The U.S. Department of Defense (DoD) will probably take years to wean itself from Asian chip supplies because of a dearth of investments in domestic production capacity, according to industry insiders and government observers.

Global chip shortages have impeded the ability of DoD contractors like Lockheed Martin and Raytheon to boost production of weapons used in the Russia-Ukraine War.

The DoD must also reckon with competition from commercial buyers for foundry wafers, as well as the military’s over-reliance on older chip architectures.

As a result, concerns are surfacing that the U.S. may have difficulty responding, for as many as 10 years, to a war involving Taiwan—which is home to manufacturers that, ironically, supply chips used by the DoD.

Taiwan is at the center of rising tensions between the U.S. and China. On Dec. 25, China sent a record 47 aircraft across the median line in the Taiwan Strait, according to Taiwan’s Ministry of National Defense. That incursion came days after the U.S. passed the 2023 National Defense Authorization Act, which allows for up to $10 billion in security assistance to Taiwan over the next five years.

The DoD may need a decade to build a reliable domestic supply chain, according to Mike Burns, an electronics engineer who has founded chip companies like iDeal Semiconductor and Agere Systems. The issue is how fast U.S.-based Intel can catch up with Taiwan Semiconductor Manufacturing Co. (TSMC), which makes Altera and Xilinx field programmable gate arrays (FPGAs) and other chips that the DoD uses in weapons systems like the F-35 fighter jet, missiles and command-and-control gear, he said.

“Maybe that’s a three-year effort,” he told EE Times. “I’m just saying that it’s many years.”

To be sure, TSMC more than tripled its overall investment in Phoenix, Ariz., to about $40 billion in December to make way for a second U.S. chip facility that’s set to start production at the 3-nm node in 2026. While Apple and AMD aim to buy chips from TSMC’s Arizona site, the fab isn’t likely to be a DoD supplier, Burns said.

“I don’t think TSMC would do that,” he said. “If you’ve seen their stance on American engineers, they don’t seem very pleased with them.”

TSMC has often said it aims to keep its most advanced production technology in Taiwan, where the world’s largest foundry makes more than 90% of its chips. The company last year began rolling out its first 3-nm chips in Taiwan—years ahead of the U.S.

TSMC has encountered a culture clash with some of the engineers it hired in the U.S. to work at the Arizona site. That challenge and TSMC’s focus on serving large customers like Apple and AMD will make Intel’s smaller foundry service a more likely fit as a DoD supplier, Burns noted.

“That will be up to Intel,” he said. “Intel Foundry Services can make application-specific integrated circuits (ASICs) that are one or two generations behind the process at TSMC. Maybe that makes up the difference.”

Years ago, Intel bought FPGA supplier Altera, which uses TSMC to make the programmable chips that are easily modified yet lag the performance of ASICs. The DoD is forced to rely on FPGAs because it can’t find a large foundry like TSMC or Samsung  to make custom ASICs for its weapons systems.

The competition from commercial customers like Apple and AMD for foundry wafers has diminished the purchasing power of small buyers like the DoD. Even U.S.-based GlobalFoundries gives little priority to the DoD, Burns said.

“GlobalFoundries has a ton of commercial customers pushing for volume. We never know with an increase in military wafers: How long will that last? Usually they’re high cost, low volume, and they’re unpredictable in volume.”

The U.S. military relies on legacy, or older, chip architectures that aren’t widely available and often must be made in small, dedicated batches, Bryan Clark, a senior fellow with the Hudson Institute, told EE Times.

“When chip shortages occur, companies that make these chips can make more money building chips for appliances and cars,” he said. “If the DoD used more state-of-the-art chips with commercial architectures and adapted them through packaging or software, the U.S. military would be able to tap into the scale of commercial production and be less vulnerable to chip supply disruptions.”

China, which the U.S. has punished with recent chip export-control measures, faces supply chain issues of its own. The U.S. considers China a “strategic competitor” that’s a threat in tech domains like 5G telecommunications and AI.

“The PRC military already uses a significant fraction of commercially available chips, but they are completely dependent on U.S. and Western technology to build those chips, or the PRC military has to buy the chips from potential rivals like Japan or Taiwan,” Clark said. “The PRC military is therefore better able to weather supply disruptions, but less prepared to address supply cut-offs that occur for national security reasons.”

FPGAs lag ASICs

The DoD needs to reserve a certain amount of capacity for surges in demand like that created by the Ukraine War, Burns said. He proposed to the U.S. Congress three years ago that the U.S. government should subsidize a joint venture with an advanced fab like TSMC for dedicated DoD capacity. Small U.S. foundries like SkyWater can’t meet the DoD’s need for advanced chips, he added.

“As much as we use the SkyWaters of the world at 90 nm and 130 nm, we still need advanced compute, but we can’t get it,” Burns said. “No trusted fab does 5-nm compute. We tend to buy FPGAs and configure them in the U.S. where the configuration can be secret. But with any FPGA, you have a tradeoff of performance because a purpose-built ASIC is going to be faster.”

The DoD did not respond to a request for comment from EE Times.

Suppliers of weapons systems like Raytheon say chip shortages limit their ability to expand production.

“There is no magic formula in the supply chain to accelerate this,” Raytheon CEO Greg Hays said in a Dec. 7 interview with CNBC. “We’re working to do as much as we can to try and cut that lead time down. But there are thousands of components. Not surprisingly, chips are a key element in this equipment. We’re working with the DoD to find other alternatives, perhaps in terms of some of the suppliers.”

