'Long, complicated and painful': Washington wrestles with strategy to boost computer chip production

April 5, 2021

By Justin Doubleday

When Mark Lewis walked into the Pentagon in late 2019, he thought his most important job as a leading research official would be overseeing the development of new, high-speed missiles.

The veteran aerospace scientist and former Air Force official was rejoining the Defense Department as it sought to regain the advantage from China in developing fast-flying "hypersonic" weapons. The year before, DOD's research chief had called it his top priority.

But on his first day back, Lewis had a conversation with Nicole Petta, principal director of DOD's microelectronics portfolio, about how the military was stuck buying outdated microelectronics.

She explained that companies in East Asia were manufacturing the most advanced integrated circuits, while the U.S. share of microelectronics production had plummeted. Meanwhile, China was advancing plans to make its semiconductor industry a global leader.

Petta was working on a strategy to get DOD to use more advanced microelectronics in its systems, in part by incentivizing manufacturing in the United States.

"Holy crap, this is really important," Lewis remembers thinking. Months later, he announced microelectronics had supplanted hypersonic weapons at the top of the military's technology list.

"It had to be microelectronics, because microelectronics are fundamental to everything the department does," said Lewis, who left DOD in January and is now executive director of the National Defense Industrial Association's Emerging Technologies Institute.

After the COVID-19 pandemic wreaked havoc on global supply chains, concerns about the semiconductor supply chain have extended well beyond the halls of the Pentagon. Top U.S. lawmakers and the Biden administration are discussing whether to pour billions into programs to incentivize companies to build new chip "fabs" in the United States.

The dual concerns about fragile supply chains and China's plans have been escalated by a global shortage of chips that has forced automakers and other sectors to slash production and shut down factories.

"We need to make sure these supply chains are secure and reliable," President Biden said in February. "I'm directing senior officials in my administration to work with industrial leaders to identify solutions to this semiconductor shortfall and work very hard with the House and Senate."

Funding for chips could be coming soon. As part of a $2 trillion infrastructure package outlined last week, Biden called on Congress to allocate $50 billion to fund semiconductor manufacturing and research programs that lawmakers have already authorized, but not yet funded.

As momentum builds toward funding a U.S. chip renaissance, policymakers are grappling with questions about how to shape what amounts to an attempt at "industrial policy" by the United States for the first time in decades.

"I don't think there's a well-defined agreement about what it actually means to lead in microelectronics," Victoria Coleman, former director of the Defense Advanced Research Projects Agency, said in an interview.

The National Security Commission on Artificial Intelligence, an influential advisory board, is also pushing for Congress to fund new semiconductor manufacturing facilities. The group is also recommending the White House develop a National Microelectronics Strategy to ensure the United States stays two chip generations ahead of China.

"We need to do the steps necessary to do that, which are long and complicated and painful," Eric Schmidt, the former chief executive of Google and chair of the commission, said during a Senate Armed Services Committee hearing in February. "But it's really important. We were in this business, and we got out of this business. We should get back into it."

'Frightening vulnerability'

The U.S. semiconductor industry remains the strongest in the world by sheer sales volume, consistently accounting for 45% to 50% of global sales in recent years, according to a report produced last year by the Semiconductor Industry Association.

However, many leading U.S. firms like Qualcomm and NVIDIA are “fabless," meaning they design their chips, but outsource the manufacturing to production facilities called "fabs."

In 1990, the United States accounted for 37% of semiconductor manufacturing. Today, it represents just 12% and is projected to dip to 10% by the end of the decade, according to the SIA report.

Chip companies began offshoring and outsourcing more than 30 years ago to cut costs and increase efficiency, according to Mike Fritze, vice president of the Potomac Institute for Policy Studies, a nonpartisan research organization based in Washington. Firms in Asia became increasingly dominant in semiconductor production as demand for commercial chips exploded with the rise of personal computing and then mobile phones.

"A rather complex international supply chain has built up as a result of natural forces in business to try to get the cheapest product possible," Fritze said.

Government subsidies also played a crucial role. According to the Semiconductor Industry Association report, the 10-year cost of owning a fab in the United States is 30% higher than in Taiwan, South Korea and Singapore, and upwards of 50% higher than in China, largely because those countries offer manufacturing subsidies and the United States does not.

The gap is "enormous," the association says in its report, considering the 10-year cost of building and operating a major advanced fab ranges from $10 billion to $40 billion.

"How do you build this huge semiconductor empire? Well, it's because those governments chose to give offsets and credits to make it work, because they wanted to build that industry and they understood how valuable it was," said Samantha Clark, of counsel at law firm Covington and Burling. Clark was previously deputy staff director for the Senate Armed Services Committee.

As manufacturing shifted to foreign shores, DOD maintained access to "trustworthy" microelectronics through a "Trusted Foundry" program established in the early 2000s. The program vets microelectronics suppliers by ensuring their facilities are secure and doing background checks on employees. All the facilities on DOD's trusted suppliers list are in the United States.

But DOD’s emphasis on using "trusted" microelectronics in its systems has also resulted in the department using increasingly obsolete products because the suppliers served the narrow defense market and had no incentive to keep pace with the rapidly evolving commercial market, according to Lewis.

In some cases, DOD had to make expensive "end of life" buys to stockpile electronics that a trusted manufacturer was phasing out, Lewis said.

"You spend a lot of money stockpiling Model Ts in your garage," he added.

Today, Taiwan Semiconductor Manufacturing Corp. is widely considered the most advanced chipmaker in the world, while Samsung in South Korea is also producing some of the most sophisticated chips. Intel is the only advanced semiconductor manufacturer in the United States, but it has fallen behind TSMC and Samsung on the most cutting-edge chips in recent years.

