U.S. Government Awards GlobalFoundries New $3.1 Billion, 10-Year Contract for Secure Chip Manufacturing

[AleksandarK]

The U.S. Department of Defense (DoD) has awarded GlobalFoundries (Nasdaq: GFS) (GF) a new 10-year contract for a supply of securely manufactured, U.S.-made semiconductors for use across a wide range of critical aerospace and defense applications.

With an initial award of $17.3 million this month and an overall 10-year spending ceiling of $3.1 billion, the new contract provides the DoD and its contractors with access to GF's semiconductor technologies manufactured at its U.S. facilities. These GF facilities are DoD-accredited to the highest security level, Trusted Supplier Category 1A, which implements proven stringent security measures to protect sensitive information and manufacture chips with the highest levels of integrity to ensure they are uncompromised.




In addition to secure chip manufacturing for DoD systems used on land, air, sea, and in space, the new contract provides the DoD and its contractors with access to GF's robust design ecosystem, IP libraries, early-access to new technologies in development, quick and efficient prototyping, and full-scale volume manufacturing. The contract was awarded through the DoD's Defense Microelectronics Activity (DMEA) Trusted Access Program Office (TAPO).

"GF is proud to begin this new chapter of our decades-long partnership with the U.S government, and to continue serving as the leading supplier of securely manufactured essential chips for the U.S. aerospace and defense industry," said Mike Cadigan, chief corporate and government affairs officer at GF. "This partnership provides DoD programs with 'front-door access' to advanced technologies in a way that is scalable and highly efficient. For this work, GF is accredited to provide the right level of security required for each program, from GF's industry leading GF Shield protections, to strictly export controlled handling (e.g. ITAR), to the highest level of accredited microelectronics manufacturing security on the planet, Trusted Category 1A."

This new contract is the third sequential 10-year contract of its kind between the DoD and the Trusted Foundry business team at GF and is the latest milestone in the longstanding partnership between the department and the company. Click here to learn more about the importance and impact of GF's Trusted Foundry accreditations.

View at TechPowerUp Main Site

 

[TumbleGeorge]

Hmm, we're almost at the end of 2023... Is there any, even the slightest rumor that GF is getting ready to host sub-12nm lithography process production?

 

[Deleted member 185088]

Hmm, we're almost at the end of 2023... Is there any, even the slightest rumor that GF is getting ready to host sub-12nm lithography process production?

I think a lot of chips don't require advanced nodes.

 

[Frank_100]

If dotgov is willing to pay, Global Foundries can probably build anything dotgov wants.

It is the economics of low yields that keeps advanced nodes limited.

 

[TechLurker]

Hmm, we're almost at the end of 2023... Is there any, even the slightest rumor that GF is getting ready to host sub-12nm lithography process production?

AFAIK, GF gave up chasing the bleeding edge and has instead shifted towards more stable, less investment-intensive, government contracts and contracts with groups that don't need the bleeding edge or even the leading edge. They haven't made any new licensing deals with Samsung for a newer node process like they did the last time either.

---

As an aside, I wonder if TSMC is also hoping to get that sweet government contract once the AZ facility is up and running proper. That fab, as far as I'm aware, isn't going to be using the hottest nodes, but would be using a more modern one than GF.

 

[Solaris17]

I think a lot of chips don't require advanced nodes.

this full stop. The DoD doesn’t look for the latest and greatest they are generations behind the “speed” of consumer land

 

[kapone32]

this full stop. The DoD doesn’t look for the latest and greatest they are generations behind the “speed” of consumer land

I wonder if a Threadripper CPU could run the Avionics on the F22?

 

[Solaris17]

I wonder if a Threadripper CPU could run the Avionics on the F22?

maybe, but its complexity right? and they want to always save money not to mention everything is custom ASIC. anyone can get there hands on a threadripper know what I mean? This has always been close to the hip.

 

[kapone32]

maybe, but its complexity right? and they want to always save money not to mention everything is custom ASIC. anyone can get there hands on a threadripper know what I mean? This has always been close to the hip.

