Intel Corp. is still the dominant provider of chips for personal computers and servers, but its reign as the chip king of Silicon Valley has ended, and the unresolved question is whether its woes stem from reaching the limits of technology, the people in charge or a mixture of both.
The chip giant announced a delay of at least six months in releasing new chips designed using its next generation 7-nanometer manufacturing process this summer, and disclosed a change that shocked longtime observers: Intel may work with a contract manufacturer to make some components of the first chip in the next generation, a graphics processor focused on the data center known as Ponte Vecchio. While Intel INTC, +0.57% already works with a foundry for about 20% of its chips, the idea that the largest chip processor would possibly farm out some of the manufacturing for one of its important next-generation chips was perceived by investors and industry analysts as a stunning fall for the standard-bearer of Moore’s Law.
“I think Andy Grove would probably be spinning in his grave,” said Nathan Brookwood, principal analyst of Insight64, referring to the famously paranoid Intel co-founder and longtime chief executive. “He regarded the manufacturing group at Intel as the heart and soul of the company, so to see them flounder, especially with regards to new-process tech, that has never happened before.”
The delay in 7-nanometer process followed a nearly four-year delay on the company’s transition to its current manufacturing process, the 10-nanometer process. Intel has been offering intra-node advancements of its 10-nanometer process, with new technologies and software optimization and other changes to stay competitive with Taiwan Semiconductor Manufacturing Co. Ltd., known as TSMC 2330, -0.88%, the world’s leading foundry for manufacturing semiconductors on a contract basis.
Intel’s prolonged descent allowed rival Advanced Micro Devices Inc. AMD, +0.04% to move ahead of Intel in the manufacturing process with its partner TSMC, a once unimaginable turn of events. At the same time, Nvidia Corp. NVDA, -1.13% has soared ahead of Intel in market value and made a bold $40 billion move to buy ARM Holdings PLC from SoftBank Group Corp. 9984, +0.10% 9984, +0.10% 9984, +0.10% which would propel the graphics chip maker into Intel’s core market for microprocessors.
Three months after the initial news of the delay, Intel has not given many more answers about its struggles, and analysts are still at odds about whether the company’s slip was due to the increased difficulty of fighting the laws of physics, company personnel issues, or some combination of both. The only thing that may be more of an unknown is where Intel goes from here.
Is Moore’s Law dead?
“One of the things...that we all know is that Moore’s Law has come to an end,” Jensen Huang, Nvidia’s co-founder and chief executive, said on a conference call to discuss the Arm deal last month.
Huang was referring to a prediction made in 1965 by Intel co-founder Gordon Moore that the number of transistors on a semiconductor would double every year, which he later revised in the mid-1970s to every two years. The bottom line of this prediction was that computers would become more powerful and less expensive, as part of that huge increase in transistors in the future, and it served as a guiding light for the semiconductor industry.
But now, the doubling of transistors, and the associated increase in computing power, has become more difficult, as engineers bump up against the laws of physics and the geometries of electronic transistors become increasingly minute, and not even visible with the human eye. It was Intel’s ability, like clockwork, to increase the amount of transistors on its chips, in its own manufacturing facilities, with every new generation of chips, that made its both the company and its chips more powerful with each new product line, at a pace of every two to two and a half years.
More from Therese: The inside story of Gordon Moore and the pre-PC days of Silicon Valley
“Moore’s Law is alive and evolving,” an Intel spokeswoman said in an email, adding that even though it is taking longer in between nodes, intra-node advancements such as its SuperFin 3-D technology and specialized architectures for specific workloads, advanced packaging and software optimizations are helping. “We can continue to deliver the benefits of Moore’s Law well into the future,” she said.
If Moore’s Law is dead, someone should tell TSMC, which still seems to be moving along at a steady clip. It is already believed to be providing Apple Inc. AAPL, -1.40% with its custom-designed chips for the iPhone 12, which are based on the latest 5-nanometer process. At its iPhone 12 launch this month, Apple touted many new features in its four new models as based on its A14 bionic chip, calling it the first 5-nanometer based chip.The new iPhones will be available next month.
Some analysts believe that as investors pose the question about its manufacturing future, inside Intel there is a debate inside about whether or not it should go completely fabless and farm out its fabrication needs to TSMC to keep up with competitors and the fabled tech path its founder laid out.