In its mid-October 2022 earnings call, Lockheed Martin said its sales this year will be little changed from last year. With continuing impacts from the Covid pandemic and supply chain challenges, the company expects a return to growth in 2024.

Raytheon and Lockheed Martin operate aging chip fabs in the U.S. that are on a list of trusted DoD suppliers.

Javelins and Stingers sent to Ukraine

Raytheon and Lockheed Martin make weapons like the Javelins and Stingers that have helped Ukraine turn back the invasion by Russia. Yet the depletion of the U.S. arsenal is concerning, according to the Center for Strategic and International Studies (CSIS), a U.S.-based think tank.

“Some U.S. inventories are reaching the minimum levels needed for war plans and training,” CSIS said in a September report.

The DoD has requested congressional approval to fund increased production capabilities for High Mobility Artillery Rocket Systems (HIMARS) and Guided Multiple Launch Rocket Systems (GMLRS), the CSIS report said.

The industry’s general position, however, is that the DoD should make commitments for multi-year acquisition to justify industry investment in surge capabilities, according to the report.

A CSIS study examined the ability of the defense industrial base to replace inventories in an emergency and found that the process would take many years for most items.

Guided rockets like HIMARS and GMLRS are enormously useful, but the numbers are likely limited, according to CSIS.

“The U.S. has an estimated inventory of 25,000 to 30,000 remaining from a production run of 55,000 rockets to date,” CSIS said in the report. “If the U.S. sent one-third of that inventory to Ukraine (as has been the case with Javelin and Stinger), Ukraine would receive 8,000 to 10,000 rockets. That inventory would likely last several months but, when the inventory is exhausted, there are no alternatives. Production is about 5,000 a year. Although the U.S. is working to increase that amount, and money has recently been allocated for that purpose, it will take years.”

More than 200 Javelin missiles and launchers and 250 Stinger systems approved for sale to Taiwan in 2015 haven’t been delivered, Mark Montgomery and Bradley Bowman noted in a Defense News report. Montgomery, a former U.S. Navy Rear Admiral, is a senior fellow at the Foundation for Defense of Democracies (FDD). Bowman is senior director of the FDD’s Center on Military and Political Power.

Given the need to supply Ukraine and restock U.S. and allied inventories, it’s not realistic to expect deliveries before 2026 or 2027, they said.

The U.S. military isn’t structured to fight or support an extended conflict, according to CSIS. “That should, of itself, spark some debate in the national security establishment about budget priorities.”

Stockpiles ‘seriously depleted’

The DoD will likely count on Intel to supply “More than Moore” chips, where added value comes from heterogeneous integration of functions like analog that don’t necessarily scale to smaller nodes in line with Moore’s Law.

“U.S. smart-munitions stockpiles have become seriously depleted, and they mostly rely on More-than-Moore solutions,” TechInsights analyst Dan Hutcheson told EE Times.

Intel will need to finish its planned merger with Tower Semiconductor of Israel to meet DoD expectations, he said. “[Intel] still has to finish the deal with Tower to have a complete More-than-Moore solution. This is looking less and less probable each week it extends out.”

Intel must also get Tower’s fabs in Israel certified for trusted-fab status and ITAR (International Traffic in Arms Regulations), Hutcheson said.

Intel Foundry Services (IFS) has stumbled in its quest to keep the DoD its largest customer. IFS President Randhir Thakur resigned on Nov. 22. Thakur will stay at IFS through the first quarter of this year to ensure a smooth transition to the next leader, according to Jason Gorss, an Intel spokesperson.

It’s difficult to transform an integrated device manufacturer like Intel into a foundry, chip entrepreneur Burns said. “Samsung did it. TSMC is built from the ground up to culturally serve third-party designers. They put a lot of extra benefits out there for people to stay with them. Switching costs are pretty substantial.”

Rogers: U.S. is vulnerable

The DoD needs years to build a secure chip supply chain that doesn’t rely on Asian nations like China, according to Mike Rogers, former chairman of the U.S. House Permanent Select Committee on Intelligence.

“We do have some unique capability in the U.S. to produce high end [chips], but the volume is not where it would need to be,” Rogers told EE Times. ”We supplement that with purchases all over the world, and China has a big part of that. If there are microprocessors that have been altered beyond what the specs are, that weapon system doesn’t even turn on when it’s supposed to, and we have a whole bunch of dead sailors, soldiers or Marines. Do I think it has made us vulnerable? I do.”

The CHIPS Act was designed to solve some of those vulnerabilities, Rogers noted.

“It certainly was a flawed bill, but it was still a step in the right direction to recognize that we have to have our own capabilities. We’re going to have to find out what is that right equilibrium of capacity in the U.S. in case something happens where we actually need those chips,” he said. “We are subject to people slowing down production or turning off production or having production delays. We should never have put ourselves in that position. We’re going to have to unwind it. That’s why you see serious amounts of money for investment in the capability of producing chips, either with strong allies or in the U.S. itself.”

The U.S. government should dedicate a portion of the $52 billion in CHIPS subsidies to military chip fabs because too much money may be handed to Micron, Texas Instruments and TSMC, Burns asserted.

“Then you haven’t changed anything for Raytheon, SkyWater and TSI Semiconductor,” he added.
“They’re tough business models. Their investors are not pushing capital in. One big problem is just how quickly that [production] ramps up.”

By EETimes