Fab, ASIC partners on Tom Caulfield's wishlist
PALO ALTO, Calif. — The new CEO of GlobalFoundries has a charter to improve financial performance of the privately held chipmaker. Thus, Tom Caulfield is seeking partners for the company that is a distant second in a hotly competitive race to make chips.
Caulfield needs a hand building a next-generation fab, probably at 3 nm, and expanding his ASIC services to attract new customers. Meanwhile, he started a reorganization geared to make the company more nimble and to hold his managers responsible for financial progress, he said in an interview with EE Times.
The new fab is probably best suited as an expansion of GF’s existing Fab 8 in Malta, New York, where it is preparing to ramp a 7-nm node. Such a facility likely would need support from federal funds, but GF has other options leveraging its fabs in China, Germany, and Singapore.
In the first 60 days in his new role, Caulfield and representatives of the company’s owners, the Mubadala Investment Company in the United Arab Emirates, spent a day and a half in Washington, D.C. It was one leg in a world tour of GF’s fabs, exploring the techno-politics of the options.
“Saxony provides 25 cents on the dollar, but the U.S. doesn’t … it’s a policy that needs to be reconsidered because these things can drive jobs away,” said Caulfield.
The U.S. Department of Defense could be attracted by the promise of access to 3-nm chips as an expansion of the trusted foundry arrangement that GF acquired with IBM’s fabs in 2015. “This is important for national security and creating jobs … access to a secure domestic supply — we’ll work that angle,” said Caulfield.
Meanwhile, he’s also angling for partners in China as GF starts to bring equipment into its new site in Chengdu, sized to make a million wafers a year. GF could “make it China’s factory, not just ours — customers or even competitors can invest in capacity to create a multi-ownership model rather than trying to fill it on our own,” he said.
Caulfield also likes the idea of another process-sharing deal. He praised the 14-nm collaboration with Samsung for giving customers a second source and providing capacity on the node “similar to” that of rival TSMC.
Indeed, one of Caulfield’s stellar achievements to date was bringing up the 14-nm process at GF’s Fab 8 in Malta. AMD used the process to make the Ryzen CPUs and Radeon GPUs, now returning it to profitability.
Given the uncertain time frame for a next-gen fab, it’s not surprising that Caulfield leans toward 3 nm.
“I don’t know if 5 nm is enough to make a fabless company invest … they need something as defined as 3 nm to get full performance, but we’re still looking at what is the right investment for the next node,” he said.
In the end, GF needs a next-generation fab more than any potential partner. It must serve legacy customers such as AMD and IBM and attract high-valued new ones.
On ASIC and packaging services
Caulfield wants new partners with IP to expand GF’s ASIC service that he sees as a way attract new customers.
“More and more system companies want to differentiate themselves with silicon, but you don’t just wake up being a silicon design team … first-time silicon engagements are typically doing it through ASICs,” he said, noting that GF is the only company other than Broadcom to have both ASIC services and fabs.
“When Apple got off standard products, it went to Samsung at 32 nm, and by the time it got to 14 nm, it had a complete capability to do its own designs — that’s how these other customers will do it.”
The ASIC service that GF acquired with IBM is already doing customer designs in networking and machine-learning chips with high-end CPUs on the roadmap. However, Caulfield has no illusions of winning the business of mobile applications processors in the near future.
“We don’t have the capacity … [and] we will be behind the leading 7-nm [foundry], but it costs more to be a leader than a fast follower,” in leading-edge nodes, he said.
Beyond the high-profile smartphone processors, “there’s plenty of business to go around … [and] some of the most profitable chips are in mid-range mature segments where assets are depreciated,” he added.
GF also won’t try to follow rival TSMC in offering a variety of packages such as its 2.5D CoWoS and its InFO wafer-level fan-out for mobile processors.
“We’ll do development, and OSATs will do the productization … who wouldn’t want a turnkey solution to offer customers?” he said. “The two TSMC technologies are very powerful, especially in mobile devices, but we will spend more time on 2.5D and 3D chip stacks for the data center.”
Financial goals and filling fabs
With the exception of the next-gen fab, Caulfield sees his job as filling the company’s existing factories with the most profitable products possible.
“Our legacy technologies in Singapore add very little value for customers, so we capture little,” he said “I’d rather convert that with targeted investments to make higher-value products.”
Owner Mubadala invested in technology for the long term, but it wants to see a path to gains, he said.
“You are getting ahead of yourself with more investment without leveraging what you invested. If you demonstrate better return on assets deployed, why wouldn’t we put more money in? Its investing and digesting before you invest again.”
Caulfield believes that he can deliver before 2020 a 1.5–2x gain in earnings before interest, taxes, and amortization (EBITA), a financial metric short of net profits. Thus, Mubadala’s goal for GF is not necessarily to get out of the red but to “generate cash, which lets us grow our business — we’ve been over-invested,” he said.
Caulfield echoed comments of Khaldoon Al Mubarak, chief executive of Mubadala. He manages a $250 billion portfolio, and GF is the largest share of the $100 billion of it in the U.S.
The Abu Dhabi-based investor “was about accumulating assets and maintaining them … [but now focuses on] growth … and monetizing them when the time is right,” said Al Mubarak in a recent interview with Bloomberg.
To meet its new goals, the company “needs to be one mission, one organization, and one Globalfoundries,” Caulfield told EE Times.
“It was four companies in a way with an AMD spinout in Dresden, the acquisitions of Chartered in Singapore and IBM in the U.S. … we started Malta before there was a company culture, so Malta was raised by wolves,” quipped Caulfield.
The changes could address criticisms that GF has been slow to deliver on promises. For example, it was early to talk about fully depleted silicon-on-insulator as a low-cost alternative to FinFETs. But Samsung won the initial FD-SOI business, and now the two companies are competing to ramp customers.
Now all GF fabs have general managers sharing responsibilities for their financial performance.
They will “own the cash generation from these assets and be more involved and engaged in their commercial operations … the real game is optimizing cash margins — some days that may be through productivity, some days through better features,” he said.
In addition, Caulfield has put IP and ASIC design teams in one customer-facing unit. Such steps aim to speed decision-making, and “that’s a key thrust for me,” he said.
For now, Caulfield is done reorganizing his reports. Follow-on changes are expected to finish trickling through the company by July. Next up, partnerships.
— Rick Merritt, Silicon Valley Bureau Chief, EE Times