The price of independence

Few got a ringside seat to the immense financial risks that surround chipmaking from its beginning as good as Arthur Rock's. Instrumental in obtaining work for the team of engineering managers who "betrayed" transistor pioneer Williams Shockley in the late 1950s, Rock again helped some of them set up Intel a decade later and helped develop what sometimes gets called Moore's Second Law.

The corollary of the more famous law, Rock observed that the cost of fitting out a semiconductor fab doubles every four years. He identified in the late-1990s that the bill had risen to $12m from just $12,000 in the late 1960s. Today's, it's well north of $12bn.

Even a single lithographic scanner, used to define circuit images one layer on one wafer every 30 seconds or so, costs in the order of $200m and that will increase in just a couple of years to more than $350m. You need ten of these, plus a bunch of scanners based on older but cheaper technology, to sustain the 150,000 wafers per month that can justify the construction of a leading-edge fab, such as those operated by TSMC in its home on Taiwan. Even the deepest pockets would have trouble buying into chipmaking, especially if they want to build up an infrastructure in a new territory. That was the original plan behind funding for GlobalFoundries from the United Arab Emirates: build up a foundry business to the point where it could bootstrap a new chipmaking cluster.

Given the enormous thirst for power that the not-quite-X-ray lasers have in extreme ultraviolet (EUV) scanners, abundant solar power in the Gulf would make sense. But with little to no semiconductor engineering infrastructure and a chip market that did not look all that healthy, the plan for a Middle Eastern gigafab to compete with TSMC and others went on the backburner.

Europe has lagged behind for several decades but still has a decent advantage in that it has several research institutes focused on chipmaking, two of which have made major contributions to the most important digital semiconductor processes we now have. First, Imec for mainstream CMOS, working closely with the only EUV lithography supplier in the world just 100km over the Belgium-Netherlands border. And second, CEA-Leti in Grenoble, France, with the silicon-on-insulator process developed with STMicroelectronics and later with GlobalFoundries. Now that the chip market has experienced surging demand, it makes a lot of sense to build a new fab for FD-SOI in the area: and this is what is happening, helped along with subsidies from the European Chips Act. And those subsidies are substantial.

They are in the order of €1m per job created in the case of the €5bn Infineon Technologies fab that just started construction in Dresden. Though not a leading-edge fab like one of TSMC's in Taiwan or the one under construction in Arizona, the plant is far from cheap. But as it will make chips for power supplies and renewables, it fits into an existing infrastructure, acting as a sister plant to one Infineon already operates in Austria. The aid for Intel's more advanced plant in Magdeburg, just west of Berlin, could wind up being more than half the cost of building and initial equipment installation if the US chip giant gets its way. The ratio of subsidies to construction cost will likely be significantly higher than those provided to Intel's two new fabs in Ohio even with the large amounts of cash available under the US CHIPS Act and $2bn allocated by the state-level government. That state money seems to have tipped the balance in favour of a location that is thousands of kilometres from Intel's existing fabs that sit on the other side of the Rockies.

Though geographically, the UK could benefit from the relative proximity of a bunch of fabs in Ireland and mainland Europe. But, having separated politically from the EU, the country has in effect split off from the economic infrastructure that if you were counting the pennies leads to a cheaper outcome: hardly anyone believes the UK needs to stump up billions in cash to attract a leading-edge fab. But it's had some pound-foolish knock-on effects that continues to chip away at the nation's competitiveness in some areas where the UK could build a strong position.

Compared to the amounts allocated to semiconductors in the US and EU chips acts, the UK's £1bn looks tiny. We can see how Rock's maxim took effect. If you adjust for inflation, the money spent by government on Inmos and what was an advanced fab more than 40 years ago actually comes to about half of the UK's expected investment. But that money has to go a long way.