“I believe that argument is going on inside the company, but I don’t have a sense if it’s going one way or the other,” said Rob Enderle, principal analyst at the Enderle Group. “But if you miss a stepping stone like this, it does force you to rethink how you are doing things.”
Some of the issues may have begun in 2011, when Intel introduced a new technology into the chip manufacturing process, to create three dimensional transistors, as a way to continue Moore’s Law, and maintain its two-year cadence of introducing new technology generations, for the 22-nanometer lithography process. But this technology change may have been trickier to scale further down the line, because a 3-D stack of transistors is taller and more fragile. When Intel moved to 14 nanometers, things started to slow. Chips based on the 14-nanometer process took three years instead of two. The company also began to introduce other innovations in new materials, such as cobalt.
With the 10-nanometer process, things slowed even further. Intel launched a chip called Cannon Lake based on the 10-nanometer process in 2018, after it was originally planned for 2016. But the first iteration was disappointing and short-lived. The big volume product is expected with its new Tiger Lake processor for PCs due at the end of this year. Intel officially unveiled Tiger Lake at a virtual product launch in September. The 11th generation chip will also include an integrated graphics processor and Intel said the chip is four times faster than the competition on common tasks such as photo uploading and 20% faster on productivity applications like Word and PowerPoint.
“Intel may have put too many things into the 14-nanometer and 10-nanometer process increments,” said Brookwood of Insight 64. “When you have that many variables, and something is not right, fixing it is a non-trivial problem.”
Despite these issues, Intel is still plowing billions into its fabs. This October marked 40 years of manufacturing operations in Ocotillo, Ariz., and it celebrated that anniversary with a multibillion expansion of its fab there, culminating in a $7 billion investment. The fab created 3,000 Intel jobs. Intel said it has invested about $23 billion in its factories in Arizona, and it is also expanding its facilities in Oregon, Ireland and Israel and investing in New Mexico, for an advanced memory facility.
“My team and I are committed to doing what is necessary to deliver a predictable cadence of leadership products for our customers,” wrote Keyvan Esfarjani, who in July was named corporate vice president and general manager, manufacturing and operations in a blog post in mid September. Esfarjani replaced Ann Kelleher, an Intel senior vice president, as leading manufacturing, while Kelleher was promoted to leading technology development, focusing on the development teams working on the 7-nanometer and 5-nanometer processes, in a management shake-up that claimed Murthy Renduchintala, just a few days after its July bombshell.
But while Intel is talking about innovations and tweaks to keep Moore’s Law alive, the bottom line is that it’s getting increasingly difficult.
“The challenge is that as Intel is maturing 10 [nanometer process], you have TSMC going to 5 [nanometer process], where they have lost the density game,” said Patrick Moorhead, principal analyst at Moor Insights and Strategy. “How they got there is up for debate, but putting all the technical gobbledy gook aside, they took some big risks on 10-nanometer to get even denser, and it didn’t pan out.”
Tumult in the executive suite
If Moore’s Law is still alive, could the problem be inside Intel, or possibly with some of those executives who have now left the company? The man overseeing the new structure, CEO Bob Swan, could be the one to push toward a solution that would be outside the company’s comfort zone, since he is not an Intel old-timer, and coming in as Intel’s former CFO, not as an engineer.
“Intel giving up the fabs is a bridge too far for the older boys,” Enderle said. “If they have a CEO who could make that decision, Bob is one who could make that better than anyone else because he is not an old Intel guy.”
Swan came to the company in 2016 as chief financial officer, joining from General Atlantic, a growth equity investor after serving as the CFO of eBay Inc. EBAY, +0.77% for nine years. He was named CEO of Intel in early 2019 after serving as interim CEO when Brian Krzanich stepped down after the board learned he had a relationship with an Intel employee against company policy.
It was under Krzanich, who moved up from the manufacturing side of the company, that Intel began to start missing some of its manufacturing milestones. Krzanich, who became CEO in May, 2013, undertook Intel’s biggest layoffs in a decade, cutting 12,000 jobs in April 2016, 11% of its workforce, as it refocused its business around data-centric and cloud computing.