Meanwhile, China has declared its ambitions to be a global leader in semiconductor manufacturing by 2030. China's most important chip company, Semiconductor Manufacturing International Corp., can't yet compete with the most advanced manufacturers, but it posted a record year in 2020 despite U.S. restrictions on buying advanced manufacturing equipment and other critical technologies.

The Pentagon's 2020 Industrial Capabilities report highlights growing concerns about depending on semiconductor suppliers in East Asia.

"China is projected to dominate global semiconductor production by 2030, and in the meantime, current suppliers in Taiwan, South Korea, Malaysia, and elsewhere are in easy range of Chinese missiles, subversion, or air or maritime interference," the report states.

The report notes other countries have invested billions to fund the construction of new fabs. "The hard truth," it argues, is U.S. chip production will decline to the point of "irrelevance" if Washington does not start investing in new production facilities.

"We will have few new fabs," the report continues. "We will have fewer semiconductor production jobs. We will have frightening vulnerability to foreign cutoffs whose impact would make our COVID-related shortages look miniscule."

Grappling with industrial policy

Despite the sizeable concerns, the Pentagon report also points to reasons for optimism, including Congress recently passing "landmark" legislation that "opens vistas for future creative pooling of federal and private capital to fund fabs in the United States."

The report is referencing the "CHIPS Act," which passed as part of the Fiscal Year 2021 National Defense Authorization Act. The legislation authorizes a Commerce Department grant program to incentivize building new fabs, a 40% refundable tax credit for manufacturing investments and new research programs.

There are signs the U.S. discussion of incentives is already paying off. Last year, TSMC announced it would build a new $3.5 billion fab in Arizona, while Intel last month unveiled $20 billion plans to expand its Arizona manufacturing capacity, including the construction of two new fabs.

But lawmakers have yet to fund the CHIPS Act programs. While Biden has called on Congress to allocate $50 billion for the CHIPS Act through his infrastructure proposal, Senate Majority Leader Chuck Schumer (D-NY) is also developing legislation aimed at technology competition with China, including more funding for semiconductor manufacturing.

But Clark said there's always a risk lawmakers could balk at the price tag.

"You either invest fully to solve it or my concern is the government will do what it often does, which is well, 'We're going to kind of split up funding and there are things here and there, so we'll just give a little bit,' and it will ultimately fail because there isn't the influx needed to actually make it successful," she said.

Others worry firms will take advantage of the subsidies to build new fabs in the United States, but won't make the most advanced chips there.

"I'm concerned that TSMC and Samsung are building fabs in the U.S., but that their most advanced fabs are in Korea and Taiwan," Schmidt, the former Google chief, said during a Jan. 26 meeting of the National Security Commission on AI. "So one of the criteria here is to make sure you actually understand what you're getting because these are very expensive things."

Meanwhile, the Trump administration largely sought to stunt China's grand plans by limiting the export of key technologies, including chips. But as U.S. officials consider how to promote more domestic manufacturing and new fabs, they'll have to weigh export controls carefully.

Coleman, the former DARPA director, offered as an example the federal government giving Intel a grant to build a new U.S.-based foundry.

"Are we going to for example say, 'You should not sell to the Chinese market.' Some people would argue that. Why would we help our Chinese competitors?" Coleman said. "On the other hand, if you're Intel, now you've taken away a significant chunk of their market. So now we've made it much harder to compete with TSMC and others because they don't have the opportunity to make as much profit."

"Thinking through regulation and export controls is a very important part of this plan," she added.

Coleman also highlighted a skilled workforce and a "thriving and diverse research community" as potentially crucial in the United States quest to bolster its microelectronics manufacturing sector.

In her view, the lingering questions about how the United States will pull off its grand chip ambitions arise from the lack of a national strategy. The CHIPS Act offers many "inputs" for semiconductor manufacturing and research, she said, but little in the way of "outputs," or objectives.

"At the end of the day, you have to figure out what your goals are," Coleman said. "Those goals, in my opinion, ought to be set by the executive branch."

As Congress and the White House weigh funding, the Biden administration is also in the middle of a review of semiconductors and other supply chains that could offer more details on the U.S. strategy. The results of the review are due in early June.

Meanwhile, the FY-21 NDAA also tasked the Pentagon with working across government to develop "a detailed microelectronics strategy to serve the national security needs of the nation" by June 1, according to the explanatory statement accompanying the legislation.

It directs DOD to consider "different models for public-private partnerships, including establishing a semiconductor manufacturing corporation, to restore American security and competitiveness in the manufacture and assembly and packaging of state-of-the-art microelectronics."

In January, the Pentagon established a "microelectronics cross-functional team" to develop the strategy, according to a Defense Department spokeswoman. The team reports to both the under secretary for acquisition and sustainment as well as the under secretary for research and engineering.

As officials debate strategy, Lewis said the Pentagon's research and engineering directorate is already making headway with its new approach to microelectronics, called "Zero Trust." Rather than relying on top-to-bottom security reviews of "trusted" vendors, the new approach assumes nothing the department buys is safe and relies on a series of processes to validate whether a microelectronics component is secure.

The approach, he said, opens the door to DOD using more commercial chips.

Earlier this year, the research directorate teamed with the Navy to launch a new program aimed at incentivizing "a U.S.-located, leading-edge foundry that can fabricate the assured leading-edge custom integrated circuits and commercial-off-the-shelf products required for critical DOD systems," according to the DOD notice on the project.

Lewis said the Pentagon needs to continue shifting toward using commercial microelectronics if it wants to take advantage of commercial technologies in areas like artificial intelligence and space systems.

"We don't want to be stuck buying sailing ships in a steam-powered era," Lewis said. "We want the department to have these available."