at militant intervention in circuit board assembly at Siemens. In issue 40, November 1986, there was another article from Processed World, “Chemicals Run Amok in the ‘Clean Room,'” about the unhealthy work in the semiconductor industry. Strikingly, the computer industry was a topic in many left-wing magazines at the time – just before many leftists ended up in the software industry themselves. All issues of Processed World can be read at libcom.org.

[1] According to U.S. manufacturer LSI Logic Corp., “productivity-reducing labor laws and high labor costs” forced the closure of the chip factory in Braunschweig; thus the manufacturer’s concept of producing its 32-bit RISC microprocessors, ASICs, ICs for image and signal processing, and PC chip sets and graphics products close to the market failed. However, the wafers with the chips came from the USA at that time. Setting up its own lithography process might have improved the cost structure.

[2] Infineon was spun off from the Siemens components plant in 1999 and floated on the stock market.

[3] Interesting: John East, “Real Men Have Fabs Jerry Sanders, TJ Rodgers, and AMD,” semiwiki.com, July 29, 2019.

[4] This used to have real meaning, it could be measured physically. Today, it’s a “virtual quantity”; see e.g.: https://en.wikichip.org/wiki/ technology_node.

[5] All TSMC fabs: 300 mm: Fab 12 (Hsinchu), Fab 14, Fab 18-1, Fab 18-3 and Fab 18-4 (Tainan), Fab 15 (Taichung), Fab 16 (Nanjing, PR); 200 mm: Fab 3, Fab 5 and Fab 8 (Hsinchu), Fab 6 (Tainan), Fab 10 (Shanghai, VR), Fab 11 (Camas, US), SSMC (Singapore, with NXP).

[6] Asia-Times, “TSMC founder doubts US competence in chip-making,” asiatimes.com, April 2021.—