If I remember it right it had 4 Cray Super Computers to run the Avionics and those are still not for the General public. My Father actually worked on that project. I still remmeber the day he came home talking about how the plane not being able to fly by itself (Of course no details). I know that we could use a Threadripper to fly to any point in the Solar system.

 

[TumbleGeorge]

I wonder if a Threadripper CPU could run the Avionics on the F22?

It can. I guess the whole story will engage several of the cores. As long as someone translates the code.

 

[Wirko]

AFAIK, GF gave up chasing the bleeding edge and has instead shifted towards more stable, less investment-intensive, government contracts and contracts with groups that don't need the bleeding edge or even the leading edge. They haven't made any new licensing deals with Samsung for a newer node process like they did the last time either.

They will have to license something newer sooner or later, probably 7nm from Samsung. But that's in the long term, when 7nm stops being one of the most profitable nodes for the big three.

 

[TechLurker]

They will have to license something newer sooner or later, probably 7nm from Samsung. But that's in the long term, when 7nm stops being one of the most profitable nodes for the big three.

True, though one would think that with the CHIPS Act and other incentives to base more high-end chip manufacturing in the States, GF would at least consider licensing one of the older "leading edge" nodes and deploy it at one of their US fabs. It worked out well for 14nm after all and gave them an early lead against Intel's own for a time. It would also increase competition against Intel's own fabs and TSMC's future AZ fab plans as far as leading edge nodes go, while letting others handle the R&D investment of the "bleeding edge".

 

[AnotherReader]

True, though one would think that with the CHIPS Act and other incentives to base more high-end chip manufacturing in the States, GF would at least consider licensing one of the older "leading edge" nodes and deploy it at one of their US fabs. It worked out well for 14nm after all and gave them an early lead against Intel's own for a time. It would also increase competition against Intel's own fabs and TSMC's future AZ fab plans as far as leading edge nodes go, while letting others handle the R&D investment of the "bleeding edge".

Intel's 14 nm was the smallest node available until Samsung started shipping 10 nm processors with the Galaxy S8 in April 2017. It's comparable to TSMC's 10 nm and not comparable at all to TSMC's 16 nm or GlobalFoundries' 14 nm node. Note that, in the table below, I've used Samsung's 14 nm as a stand-in for GlobalFoundries' 14 nm even though the latter has slightly larger dimensions for performance.

Process	Gate pitch (nm)	Metal pitch (nm)
Intel 14 nm	70	52
GlobalFoundries 14 nm	78	67
TSMC 16 nm	88	70

 

[dragontamer5788]

this full stop. The DoD doesn’t look for the latest and greatest they are generations behind the “speed” of consumer land

Not just DoD, but auto-manufacturers ran out of 40nm, 90nm, and 180nm chips just two years ago. Security to higher nodes (aka: "cost optimized" processes) is important.

Us techies like 3nm or 5nm chips. But in practice, 5nm chips are luxury items (iPhones / Androids) while 40nm or 180nm chips are the workhorses of our economy (manufacturing equipment, power-management ICs, IGBTs, etc. etc.). In many ways, these older chips are "more important", even if they're far less ambitious.

Even today, there are new chips on 40nm (such as STM32U5). I think TI's new MSPM0+ (cost-optimized 50-cent microcontroller) is on 28nm but I don't remember where I read that so I could be mistaken.

Remember that chips on 40nm and 28nm are much much cheaper: $1 to $5 per chip. So while they're "only making" $2.2 Billion worth of chips or $1.4 Billion worth of chips per year, this represents many-many-many more chips than the more advanced nodes. The vast majority of our livlihood depends on 28nm or higher nodes even today. (Ex: Car antilock brakes, tire pressure sensors, windshield wiper timers, CAN bus to handle our steering wheel, accelerator pedals, engine timing control, electric car charge managers, airbags, etc. etc.). Just a brief overview of some common applications.

