Chips with everything

The technology stack powering the European economies was supposed to be impervious to geopolitics; entrusting its construction to the market seemed to be the way to go. It was, in the end, an extremely foolish bet.

  • Español
  • English
  • Français
  • Deutsch
  • Português
  • Análisis
-A +A

As crises go, this year’s chip famine — the global shortage of semiconductors that run electronic devices, from PCs, cars and washing machines to toasters and gaming consoles — has been a weird event, not least geopolitically. In May, US companies wrote to South Korean president Moon Jae-in, urging him to pardon the disgraced Samsung chair Lee Jae-yong, currently serving an 18-month prison sentence for bribery. Given American chip dependence, it was imperative Samsung should proceed with its planned multi-billion dollar investment in chip manufacturing facilities in the US. With the US’s ‘semiconductor sovereignty’ at stake, there was no mention of the rule of law.


The dialectic of the chip crisis would have delighted scholars of the Frankfurt School, if only for exposing the inner dumbness of today’s smart society. The scarcity of chips has forced us to wait for the latest consumer electronics: without semiconductors (some at only $1 a piece), they cannot function. All those electric cars, smartphones, smart fridges and toothbrushes have suddenly disappeared into the black hole of global capitalism, as if an invisible spoilsport had cancelled the Consumer Electronics Show in Vegas.


Today’s crisis isn’t exceptional. This time, however, it has come amidst wider anxiety about globalisation, the decline of western industrial activity, and politicisation of advanced technology such as AI, now a strategic domain in the US/China standoff. This explains how a boring technical issue, which ten years ago would have had little impact outside the directly affected industries, has become a massive headache for governments.


This had something to do with the pandemic. Global lockdown was made bearable by an unprecedented increase in consumption of digital services, all requiring chip-dependent devices, routers and servers. It also pushed bored consumers to upgrade to slicker household appliances, boosting demand for blenders, rice cookers etc.


Slowed production


The pandemic (briefly) disrupted the operations of semiconductor manufacturing facilities. Most are in Asia: Taiwan, South Korea and China (Wuhan is home to one of the country’s top chipmakers, Yangtze Memory Technologies). While the region was praised for its initial handling of the pandemic, the inability of Seoul and Taipei to secure enough vaccines resulted in outbreaks at chip manufacturing facilities and further slowed production.


A kind of chips-for-vaccines diplomacy followed, with Taiwan aggressively leveraging its chip-making prowess to procure vaccines from chip-hungry allies. Japan, keen for Taiwanese chipmakers to set up shop there, donated 1.24 million doses of the AstraZeneca vaccine. The US, initially planning to donate 750,000 doses of Moderna’s vaccine to Taiwan, tripled its commitment. In mid-June Taipei gave the Taiwan Semiconductor Manufacturing Company (TSMC), its most important chipmaker, and Foxconn, its other homegrown tech giant, the authority to directly negotiate with Germany’s BioNTech for the purchase of 10 million doses.


The current shortage feels more consequential because it’s the car industry — still the engine of economic growth and source of hope for the post-Covid recovery — that has borne the brunt of shipping and manufacturing delays. For decades, carmakers have worshipped the just-in-time ideology, the true gospel of globalisation. It promised them immense cost savings in exchange for keeping their inventories low. With well functioning global supply chains, fluctuations in demand could be addressed in real time, without forcing the car companies to buy and store excess inputs.


As the pandemic hit, carmakers slashed their sales projections, cancelling or reducing chip orders. But they didn’t anticipate that the global appetite for chips would stay strong or that demand for vehicles would recover relatively quickly. Reluctant to take public transport, customers splashed out on new cars. These might contain anywhere between 1,400 and 3,500 semiconductors, 40% of the car’s total cost.


In normal times, it might have been possible just to ramp up chip production. However, Covid-related factory closures combined with some extraordinary events made this all but impossible: cold weather in Texas (where many US chip manufacturers are based), drought in Taiwan, a fire in a major chip factory in Japan, the blockage of the Suez Canal, Chinese companies’ efforts to stock up on chips before being hit with US sanctions.


Why can’t carmakers make chips?


The auto manufacturers were caught by surprise. Many, even the largest, don’t have direct relationships with chipmakers; they source through car parts suppliers like Bosch and Continental. Moreover, as demand for semiconductors of all kinds increased, manufacturers rationally reallocated production capacity to the far more lucrative computer and smartphone chips.


Why can’t carmakers produce their own chips? Elon Musk is trying to find out: Tesla, which, in addition to mulling over a plan to prepay chipmakers for the semiconductors it needs — another nail in the coffin of the just-in-time dogma — has considered buying a chip factory. Volkswagen will soon be designing chips for autonomous cars.


