IBM Exec on Intel Chipmaking Collab: “Choose Your Strength”


IEEE Spectrum

April 1, 2021

Intel and IBM—two of the titans of the U.S. semiconductor industry over the last 40 years—have joined forces to advance next-generation logic and packaging technologies. This new partnership comes as a bit of a surprise to anyone who has observed the U.S. semiconductor industry over the last four decades and witnessed the rivalry between the two companies.

The backdrop for these two long-term competitors joining forces may help explain the timing for this unusual partnership.  Last week, Intel announced a strategy it’s calling IDM 2.0. (IDM, stands for integrated device manufacturer, as opposed to say TSMC, which is a pure-play foundry.) A major part of that scheme is to double down on manufacturing, pumping $20-billion into new fabs, and jump start its foundry services, manufacturing chips for other companies.

Industry observers say reviving the foundry service business in order to go toe-to-toe with Taiwan-based TSMC is a bold move. However, this would not be the first time Intel had tried this strategy, with its previous attempt resulting in little success (some would describe it as a “failure”).

However, this time around Intel is partnering with Big Blue, which, as an IDM itself, had managed to successfully run a foundry service for nearly twenty-five years, until it ultimately sold the business to Global Foundries in 2014. This latest R&D partnership with IBM should be beneficial in making Intel’s foundry business—Intel Foundry Services or IFS—a success.

Based on IBM’s background, it certainly should have some friendly advice on how to run a foundry service to its new partner. The key for the IFS initiative being a success, according to Mukesh Khare, vice president of Hybrid Cloud at IBM Research in Albany, NY, will be Intel committing to IFS both in terms of money and also to a persistent focus on its core capabilities.

“There are leading-edge foundries and those that are not. Pick your battle and invest in big partnerships,” Khare said. “Every company, however big they are, is good in only certain areas. Choose your strength and double down on it and pick that market where you can differentiate yourself.”

For IBM, the focus has clearly moved on to hybrid Cloud and Artificial Intelligence (AI) since selling its foundry business seven years ago.  However, since then, IBM has continued to work on making innovations to chip technology, largely through its $3-billion “Seven Nanometers and Beyond” research initiative launched soon after selling its foundry service.

With its continued focus on developing state-of-the-art chip technologies for its own business, IBM has kept its eyes downfield to the key challenges facing the semiconductor industry. These challenges Khare has identified as scaling (continued density improvements) and packaging (how do you make connections with ever-increasing density).

“These are the two areas we are going to work with Intel in this partnership,” said Khare. “How can we continue to progress the logic technology roadmap? And how can we create additional value in packaging technology so that we can sustain economic value?”

Both Intel and IBM have been working on the next generation of chip design after finFet, named for the fin-like ridges of current-carrying silicon that project from the chip’s surface. The life expectancy of finFet has been more or less set at the 7-nanometers node. If it were to go any smaller, transistors will become difficult to switch off: electrons will leak out, even with the three-sided gates.

To address this, Intel and IBM have been initiating key developments in so-called gate-all-around devices, in which nanowires completely surrounded the gate, preventing electron leaks and saving power. Both companies have also been making advances in so-called nanosheet devices in which each transistor is made up of three stacked horizontal sheets of silicon, each only a few nanometers thick and completely surrounded by a gate.

The partnership should also foster new approaches to integrated circuit (IC) packaging, such as chiplets. Intel has demonstrated particular prowess with chiplets, which involves making processors by bounding together collections of smaller, less-expensive-to-produce chiplets via high-bandwidth connections all within a single package.

Khare envisions the IBM and Intel engineers working together on these projects. There is no intention of dividing the responsibilities between the two companies based on their expertise. But instead to work from the premise that both companies have expertise in the same areas. He says it will be the cross pollination of ideas that come from working side by side that will create the synergies between the two companies.

The partnership between the two companies will start to physically occur with the exchange of researchers and engineers probably by the second half of this year, according to Khare. It is likely to be a delicate process involving intellectual property concerns for the individual engineers and companies.

“For Intel, process technology and packaging technologies are like the crown jewels of the company so this will be the first time for these engineers to partner in such a super-sensitive area,” said Khare.

An undercurrent influencing the partnership is an explicit national-initiative element. In January, the U.S. Congress passed landmark legislation aimed at boosting semiconductor and advanced packaging manufacturing in the United States. Khare envisions the partnership as a “super team for winning the game for the U.S. semiconductor business.”

“This partnership is not only great for our two companies, but it’s great for U.S. semiconductor leadership,” he says. “It means the two titans of semiconductor innovation in the [United States], who, to be honest, competed for many, many years, are coming together. That’s great for the country and it’s going to be great for the world.”