Even the majority of chips in your computer are likely 28nm. VRMs to send power across the motherboard, the CMOS sensor in your mouse to detect movement, the microcontroller to compare the CMOS sensor and determine direction and heading, the USB-controller to convert those signals into a form usable by your computer. The fan-speed controller to make sure you're at 500 RPM at idle but ramp up to 3000RPM when things get hot. Your CPU and GPU are the big chips, yes, but they only work because of hundreds of individual 28nm, 40nm, or even 180nm chips scattered across the motherboard that actually perform the information gathering.

 

[Wirko]

Not just DoD, but auto-manufacturers ran out of 40nm, 90nm, and 180nm chips just two years ago.

That was also the time when chip manufacturers started expanding their capacities on older nodes, for the first time in history. (But I'm not sure if this applies to 22/28 nm or older nodes as well.)

The distribution graphs such as the one you posted always show revenue per node. When you take into account that a 28 nm wafer is many times cheaper than a 5 nm wafer, and an older-tech wafer is cheaper still, the distribution (number of wafers per node) becomes a lot different. Unfortunately I've never seen a volume-per-node graph.

 

[Frank_100]

If I remember it right it had 4 Cray Super Computers to run the Avionics and those are still not for the General public. My Father actually worked on that project. I still remmeber the day he came home talking about how the plane not being able to fly by itself (Of course no details). I know that we could use a Threadripper to fly to any point in the Solar system.

A 2008 Cray XT4 had a theoretical max of 51.06 T flops.

The Thread-ripper Pro 5995WX is a little less then 2 T flops.

Obviously, our Thread-Ripper guided space craft is going to need super optimized code.

 

[dragontamer5788]

A 2008 Cray XT4 had a theoretical max of 51.06 T flops.

The Thread-ripper Pro 5995WX is a little less then 2 T flops.

Obviously, our Thread-Ripper guided space craft is going to need super optimized code.

Threadripper is a CPU. Cray was a SIMD (closer to a GPU).

AMD's Rx 7900 XT GPU is 52 TFlops (32-bit though instead of 64-bit, as were the standard for Cray / scientific compute).

So its a bit of apples vs oranges, but ~2x modern high-end GPUs beats a 2008-era Cray.

---------

EDIT: Wait, apparently I'm wrong. Cray XT3 is saying its CPU-based. Really weird, I thought Crays always used SIMD boxes. Well... whatever. I'm wrong but I'll leave my post up anyway, lol.

 

[TechLurker]

this full stop. The DoD doesn’t look for the latest and greatest they are generations behind the “speed” of consumer land

There were a few articles about 2-3 years back where the US Military specifically said they would start chasing the leading edge where viable, specifically using attritable drones and more compact electronics to offset power usage as examples.

Now that's not to say that everything in the military will be replaced by Intels or Ryzens and off-the-shelf memory kits, but that they're now considering more vetted "off-the-shelf" stuff on newer nodes if it proves viable for their use-cases, as opposed to sticking to older and slower legacy systems. That also included plans to talk about custom processors or accelerators based on existing stuff with US-based chip designers and fabs, kind of like how AMD does a lot of semi-custom designs for partners while leveraging their existing IP and integrating partner-specific elements into the design, as with Sony's PS5.

Yeah, I don't expect them to adopt <28nm nodes for everything, but they claim to want to be able to also keep up with peer rivals in the computer space as well as look into more efficient and modern ways to harden specialized chips on modern nodes, as well as make more use of off-the-shelf equipment to speed up tech capabilities while reducing cost.

That being said, it's kind of humorous to imagine AMD having a "Mil-Spec" Ryzen and Radeon line that focus more on reliability and stability than top speeds (like a militarized PRO line), or Intel sweet-talking the military into funding their GPUs by going the accelerator route, or NVIDIA actually taking up semi-custom designs proper and throwing super-stable RTX systems inside tanks (esp. now that many tanks are shifting towards digital displays for better protection).