However, designing is one thing, manufacturing is another. Europe is not happy. ‘If a big bloc like the EU is not in a position to produce microchips, I don’t feel comfortable about that,’ Angela Merkel said in May. ‘If you are a car nation, it is not really good if you cannot produce the main component’. Merkel’s logic is solid: Europe has gone from having 44% of global manufacturing capacity in 1990 to just 10% today. However, this overdue soul-searching is unlikely to yield results without a much broader examination of the key assumptions (about globalisation, trade, national security and industrial strategy) that have shaped US and EU semiconductor policy for decades.


Chip manufacturing is a complex process, requiring a number of distinct operations that can span months. These operations can include more than a thousand steps, as sand — the source of silicon, still the most common material in semiconductors — is transformed into integrated circuits of almost infinite complexity.


For all the complexity, the underlying principle remains simple. Today’s chips are made up of millions or even billions of transistors; the more transistors fit on a chip, the greater its power. Transistors allow the control of the flow of the electric circuit, on or off. This binary language of 0s and 1s underpins modern computing, mediating between electricity and information.


Let’s do ‘more with more’


Like every competitive industry, chip-making is subject to the pressure of having to do more with less. For most chips, at least those of the logic variety, this means maximising their computing power while minimising the associated financial and energy costs. This tendency is known as Moore’s Law (after Intel’s co-founder Gordon Moore): should past trends continue, the number of transistors on a chip will double every year, with a corresponding reduction in the chip’s cost and an increase in its computing power.


Ironically, the ‘more with less’ benefits associated with Moore’s Law have pushed the semiconductor industry to do ‘more with more’. As its leading firms push to the limits of physics, they are forced to invest ever greater amounts in equipment. TSMC’s projected capital expenditure for 2021-24 is $100bn; Samsung aims to spend $151bn by 2030. Other giants are investing similar amounts. What’s needed is not just dollars but brains: according to one study, keeping up with Moore’s Law today requires 18 times more researchers than in the early 1970s.


Chips are usually grouped into ‘technology nodes’, according to how finely they are etched. A bit like generations in a family, different nodes usually involve different architectures and manufacturing processes. Generally, the smaller the node, the smaller, faster and more energy efficient the transistors. The chips in the latest smartphones and tablets are of the 5nm (nanometre) node. In 3nm node chips — not in mass production until 2022 at the earliest — transistors will be just 1/20,000th of the breadth of a human hair. Such breakthroughs, however, are of relevance for only some electronics. Save for chips used for AI and infotainment, cars run on semiconductors of much older technology nodes.


In the past, a single company — known as an integrated device manufacturer — would typically orchestrate the whole process, from designing the chip to manufacturing, testing, and packaging it. This was the story of Intel, Texas Instruments, IBM, and many others.


All that began to change in the late 1980s, when Morris Chang, a Chinese-born and US-educated engineer with decades of experience at Texas Instruments, the US-based semiconductor giant, founded TSMC in Taiwan. Chang saw that chip-making was becoming so capital intensive that another model was needed. He envisioned chip-making as a service, whereby TSMC would offer superb fabrication facilities to allow chip-companies to get rid of their own factories and concentrate on design.


Chang’s big break


Chang’s big break came in the early 2010s, when Apple gave the company the contract for iPhone chips. TSMC has always been a tight-run ship, with almost paranoid insistence on workplace secrecy and a Stakhanovite work culture; at one point, the company’s R&D division worked on a 24-hour schedule, in three shifts.


Today, TSMC’s market capitalisation of more than $600bn — 2.5 times that of Intel — puts it among the dozen most valuable companies in the world. Technologically, it’s several years ahead of its closest competitors. Its latest plant, due to enter service next year, has cost $20bn and will feature a clean room — crucial for semiconductor production — the size of 22 football pitches.


It is partly thanks to TSMC that chips are no longer viewed as take-it-or-leave-it, all-purpose components. The needs of technology giants like Alphabet and Amazon are so specific — and their resources are so vast — that they can afford to spec out and design their own chips. In just a few years, the carmakers are likely to follow the same course for the more advanced AI chips.


The shift to bespoke design has also meant business models built around intellectual property. UK-based Arm Holdings — owned by Softbank but currently the target of a controversial $40bn takeover bid by US chip giant Nvidia — is a salient example. Arm holds an impressive array of intellectual property rights. The company devises abstract solutions that, once implemented, make it possible to improve chip architecture. Clients pay Arm a license fee and royalties in return for instructions on how these abstract rules are to be executed in each specific use context.


From the 1950s onwards, the US was the undisputed leader in the field. Vast research funding and the Pentagon’s blessing ensured hegemony for American companies. This began to change in the 1970s, when Japanese firms challenged US leadership, especially in memory chips and sensors. Japan’s chipmakers embarked on bold acquisition bids in the US, while keeping their own domestic market closed to foreign entrants.


