Intel’s success came with making its own chips. Until now


Mint

November 7, 2020

Intel Corp., more than any other company, represents the historical silicon in “Silicon Valley.” It has held its reign not only by designing compelling new circuitry but also by etching it into silicon at its own factories.

It was orthodoxy inside Intel. To thrive, it had to remain the maker of its flagship chips that are the brains of computers—long after many rivals had outsourced manufacturing and stuck to design.

So on July 23, when Chief Executive Bob Swan on an earnings call said Intel would consider outsourcing the manufacture of some of its most advanced chips, it was a milestone in the story of America’s losing its manufacturing primacy.

Intel’s plants had stumbled for the second time in a row in making the circuitry that underlies a new generation of chips called central processing units, or CPUs, each with tinier transistors than the one before. It had needed to scrap much of what its factories had made during trial runs and push back the promised delivery time frame. Intel would now reconsider how to make its chips that hit the market in the year 2023 and later, Mr. Swan said.

The company might end up outsourcing them, might keep making them in-house, or might use a new approach of processing a chip partly itself and farming out certain other processes on the chip, Mr. Swan said. Intel saw the decision, he said, “as a sign of strength, not as a sign of weakness, that gives us much more flexibility to make the decisions on where is the most effective way to build our products.”

The Santa Clara, Calif., company’s concession that it has hit the factory wall—that the circuitry has gotten so small it may not be able to economically build some of the most important chips itself—reverberates beyond Silicon Valley.

“Intel’s inability to solve this problem,” said David Yoffie, an Intel director for 29 years until 2018, “is something that the country is going to feel as a loss.”

Intel declined to make Mr. Swan available for an interview, citing his earlier pledge to give an update on the strategy in the coming months.

The chip maker remains America’s largest by revenue, and its latest stumbles might not threaten its sales for years, given the industry’s yearslong development cycles. It has said it expects to report record sales and near-record earnings per share for 2020, boosted by a pandemic-era surge in technology spending.

In the past, manufacturing problems have delayed the launch of new generations of CPUs and other advanced chips. While Intel is already behind on its initial schedule for cutting-edge chips set for production in about two years’ time, the company is trying to avoid any further delays even if that means outsourcing production of critical chips.

Intel must decide in the coming months where to produce those chips, Mr. Swan said on an October earnings call. Intel has already decided to turn over production of parts of a new so-called graphics-processing unit—a powerful chip used in data centers—to a contract manufacturer, he revealed in July.

“It’s much more of a company turning itself inside out a little bit to go engage the outside world,” said Navin Shenoy, who oversees some of Intel’s most important projects, including those involving chips for servers and artificial-intelligence calculations. The shift, he said, has involved “lots of debate.”

Intel’s stock has fallen since Mr. Swan’s announcement, closing at $45.68 on Thursday, down 24% from July 23. The S&P Semiconductor Select Industry Index climbed 23% in the period.

Moore’s Law

Large chunks of American business have farmed out the manufacture of products they design. Better to put resources into the creative tasks you do best, the argument goes, and let someone else risk the capital it takes to specialize in making things, often where labor costs are low. Companies from garment makers to Apple Inc. have consigned production to contract manufacturers abroad.

Intel was different.

Founded in 1968, it became a paragon of American technology leadership by designing and building the engines inside generations of personal computers and other devices. Co-founder Gordon Moore literally wrote the law that has powered one of the greatest economic advancements in world history.

Moore’s Law, which describes how engineers will find ways to shrink circuits at a predictable pace year after year, has given the world the powerful, inexpensive chips that underpin Google searches, Facebook pages, smartphone apps, streaming video—everything the world considers a technological given.

Intel’s orthodoxy played off that law: Its chip engineers could create better circuitry if they worked directly with manufacturing engineers to match new chips to new production equipment. That would be harder with an outside manufacturer.

The company’s 2020 strategy shift recalls another crisis at Intel. In 1985, it abandoned the huge market for so-called random-access memory, chips that store data. It was a major player in the technology and felt it needed to stay in the business to be competitive. But as Moore’s Law made memory chips cheaper and cheaper, Intel couldn’t keep up with huge Japanese rivals that poured capital into ever-newer factories. The chips had become commodities.