Leadership issues may have begun even before the big layoffs and Krzanich’s abrupt resignation, however.
“For the longest time, Intel’s big advantage was the ability to manage an R&D team,” said Dan Hutcheson, president of VLSI Research. “They were the best at that, maybe IBM IBM, +0.83% too, but there was nothing in Asia that came close.”
Hutcheson noted that the retirement of manufacturing chief Bill Holt in 2016, and followed three years later by former senior fellow and director of process architecture Mark Bohr in 2019, may have created vacuums in the absence of longtime leadership roles, and instead some may have been focused on the machinations inside Intel, not on customers.
At the time of his retirement, Bohr told The Oregonian, “We kind of overshot, I think, with our 10 nanometer technology,” adding that Intel was “too aggressive” in its goals of packing transistors on its semiconductor wafers.
“We have a strong bench of technical talent at Intel. We’ve continued to invest and innovate to deliver leadership products,” Intel’s spokeswoman said. She pointed out how Intel is advancing its 10-nanometer process with its new Superfin technology, which is contributing nearly a node-sized advance in performance, which will be in the new Tiger Lake family.
The abrupt resignation of Jim Keller, a microprocessor guru earlier this year, who has worked at AMD, Apple and Tesla Inc. TSLA, -2.05% , after just two years at Intel also raised some eyebrows. One of the class-action lawsuits against Intel contends that Keller was pushing for Intel to do more manufacturing with foundries in order to be more competitive, but no source was cited for this statement. Intel at the time said that Keller was resigning due to personal reasons.
A few weeks after Keller’s resignation, Apple announced at its virtual developer conference that it would be breaking up with Intel as its long-term microprocessor partner, after a 15 year relationship. In a two-year transition period, Apple will move the Mac over to custom-designed ARM-based chips.
How does Intel move forward?
Longtime industry analysts could not point to a single event or person that led to Intel’s current situation, but a combination of failed process and inadequate leaders. Daniel Newman, principal analyst with Futurum Research summed it up succinctly. “It was clearly a people and tech problem,” he said.
Whatever the root cause of the issues, Intel is now at a major crossroads. Already, its delays are having ramifications. In late August, the New York Times reported that Intel’s delays with its 10-nanometer technology could delay the $500 million supercomputer, called Aurora, one that counts Intel among its major tech providers. On top of a possible delay, the possibility that Intel may have some parts of the chipset manufactured by TSMC will also dash the hopes of a project with all American products. Intel said that beyond Aurora, it has a long-term deal with the Energy Department to further support the U.S. leadership in advanced computing systems.
Brookwood noted that Intel has gone through major shifts in its strategies in its past, especially when Grove was CEO, but was unsure if Swan could navigate the same treacherous waters as his legendary predecessor.
“A couple of weeks ago, I was thinking about Andy Grove and how he got them out of the memory business, because they just couldn’t compete with the Japanese,” Brookwood said. “I could see how someone with Andy Grove’s vision today could look at Intel and say, ‘you know, the manufacturing part of what we do has become a commodity, and the TSMCs and Samsungs can do it better than us. Maybe it is time for Intel to get out of manufacturing entirely and focus on what we know we do well’....But they need someone to have Andy Grove’s insight and business judgment and I don’t know if Bob Swan is that guy.”
Bernstein Research analyst Stacy Rasgon said it was difficult to envision the impact on Intel if it began contracting out its manufacturing on a larger scale. Rasgon said in an interview that Intel will probably never get rid of all of its plants, but it needs to pursue a sort of dual-pronged strategy, of contracting out more parts.
“They have to pursue both options. If they want to get first products in 2023, they have to place orders at TSMC by end of 2021.” Analysts may ask for more details on the alternative plans during Intel’s third quarter earnings call later this month.
Intel indeed has a massive manufacturing footprint and expenditures that employs thousands around the world. In July, Intel said it would spend $15 billion on capital expenditures this year.
Perhaps the answer will be scaling down of some manufacturing in the future, as Swan hinted at in an op-ed piece for USA Today.
“Forty years from now, how and what Intel manufactures may look quite different,” he wrote. “But what won’t change is our belief in the power of technology to enrich lives and our relentless effort to provide the technology foundation for the world’s innovation.”