This didn’t go well with the Reagan administration, which eventually used its trade and geopolitical might to neutralise Japanese competitors. The US government also pushed for closer ties between industry and academia. The rise of the no-factory model in fact proved beneficial to the US, as it left Japanese chipmakers with idle (and relatively expensive) production facilities, while allowing American chip giants to refocus their efforts on design. Japan has never quite regained its chip-making prowess; its share of global semiconductor sales has dwindled from 50% in 1988 to 10% today.


America’s pyrrhic victory?


Was America’s a pyrrhic victory, given that its own share of global manufacturing capacity fell from 37% in 1990 to 12% today? It certainly doesn’t feel that way: the American semiconductor industry — represented by Nvidia, AMD, Broadcom, Qualcomm, and even the struggling Intel — has played the globalisation game as it was meant to be played: low-margin manufacturing was outsourced to Asia, while high-margin design and other intellectual property-related activities stayed in the US.


One country that has kept a close eye on the rise of TSMC is China. For most of the 1990s, the hands of its tech firms, many of them close to the military, were tied by the Wassenaar Arrangement, a successor to the cold war-era multilateral arms control regime that controls transfers of conventional weapons and dual-use goods and technologies, and severely restricted China’s room for manoeuvre in semiconductors.


China needed national champions that would look like respectable independent concerns, not just puppets of the regime. It found such a champion in Semiconductor Manufacturing International Corporation (SMIC), founded in 2000 by Richard Chang. Like Morris Chang, Richard Chang (no relation) spent years at Texas Instruments, then worked for Morris Chang at TSMC. In the late 1990s Richard Chang left TSMC to found a competitor called Worldwide Semiconductor Manufacturing Corp (WSMC), though he eventually lost control of the firm and it was sold.


Angered, Richard Chang left for Shanghai, taking 100 engineers with him, to set up SMIC. A deeply religious man, he was not an obvious ally for the Communist Party. Yet he had no problem raising money from the likes of Goldman Sachs, who became one of the leading early investors in SMIC. Neither the Taiwanese government nor TMSC were happy about Chang’s departure; years of litigation forced him to leave SMIC in 2009.


Chinese investments help SMIC


SMIC pressed on, attracting vast investments from various Chinese state agencies and funds (the most recent, $2.2bn, was in May 2020). For all this funding, it still trails behind TSMC and Samsung; for now, it confines itself to chips in the 14nm node. Its efforts to move up the node ladder have been hobbled by US sanctions, preventing it from getting extreme ultraviolet lithography machines from their only supplier, the Dutch company ASML. However, SMIC claims to have found a way around this; its own in-house innovations might help it to jump directly to making the equivalents of 7nm node chips.


SMIC’s success is due to the laser-focus efforts of the Chinese authorities to develop a domestic chip industry. And these efforts seem to be paying off, in part: China has more factories under construction than any other country in the world. There have been more than 1,000 official plans that promised some form of support to the semiconductor cause. The most important of these is the 2014 National Integrated Circuit Plan, which established a $150bn fund to support the domestic chip industry, facilitate foreign acquisitions, and secure critical components from abroad; in 2019 it was extended with another $28.9bn.


Then, there’s the broader commitment by President Xi Jinping to spend up to $1.4tr over the next six years to secure China’s lead in strategic technologies. Liu He, the Harvard-educated vice-premier, has been appointed as China’s chip tsar and tasked with overseeing the development of cutting-edge chip technologies.


Beijing has plenty of measures for stimulating its chip industry. They range from forcing foreign tech companies to form joint ventures and share their intellectual property with Chinese firms to pushing China’s existing tech giants to source more of their chips from the fledgling makers or lose all government subsidies.


Hundreds, or thousands, go bust


Why is Beijing paying so much attention to semiconductors? As long as China aspires to remain the world’s factory, it needs to have enough chips to power all the electronics it will be manufacturing. For now, it still does not: in 2020 alone, it imported $350bn worth of chips — spending more on these than on oil. Since 2005 it holds the dubious title of being the world’s largest importer of semiconductors, underlining the immense gap between its production and consumption.


The world of semiconductors is a cruel one. For every SMIC, there are hundreds, if not thousands of firms, that go bust. According to an estimate from People’s Daily, the period between January and October of last year saw the birth of more than 58,000 chip companies — roughly 200 a day.


