It was nearly a decade ago when Intel, then the undisputed leader in global semiconductor manufacturing, made a fateful decision.
A new technology, extreme lithography, was offering a way to pack more computing power on to the silicon wafers from which tiny chips, essential for widely used products like smartphones and PCs, are cut.
Using light to etch complicated integrated circuits, EUV promised an unparalleled degree of miniaturisation, but Intel executives believed it would take years for the method to become practical. Instead, they stuck with older manufacturing techniques for their next generation of chips.
This turned out to be a historic mistake, one with consequences that are being felt at a time when the US has put advanced chipmaking at the centre of its national industrial policy.
Taiwan Semiconductor Manufacturing Company, which adopted EUV in 2019, has leapfrogged Intel to become the world’s most advanced chip manufacturer, closely followed by Samsung. Along with other slips, the judgment call has left Intel – and the US – scrambling to catch up.
“Hindsight is 20/20,” says Ann Kelleher, head of technology development at Intel and the executive charged with restoring the US chipmaker’s manufacturing processes. “It’s very easy to look back and say, ‘If something different was done ...’”
Intel is today at another crucial juncture. If, as planned, the company finally produces chips made with EUV in large volume later this year, it will be an important step on the road back. Nowhere will progress be watched more anxiously than in Washington, where the Biden administration is facing an imminent decision about how much financial backing to throw behind the company.
Last year’s US Chips Act committed $52 billion (€47 billion) in direct subsidies to support semiconductor manufacturing and boost research and development, along with an estimated $24 billion worth of tax credits over the next eight years. The law was designed to reverse a slide that has taken the US share of chip production to 12 per cent, from 37 per cent in 1990.
The centrepiece of that plan is to bring leading-edge manufacturing back to the US. For better or worse, that leaves Washington with little choice but to bet heavily on Intel, despite it being the laggard in one of the tech world’s most important races.
Yet falling behind in advanced chip production is not the only problem hanging over Intel. Big shifts in its customers’ needs – such as the rise of artificial intelligence – are threatening to sideline its traditional PC and server chips. Its attempt to go into direct competition with TSMC by becoming a so-called chip foundry, manufacturing chips on behalf of other companies, represents the biggest change to its business since it abandoned its original memory chips for processors nearly 40 years ago.
To make things even harder, a yawning financial hole has opened up under the company at just the moment it is trying to make up for years of underinvestment with a surge in capital spending. The depth of the reversal, which the company says is caused by a temporary inventory correction, shocked Wall Street in January, when Intel warned its revenue would tumble 40 per cent in the first three months of this year.
The setbacks mean that a central piece of US industrial policy is now riding on one of the most difficult and complex tech turnarounds ever attempted. As the US Department of Commerce begins to weigh how to distribute the Chips Act subsidies, deciding how fiercely to back Intel will be a central question.
“It’s a very hard problem, people are underestimating the degree of difficulty,” says Willy Shih, a professor at Harvard Business School. Even if subsidies help to make up for some of the higher construction costs in the US, Intel’s plants will still face much higher operating costs than its Asian rivals, he adds. At the same time, rival chip subsidies in other countries also mean that even if the US halts the relative decline of its chip manufacturing base, it will struggle to win back global share.
The extent of Intel’s woes has spooked investors. Despite rebounding 22 per cent this year, its stock price has halved in the past two years. In the same period, the Philadelphia semiconductor index, a measure of the broader health of the industry, has only fallen 2 per cent.
“They’re in a tough spot,” says Stacy Rasgon, a chip analyst at Bernstein Research. “The best you can say is that the worst news is already out.”
Intel’s failure to keep its lead in manufacturing technology has been at the centre of its problems. For half a century after its co-founder Gordon Moore famously predicted in 1965 that the number of transistors on a chip would rise exponentially, Intel maintained a roughly two-year advantage over rivals.
After 2014, things started to go awry. The planned “shrink” to chips with features only 10 nanometres wide was thrown off course by the complex manufacturing steps put in place to get round the lack of EUV. Further delays mean its upcoming “node”, using a 7nm process that has been renamed Intel 4, will be roughly five years late – assuming the company succeeds in getting it into production later this year.
In less than a decade, Intel has slipped from being one generation ahead of its rivals in the latest chip technology to being one generation behind. Comparable chips from TSMC, using a process known as 5nm (confusingly, the actual sizes have diverged from the naming systems used to identify them) went into volume production in 2020. As a result, the latest products from rivals AMD and Nvidia – companies that design chips and outsource them to TSMC to be manufactured – have achieved higher levels of performance and eaten into Intel’s market.
According to Kelleher, the Irish-born former head of manufacturing who was put in charge of Intel’s technology development more than two years ago, turning Intel around will require nothing less than a cultural transformation. One of her first steps was to take on the “not invented here” mentality at a company where success had bred an insular approach to engineering.
“Because we were in a leadership position, we weren’t as open to the industry,” she says.“ We don’t need to invent everything going forward ourselves.” In one big break with the past, Kelleher brought Intel’s chip design processes in line with industry standards, enabling it to use the same design automation technology from outside suppliers as other chip producers.
To make Intel less dependent on the kind of risky leaps that hurt its move to 10nm, Kelleher adopted what she calls an “incremental and modular” approach. That means some parts of each chip platform can be reused in later releases, or brought forward and tested alongside current technologies, such as PowerVia. This method of powering a chip from the back of the wafer to free up space for logic circuits on the front is designed for future releases but has been packaged with other components on a trial basis.
Along with PowerVia, Intel is also gambling on its first new transistor architecture since 2011, called “gate all around”, designed to reduce the leakage of electricity as transistors move towards sub-nanometre scales to regain its edge. “Both of those innovations are essential to us getting back to leadership,” says Kelleher.