The UK's National Semiconductor Strategy has three wide-ranging aims: grow the domestic sector; mitigate the risk of the kind of supply-chain disruptions seen in the post-Covid crunch; and protect national security. These aims have much in common with the US CHIPS Act. Part of the rationale for supporting TSMC's Arizona fab, which had already started construction by the time the act passed, was to make devices locally for defence programmes so they would not have to be obtained from potentially vulnerable far Eastern fabs.

When it comes to the national-security objective, the National Semiconductor Strategy does not have a lot that it can do in practice other than tell suppliers "be careful out there". However, the UK is not exactly on its own here. The country's defence and national-security infrastructure will have to rely on the US and Europe to supply devices for defence if the far Eastern options get closed off. However, though they would be significantly more self-sufficient, without major ongoing changes to how it sources components the US defence industry would still find it difficult, if not impossible, to close the kind of supply gap that would result from Taiwan's production being shut down in the event of a Chinese invasion of the island. For everyone else, a peaceful future continues to rely on a predominantly globalised industry and the supply-chain headaches that implies.

That leaves one objective where the government might be able to exercise some practical influence: grow the domestic sector. That too is a tall order but far from impossible given the country's existing strengths though this is where we see the how prior government actions have hobbled potentially lucrative options. The report spends a long time on the prospects for the compound-semiconductor industry but government decisions taken over the years have done more to weaken the country's position than strengthen it.

Take one element of the strategy: it puts an emphasis on easy and low-cost access to prototyping and low-volume manufacturing in sectors such as photonics and compound semiconductors of the kind found in mobile telecom systems and the power converters of electric vehicles. But this has been planned – on and off – for close to a decade.

The Compound Semiconductor Centre (CSC) that opened close to Newport Wafer Fab (NWF) in south Wales at the end of 2015 was originally intended to play a key role in an EU-wide network of pilot plants for Key Enabling Technologies (KET) as part of its the Horizon 2020 research programme, a programme that the UK found itself negotiating its way out of just a few years later. These centres were established to try to deal with the "chasm" that faces hardware startups as they move from exploratory experiments to trying to make prototypes.

Located close to specialist wafer supplier IQE in Cardiff, the centre then became part of the compound-semiconductor cluster CS Connected. This cluster for a while included NWF, which in turn was expected to act as a foundry to make high-voltage gallium nitride devices until Chinese-owned Nexperia bought the facility and shut down external activities including sales of devices to photonics specialist Rockley, which was forced to look overseas for production.

Now, the prospect of a local prototyping plant is back on the table and being considered by a consortium led by the Institute of Manufacturing in Cambridge to work out how to spend some of the money. However, though such a fab would be far from the leading edge, IQE last year estimated it would cost £1bn alone to fit out a dedicated compound-semiconductor fab for handling prototypes and regular manufacturing. A similar decision faces the consortium on multichip packaging, seen as another important element in the future of semiconductors, especially chips that combine photonics, electronics and other technologies that are likely to be important to the UK's industry.

Had the UK made better decisions, this would not be such a dilemma. Just this week, the EU launched the PhotonixFAB initiative as part of a successor to the KET programme. This brings together a bunch of suppliers that includes German foundry X-Fab to build a network of suppliers who can support prototyping through to full manufacture. But the UK government decided it would make cooperation like this and the Horizon Europe R&D programme part of its "bargaining chip" selection in the Brexit negotiations. And it found itself locked out.

You can see the effect in compound semiconductors as well. Infineon and others have pressed ahead with initiatives such as All4GAN, a project to develop more efficient power semiconductors based on similar technologies to those provided by IQE and the companies around CSConnected. Some 45 companies and research institutes are taking part in the €60m, three-year project. None of them from the UK.

Cooperation does continue at some levels. A photonics-focused joint project between the Universities of Glasgow and Southampton takes part in the Europractice network of chipmaking facilities, which offers low-cost chipmaking services to academic researchers and spinouts from those institutions. But as with defence, the UK is gradually learning that it remains a globalised industry, though perhaps a bit less globalised than before and that alliances are important. Whether the government is learning is another matter.