Nvidia’s attempt to take over Arm Holdings has caused some alarm in Beijing. If Arm becomes part of a US company, Washington might pressure it not to license its intellectual property to Chinese companies. In the short term, Beijing might simply block the Nvidia-Arms merger, as it has done in the past. In the long term, however, China, India, and Russia are pinning their hopes on RISC-V, an open-source alternative to Arm’s RISC technology. This began as an open-source project at the University of California, Berkeley, but has since grown into an impressive non-profit association, RISC-V International. In November 2019 it moved to Switzerland. This was done in order to avoid any problems with US trade regulations, given the heavy presence of Chinese companies in its ranks (more than two dozen of them have joined). ZTE, Huawei, and Alibaba are also busy experimenting with RISC-V technologies.


Perhaps, the most entertaining aspect of the chip crisis has been watching American politicians question the policy consensus of the last few decades. Speaking to the hawkish Atlantic Council at the end of June, Joe Biden’s top economic advisor Brian Deese complained of a ‘policy-induced coma’ and insisted that ‘strategic public investment to shelter and grow champion industries is a reality of the 21st-century economy’.


US policy on microprocessors has been shaped by the twin imperatives of creating jobs and preventing the rise of China. Since Biden has promised to bring manufacturing jobs back to America, few politicians could object to giving the semiconductor industry some priority; after all, such jobs pay twice the wage of an average US manufacturing job.


The US Innovation and Competition Act, passed by the Senate in early June, sets aside $52bn to restore America’s chip manufacturing to its former glory. Some of this money could be used to entice TSMC and Samsung to go ahead with plans to set up advanced chip plants in the US. But although $52bn might seem like a lot of money — and, in most industries, it is — it pales in comparison to the $450bn South Korea plans to spend over the next decades. And the long-term benefits of reshoring factories are not obvious: one study estimates that the cost of operating a new chip-making facility in the US over ten years will be 30% higher than in Taiwan or South Korea and 50% higher than in China.


Can Europe make globalisation work?


The Biden administration has continued Trump’s tough line on China, refining some of its initial measures. For example, an executive order signed by Biden in early June prohibits Americans from investing in 59 of Chinese companies with alleged ties to its military, including Huawei and SMIC.


Where is Europe in all of this? The current attitude of European policymakers resembles that of their peers in America: panic. In May, Thierry Breton, the French EU commissioner in charge of digital policy, promised to restore Europe’s lost market share in chip manufacturing, securing at least 20% of the world supply of chips by 2030. The EU, he said, has been ‘too naïve, too open’.


This is a polite way of saying that Europe has not been as successful as the US in making globalisation work for its own agenda. When it comes to factoryless semiconductor companies, Europe holds just 3% of the market. The only European company in the top 50 factoryless chipmakers is Norway’s Nordic Semiconductor; the only other recent entry on the list — UK-based Dialog Semiconductor — was sold to Japan’s Renesas Electronics in February.


Europe’s better-known chipmakers — NXP (Netherlands), Infineon and Bosch Semiconductors (both Germany), STMicroelectronics (France/Italy) — have retained some basic manufacturing capacity but they also rely on the likes of TSMC. They cater to a very particular customer base — mostly in the industrial and automotive sectors — and specialise in sensors and power and radio frequency semiconductors. The latter, unlike logic chips, are not subject to Moore’s Law, and ‘node shrinkage’ is, consequently, of less importance.


Many of Europe’s chip companies are actually doing well, as demand from the automotive sector is as strong as ever. Many of their chips, however, are far from cutting-edge, even if they are good enough for the European car industry. Since Europe gave up its ambitions to compete with Apple and Samsung in smartphones and tablets, there is no assured European demand for advanced logic chips. And, without European demand, it’s not obvious why new factories should be sited in Europe, with its high labour costs, rather than Asia; American companies certainly wouldn’t rush to get their chips made in Dresden instead of Taipei.


Predictably, none of the big European chipmakers rallied to Breton’s call to pour billions into ensuring that Europe can manufacture 2nm and 5nm chips by 2030. Intel — which, together with TSMC and Samsung has been approached for help — has offered its services, as long as each factory receives at least €4bn in government subsidies.


Breton is convinced that, even if Europe currently has no market for 2nm chips, it’s his job to create technologies that will make them possible. This seems like magical thinking. Europe’s chip dependency is a symptom of a much deeper malaise that cannot be cured by injecting funds. Having outsourced its defence strategy to the Pentagon and its industrial strategy to the carmakers, Europe has lost the ability to think strategically about how to source its electronics. Nor does it know why this is something worth thinking about.


The technology stack powering the European economies was supposed to be impervious to geopolitics; entrusting its construction to the market seemed to be the way to go. It was, in the end, an extremely foolish bet. The ‘Airbus for chips’ that so excites European technocrats will surely fly under Chinese national colours.



- Evgeny Morozov is the founder and publisher of The Syllabus, a knowledge curation initiative. He is the author of several books on technology and politics.


Copyright ©2021 Le Monde diplomatique — used by permission of Agence Global
Subscrever America Latina en Movimiento - RSS