It is not just cellular infrastructure that is suffering from constriction in the supply chain but also the semiconductor industry underpinning it, especially among the foundries where the integrated circuits are manufactured. This has reared its head recently in the light of global shortages among both leading edge and standard chips that are now afflicting a wide range of industrial sectors, including telecoms.
The crisis has been acknowledged by leading ICT manufacturers including IBM and Dell, as well as by Intel, which is still by some distance the world’s biggest dedicated semiconductor company by revenue at around $77bn a year. Taiwan’s TSMC, the world’s largest foundry, echoed these sentiments, with the consensus being that supplies would not catch up with demand until well into 2023.
That is about the timescale needed to increase capacity at the foundry level, as TSMC CEO Chung Wei recently indicated. “We see the demand continue to be high. In 2023, I hope we can offer more capacity to support our customers. At that time, we’ll start to see the supply chain tightness release a little bit,” he said.
IBM president Jim Whitehurst added that in the meantime, his company, as well as others, would have to consider how to extend the life of certain computing technologies, while accelerating investment in the fabs. Intel, as the only major fully integrated semiconductor company involved in fabless design as well as manufacture, admittedly outsourcing some fabrication to TSMC, is in the spotlight of US efforts to ease the supply chain. “We do believe we have the ability to help,” said Intel CEO Pat Gelsinger recently. “However, it just takes a couple of years to build capacity.”
Clearly these companies failed to anticipate current demand or else they would have planned these capacity increases two or three years ago. It is true they could hardly have predicted the Covid-19 pandemic which boosted demand for some systems consuming chips, but to a large extent the crisis is a consequence of the trajectory the semiconductor industry has taken over the past decade or more.
The pandemic was in fact responsible for the first indications of chip shortage, but that was almost a symptom rather than a root cause of the problem. That shortage first showed up towards the end of 2020 in the automotive industry, as a number of manufacturers reported they would not be able to deliver as many cars as they had hoped during 2021 on account of a deepening chip shortage.
The crisis partly reflects the auto industry’s use of lean manufacturing that keeps stocks light and relies on just in time delivery from suppliers of components, including chips, to minimize inventory costs. At the same time, chips had become crucial components for vehicles that were fast becoming essentially computers on wheels with proliferating monitoring, control, autonomous and infotainment features.
Then early in the pandemic, demand for new cars slumped temporarily, causing auto makers to slash orders for chips. Meanwhile, demand for consumer electronics products that also require a lot of silicon, such as PCs and games consoles, was booming and so they took up the slack. Then by the time demand for cars picked up later in the year, the required chip production capacity had already been diverted. The upshot was serious, with Germany’s Audi for example reducing production of vehicles by around 10,000 a quarter and furloughing 10,000 staff.
This turned out though not to be just a local difficulty for the auto industry resulting from poor forward planning, but spilled out across other sectors including consumer electronics and telecoms. The impact on 5G roll-outs so far has been mixed, with T-Mobile warning that its 5G fixed wireless access service would be “limited by global supply chain constraints at launch”. The,n also in the USA, Gogo delayed launch of its 5G service aimed at aviation until 2022, citing the chip shortages.
Verizon, on the other hand, insisted that its supply chain was robust enough to cater for the shortages without any delays in deployment. This appeared to endorse the company’s preparations for the upgrade of 8,000 cell sites to operate in midband spectrum. The operator had spent just over $50bn on C-band spectrum licences across the USA and determined it was worth shoring up the supply chain to ensure it started recouping that outlay as soon as possible.
One consequence of this shortage – coupled with the geopolitical schism opening up over semiconductors between the USA and China, centering on TSMC in Taiwan – looks like being a reversal of the historical consolidation among foundries. This dates back ultimately to the 1980s when the semiconductor industry was in a state of early maturation and the overall process of designing, fabricating and testing chips split into distinct operations, even if some players, notably Intel but also at that time others, remained fully integrated.
In this process, described as “disarticulation of the value chain”, some became foundries, concentrating on manufacture, while others focused on design. Yet others specialized in packaging, that sector being called outsourced assembly and test (OSAT), while a fourth key sector emerged in developing and making the dedicated assembly equipment used by the foundries.