Intel’s departure from that memory market was cited at the time as evidence America was losing its technological edge. Instead, for Intel the move ushered in an era of prosperity. It focused resources on its strong suit—designing CPUs, which commanded higher profits in part because their circuitry was protected by Intel’s intellectual-property rights.

Combined with Microsoft Corp.’s Windows operating system, Intel’s chips formed the “Wintel” duopoly that ruled the world for years—giving the companies such dominance they became targets of federal antitrust regulators.

Intel lost some Silicon Valley luster in the 21st century as new internet giants emerged, but it remained a powerful force in the market to supply processors that help drive the new giants’ offerings. Intel’s chips power most of the world’s PCs and are in nearly all server computers in data centers that process companies’ data and that Microsoft, Alphabet Inc.’s Google and Amazon.com Inc. use in their cloud-computing operations.

The company now has 10 major manufacturing sites, four of them in the U.S.

Rival strategies

Other chip makers have embraced a divorce from manufacturing. Advanced Micro Devices Inc., a longtime Intel competitor in the processor market, spun off its factories about a decade ago. Nvidia Corp., which specializes in graphics-processing chips, has always relied on outside manufacturers and overtook Intel this year as the largest U.S. chip company by market value.

Among the advantages: Chip companies don’t have to sink capital into factories and don’t have to worry about running under capacity in their own factories when demand flags. When a contract manufacturer has problems, chips designed to be built by others tend to be easier to shift to other producers in the global chip-production “ecosystem,” as the industry calls it, which has gravitated to standardized design and manufacturing approaches.

Intel does outsource some manufacturing—the amount totals about 20% of its production today—when demand for fast-growing new products such as chips inside cars outstrips Intel’s capacity, according to the company. The company also often lets chip makers it acquires continue to use external manufacturers.

But it always made CPUs in-house. A confidence pervades Intel’s engineering and management culture that it can overcome the challenges of keeping up with Moore’s Law in factories, according to current and former Intel employees.

By 2018, competitive pressure was growing. Chip giants like Taiwan Semiconductor Manufacturing Co. and South Korea’s Samsung Electronics Co. had caught up with Intel in the number of transistors their factories could fit on a square millimeter of silicon. They make chips under contract for some Intel competitors, including AMD and Nvidia, so their factories are, in essence, competing with Intel’s plants.

Intel sought to cope by inventing new types of transistors that performed better without getting smaller and by inventing new ways of knitting chips together that gave them a performance advantage.

Modern chip factories can cost tens of billions of dollars to set up and involve machines capable of imprinting circuits only several atoms wide. The chip industry continually migrates to smaller circuitry and to new factory equipment that can produce it. In 2015, the most-advanced chips in production had 14-nanometer transistors, a measurement loosely referring to their size and equaling about one 10-thousandth the width of an average human hair.

In 2018, the industry had moved toward smaller, 10-nanometer chips. Intel’s then-CEO Brian Krzanich gambled the company could do something unprecedented with that generation: cram 2.7 times as many transistors into the same space as before. Typically, the industry about doubled transistor density with each step forward, and success could have vaulted Intel ahead.

The move, Mr. Krzanich told analysts and shareholders in 2018, turned out to be overly bold. To get there, he said, engineers took risks in the development process that caused problems that were difficult to overcome.

As a result, Intel hit repeated manufacturing snags and product delays, which, along with surging demand for chips, contributed to a shortage of its CPUs that crimped sales across the global personal-computer industry. One problem was physics. As transistors shrank, electricity started to behave in unexpected ways, requiring new combinations of materials and chip designs—a growing hurdle for all chip makers.

Mr. Krzanich, who was forced out at Intel in 2018 for unrelated issues—having what the company described as a consensual relationship with an employee—didn’t respond to requests for comment. Mr. Swan, then interim CEO, apologized to customers because of manufacturing delays that reverberated through the industry.

