When the chips are down, Intel's biggest gamble isn't what to do – it's whom to do it with
Trade you architecture and production tricks for lithography and yield plans?
Column Political America likes to judge its presidents by their first 100 days. Corporate America thinks more in 90-day cycles, so as today is Pat Gelsinger's 90th day at the helm of Intel, it's an apt time to look at how he and the company are doing.
The market remains cool on Gelsinger, perhaps because he said that the $80bn of Intel cash spent on share buybacks this past decade may have been better invested in, y'know, making chips.
He has set about proving that by announcing $20bn for a couple of new fabs in Arizona last month, and he's up for more. A quick tour of Europe saw him ask for $10bn in subsidies to build new plants somewhere in the EU, expressing strong interest in Germany and Benelux (sorry, UK, you don't get to play). He also broke bread with BMW and had a neither-confirmed-nor-denied visit to Volkswagen.
All of this fits Plan Pat, which is to reinvest and reinvigorate manufacturing, regaining leadership of, or at least parity with, the best of the rest, while at the same time pushing into the chip-starved automotive industry with lots of chips on less-advanced nodes, where Intel has a lot of older capabilities.
The other legs of Plan Pat are to make chips for other people as a foundry, and to buy in chiplets from other people to make up multi-die circuits that mix and match different functions with the most appropriate technologies.
- The strife of Brian: Why doomed Intel boss's ex86 may not be the real reason for his hasty exit
- As Samsung breaks ground on new EUV semi foundry, SK Hynix mulls expanding its logic and contract fab businesses
- FCC starts probing effects of semiconductor drought on the US telecoms supply chain
- Intel throws sand in the face of 'musclebooks' with 10nm Tiger Lake tech
- UK's competition watchdog sniffs around AMD's proposed $35bn all-stock buy of Xilinx
- 'It's dead, Jim': Torvalds marks Intel Itanium processors as orphaned in Linux kernel
And there are some contradictions. Gelsinger is promoting the build-up of fab capacity as one of national interest to Western nations, a way of securing the highest technologies against supply line woes and over-reliance on Asia. Which is fair enough – government investment in the US semiconductor industry helped it regain global share after a big slump in the late 1980s, and even with recent mis-steps, Intel is still the only Western silicon company with the scale and experience to look like it can bring in the very best for its sponsors.
But it's hard to say that your strong point is independence when you're also proposing to become dependent on those selfsame Asian suppliers for your chiplets. Which is it to be?
Likewise, becoming a foundry for other companies sounds easy and obvious; it is neither. Intel has tried and foundered before, not least because of its half-century history of almost exclusively designing and making its own chips means it also designed and made its own software tools and internal standards. Migrating to those used by the rest of the world, which has to be done to compete, is going to be long, hard, and risky for a job that's already one of the longest, hardest, and riskiest in the business.
Perhaps the biggest unknown and the most important is how and with whom Intel will form new long-term partnerships. There are many dimensions here – technical, economic, strategic, and political – and at every level there is little clarity.
TSMC already fabs for Intel, and Intel would like more – but TSMC wants a long-term commitment that probably does not align with Gelsinger's plans.
TSMC, meanwhile, is building its own big new next-generation fab in Arizona that may or may not have some Trump-era subsidies locked in.
Biden has announced $50bn for the domestic semiconductor industry, and Intel would like some of that. But so would the rest of the new US industry lobbying group, the Semiconductor Industry Against China Semiconductors in America Coalition, which counts every red, white and blue-blooded American tech firm, including Intel, Google, and Apple, plus Samsung and TSMC among its members. Those board meetings are going to be fun.
But Intel will know, as they all do, that the chances of getting investment depend on presenting convincing partnerships with each other that demonstrate maximum chance of success.
And this brings us back to that thing that Gelsinger keeps going on about – making the darn chips. Intel's partnerships here will be critical, but not obvious.
When IBM unveiled its 2nm chips last fortnight and said "they're working, all we need is a fab," many assumed that existing partner Intel would be that prime pal. That's not guaranteed by any means, not only because IBM might choose someone with a better recent track record like TSMC but because Intel itself may have other ideas.
Nobody buys a nanometre
The IBM process, although it follows the basic ideas of gate-all-around transistors formed from silicon nanotubes and nanosheets in stacks, may be quite different from Intel's own extensive developments along those lines – and it doesn't offer the sort of leap from Intel's 7nm that "2nm" implies. Nanometre node numbers are not comparable between companies as the days when everyone made basically the same transistor geometry are long gone.
Nobody buys a nanometre. Instead, purchasers care about power consumption versus performance versus price, and chip makers care about process complexity, yield, and device packaging. In some important ways like transistor density per square millimetre, Intel's new 10nm process is equal or better than others' 7nm, while the company says it has finally fixed its own 7nm process.
The company still has a lot of crown jewels in architecture and production, and the judicious trading of those for the lithography and yield tricks that others have will be vital.
The next big round of new 5nm or lower nodes are destined to be in place by 2024/5. This means Gelsinger has to make sure Intel has the cash and needs to know where he'll be placing his bets in the next six months or so. That means whom he's going to work with; how to retool his fabs; how to manage the supply chain for incoming as well as outgoing silicon; how to manage the EU, European, and American state and federal agencies – which may give him long-term money that the market may not; and how to have chips that people want to buy – not just at the end of it all, but every day until then.
By all means, enjoy the big reveals about new chips and new technologies that Intel will be making along the way. But it's the deals in place at the end of the year that will seal Gelsinger's place in history. ®