Over the following years, the fabless model gained the most traction because it allowed greater flexibility and demanded less heavy investment, being relatively asset-light. While a fab plant can cost $10bn or more and require regular replacement or at least substantial refitting, the fabless group, in which Qualcomm has just overtaken Broadcom as the largest, came to be seen as the font of innovation and agility.
The majority of previously integrated chip companies became fabless, with AMD parting company from arch-rival Intel by doing so in 2009. Nvidia of GPU fame was one of the most successful to grow up purely as a fabless chip company after its launch in 1993.
Then ARM was notable for great success by taking the fabless concept one stage further back by not producing any chips of its own at all. Instead, it develops the intellectual property specifying the components of an integrated circuit such as the instruction set design, graphics and interconnects. Its customers, which can be foundries or more often fabless companies, then incorporate these designs in silicon.
The upshot was that by the end of this millennium’s first decade, the global foundry industry had consolidated into just three dedicated major players – TSMC, UMC and Global Foundries (the latter owning the remnants of AMD’s former foundry). Of these, TSMC is dominant today, almost 10 times larger by 2020 revenue of $47.95bn than either of the other two. Samsung is more of a match, with revenues around $57bn across its whole semiconductor business including foundry, but this is set up largely to supply its own consumer and enterprise devices. Then there is Intel, which entered the foundry business a few years ago, though only in a small way so far.
The dominance of TSMC in particular, coupled with the potential military threat posed by China to Taiwan, has concentrated minds in the business and led to growing momentum behind building foundry capability in the USA. Suggestions that TSMC’s output could be diverted to China in the event of a military conquest are somewhat wide of the mark, because TSMC itself is highly dependent on equipment and IP coming from the five leading semiconductor equipment makers, Applied Materials, ASML, Lam Research, Tokyo Electron and KLA. Of these three are American, while ASML is Dutch and Tokyo Electron Japanese. In the event of a Chinese takeover, all of these would probably cut off TSMC and it would take some years to catch up. TSMC, effectively given an ultimatum last year by the US government, agreed to break ties with Huawei rather than lose US customers such as Apple, its largest, and Qualcomm.
However, the Chinese threat has accelerated moves to build up advanced foundry capability elsewhere, primarily in the USA. TSMC itself announced early in 2020 that it was building a $12bn fab in Arizona, citing the need to be closer to its major US customers but also for geographical diversification and some hedge against a Chinese invasion of Taiwan. Government support for this build-out was one of the incentives to turn against Huawei, previously its second largest customer.
Samsung Foundry, meanwhile, is planning to build a leading-edge semiconductor manufacturing facility in the USA, near Austin, Texas, expected to cost over $17bn, and scheduled to start production late in 2023. While Samsung has not indicated what process it will be using there are rumors it will go for the most advanced current geometry, 3-nanometer. TSMC has also indicated it will go beyond the current relatively seasoned 5nm process to 3nm at its impending Arizona plant. The first products using 3nm process are expected during 2022, and IBM has already demonstrated 2nm technology.
Intel is also aware of the foundry bottleneck and Gelsinger has hinted at expanding this aspect of its business as a priority. Meanwhile, the Biden administration has also outlined its concern and given its more interventionist strategy could well intervene with investment to propel Intel into the forefront of the foundry game from its tiny position now (its largest foundry customer was previously FPGA maker Altera, but Intel then acquired that company).
One, at least temporary, impediment to this gathering foundry goldrush in the USA could be shortage of top personnel. TSMC has hired 250 engineers for its Arizona facility but has signaled its concern over the quality of relevant graduates coming out of American universities and colleges.
“In the United States, the level of professional dedication is no match to that in Taiwan, at least for engineers,” said TSMC founder and chairman Morris Chang recently. This coupled with higher pay in the US would make foundries there less competitive than in Taiwan for the foreseeable future, Chang added.
One thing that is certain is that the historical consolidation in foundries towards one company and one country is going to be reversed, at least to some extent. The European Union also has ambitions to become host to a foundry and has been in discussions at least with Samsung and TSMC.