Intel has since ironed out difficulties with its 10-nanometer chips, but the company is continuing to deal with the aftermath of the repeated delays. Overcoming production problems was complicated by the tight linkage between Intel’s chip design and manufacturing operations that had evolved over decades. Because Intel optimized designs for its own chip-making tools, it couldn’t easily turn to outside manufacturers for help to catch up. That made it hard to swiftly recover from missteps.

Increased flexibility

Intel was taking technical steps to shift toward increasing its manufacturing flexibility. Meanwhile, engineers joined Intel from chip makers that relied on outsourcing and wanted to accelerate that shift. To them, constraining chip design because of the idiosyncrasies of the in-house factories made no sense, said an engineer familiar with Intel, even though they also saw the advantages that came with Intel’s keeping its own factories.

Intel’s acquisition of NetSpeed Systems Inc. and the arrival of NetSpeed CEO Sundari Mitra helped accelerate chip-design standardization that let chip architects more easily take advantage of any manufacturing process, whether internal or external, the company said.

By 2019, Intel engineers and executives were debating how to manufacture future 10-nanometer CPU chips that had been held up because of earlier engineering delays. The debates were sometimes fierce, with some engineers urging management to consider letting someone else fabricate the chips if in-house facilities couldn’t, and some executives arguing that the factories could fix their problems.

Chief Engineer Venkata “Murthy” Renduchintala told analysts in May 2019 that Intel had learned lessons from earlier stumbles and that its 10-nanometer chips were on track. Intel’s next generation—7-nanometer CPUs—were on track to start production in 2021, he told them.

That didn’t happen. The manufacture of the next generation of CPUs is now a year behind initial plans, which will delay the arrival of products on the market by six months, Intel said. Intel shook up its technical team and announced Mr. Renduchintala’s departure. He declined to comment. Intel declined to comment on the departure, citing a statement at the time that he left amid a management shake-up aimed at improving the company’s chip-technology execution.

Mr. Swan in the July call told analysts: “We’re going to be pretty pragmatic about if and when we should be making stuff inside or making outside.”

The company’s new approach, Mr. Swan said, would be to make market-leading chips on schedule. Intel’s factories would be the preferred manufacturing option, but, if needed, production could be outsourced. Intel still plans to invest heavily in its own factories and future cutting-edge transistor technology, Mr. Swan has said.

As part of its move toward more outsourcing, Intel is adopting for some chips what it calls “disaggregation”—a process that lets it make a single chip using manufacturing processes in different places. Intel might start a chip in one in-house factory and then move it to another, or might start making a chip at an Intel plant and then ship it to an outside manufacturer to add elements Intel doesn’t produce as well. The company said it is beginning that type of mixed manufacturing, but on a limited basis with chips including a coming graphics-processing unit.

“As you move to this disaggregated design or modular design approach,” Intel’s Mr. Shenoy said, “we can take various pieces of a chip and choose different foundries.”

Mr. Swan said during the October analysts call that Intel would decide by early next year how to handle chips in 2023 and 2024.

If Intel had embraced the new design approach sooner, it would have spared itself problems with the 10-nanometer chips, said Raja Koduri, Intel’s chief chip architect, in an August virtual presentation. Now it will be late 2022 or 2023 until its design flexibility will be in use in a wide range of chips, he told The Wall Street Journal that month.

Outsourcing poses risks, said François Piednoël, a former Intel engineer who left in 2017 after two decades, because it involves changes in the chip-design process that sacrifice performance. Mr. Piednoël, an Intel shareholder, is pushing for more chip-design expertise on Intel’s board. The board includes just one semiconductor expert, although it does include people whose current or former employers have been major semiconductor consumers.

Mr. Piednoël ruffled feathers in August and September with YouTube videos that cast doubt on Intel’s chip-design choices and critiqued its corporate culture, which he says has veered away from the combative yet collaborative ethos set under its longtime CEO, the late Andy Grove.

Intel still sees having chip factories as an advantage over rivals that must rely on others, said Intel’s Mr. Shenoy. “Having advanced technology manufacturing in the United States,” he said, “is a super important competitive advantage that is not lost on us.”