With other manufacturers also planning to adopt new transistor architecture, this presents an opportunity to shake up industry leadership, says Shih, as companies vie to be first to perfect the technology. However, there is not yet a sign that this will play to Intel’s advantage.
Intel is counting on the changes Kelleher is making to race through five new process “nodes” in only four years, something the company says will enable it to regain a manufacturing lead by 2025. “Overall, we’re doing very well,” says Kelleher after two years in.
With one release completed and four still to come in quick succession, followed by the need to scale up production and embed the new technologies in future generations of products, most of the hard work lies ahead. According to Rasgon at Bernstein, it will take another five years to tell whether Intel can become globally competitive again.
To support its new chip designs, Intel has announced a spate of giant new manufacturing plants, known as fabs, with the economies of scale needed to justify the capital-intensive processes.
There are two fabs planned outside Phoenix, two more in Ohio and a new €17 billion mega plant in Germany that represents the country’s biggest investment since the second world war.
The cost for the first phases of these developments has already reached around $60 billion, and the German government is pushing Intel to expand its plans in exchange for the higher subsidies the company is seeking.
Under the Chips Act, Intel could receive up to $12 billion to support its new US facilities, with extra subsidies for an advanced chip packaging plant in New Mexico and further tax credits.
Throwing so much support behind the company, however, may not be the quickest way for the US to become self-sufficient in chipmaking. In a report two years ago, Georgetown University’s Center for Security and Emerging Technology (CSET) estimated that around 55 per cent of the advanced chips consumed in the US were made in TSMC’s fabs, with a further 25 per cent coming from Intel and the rest from Samsung.
The most effective use of the Chips Act subsidies geared to advanced chipmaking, according to CSET, would be to apportion them based on market shares, in effect leaving Intel with only half the money it is seeking.
However, Intel’s greater willingness to pour money into new US manufacturing has put it in pole position to receive a much bigger share. Last year, TSMC expanded its plans for a new fab under construction in Arizona, but output would be “a drop in the bucket” compared with the giant fabs the company runs in Taiwan, says Shih.
That has left Intel as the US standard-bearer in chip manufacturing “on a de facto basis”, says Chuck Wessner, a senior adviser to the Center for Strategic and International Studies in Washington. “It’s not a government policy,” he adds, but the company’s eagerness to invest significantly in the US makes it the only realistic option.
This message is one Intel itself has been driving home in Washington. The main goals set by the commerce department – to boost domestic chipmaking in order to increase economic and national security – leaves the country with no choice but to back Intel as “America’s champion, which has been investing heavily for decades,” says Al Thompson, the company’s head of US government relations.
Yet even if Intel succeeds in its manufacturing plans, there is no guarantee it will have enough business to fill its giant new fabs, or to operate them profitably.
Sales of PCs – still Intel’s main market – have fallen back after a pandemic-era boom, and many Wall Street analysts believe the company’s predictions about the market in the long term are unrealistic. To make matters worse, Apple recently dropped Intel in its Macs in favour of its own silicon chip designs, while AMD has taken advantage of TSMC’s superior manufacturing to claim an estimated 35 per cent of the PC market.
“Thirty per cent of their [PC] market has vanished,” Rasgon says of Intel, once synonymous with the PC industry. But now, even some of its biggest customers seem ready to move on.
“I think it’s important for Intel to succeed, and they’ve been a great partner,” says Michael Dell, chief executive of Dell Technologies. “But if they don’t succeed, we’ll use something else.” That could include new chip designs not based on the core chip architecture found in Intel’s main products, he adds. “Competition’s a good thing.”
Meanwhile, in servers, Intel processors face a barrage of competition, as cloud computing giants such as Google and Amazon have turned to designing their own chips. The data-intensive work of training AI systems has also boosted demand for different classes of chips. Wall Street’s belief that Nvidia will be the main winner from the AI race has lifted its shares by 90 per cent this year and added $360 billion to its value – or more than two and a half times Intel’s entire market capitalisation.
One response from Intel has been to diversify into new chip architectures to compete. Another has been to open up its manufacturing to other chip companies, in the hopes of bringing in enough outside business to support the huge investments it needs to make.
Potential customers for this new chip foundry business have cautiously welcomed the move. Cristiano Amon, chief executive of mobile chip company Qualcomm, says it will be “a good thing” for the US if Intel succeeds, and that it will give Qualcomm another choice of manufacturer to turn to. Whether or not the Intel plan will work, says Amon, is “a tricky question to answer”, adding: “We haven’t committed a product yet.”
For Intel, competing with TSMC will mean learning a new way of doing business, including persuading customers that Intel will not put its own interests first if capacity is ever in short supply. It will also mean matching a fearsomely efficient rival, despite facing a likely cost disadvantage from being based in the US.
Intel has said that from early next year it will report the results of its manufacturing operations separately – even though, for now, the only customer will be Intel itself. The company’s executives hope the division will instil greater discipline throughout the company, forcing its manufacturing arm to show it can stand on its own feet, while its chip design business has to match the best of the “fabless” companies, such as Nvidia.
One day, they concede, that could even lead to a full break-up of the company – something some investors pressed for when Pat Gelsinger returned to head Intel in 2021. But it is likely to take years for Intel to win over enough big customers willing to bet on its advanced new fabs, making such a break a distant prospect.
For politicians in Washington, where reducing the country’s dependence on foreign-made chips has become an urgent matter, such long timescales will be a challenge.
“It took Intel 10 years to lose its lead, and the US giving up its chip manufacturing has taken 30 years,” says Shih at Harvard Business School. “Don’t expect results by the next election cycle.” – Copyright The Financial Times Limited 2023
