Pat Gelsinger Disappointed with Sam Altman

[blueone]

This is critical imo. I seriously wonder if Intel have figured this out even today. I fear they still don't get this. Maybe Lip Bu does but do those at the levers of IFS?

If LBT hasn't set clear objectives for IFS, along with metrics and progress reviews, he wouldn't be a competent CEO. IFS is the most expensive investment Intel is making, by far, I suspect.

 

[tooLongInEDA]

What is ICF? What were these common mistakes?

ICF was Intel Custom Foundry - the previous iteration of Intel attempting to establish an external foundry business. Believe it ran from around 2010 to 2018. In its later/final stage it was using Intel 10nm (probably the most delayed process in Intel history). Believe they had a small number (likely less than 10) of large account engagements - but nothing of huge volume. Ericsson was apparently one. Some of these customers apparently got badly burnt by the ICF project failures. Altera may have been an ICF account before Intel acquired them. There's a SemiWiki discussion thread about ICF.

I'm fairly confident that LBT won't be repeating the ICF mistakes (though Pat may have done if left in charge).

 

[Daniel Nenni]

I'm fairly confident that LBT won't be repeating the ICF mistakes (though Pat may have done if left in charge).

One experience I had Lip-Bu and foundries was when he first joined Cadence. TSMC did not have a great relationship with Cadence at that time but Lip-Bu turned that around in a hurry as a customer and a partner. I saw Lip-Bu in Taiwan more than Silicon Valley. Having TSMC as a customer is big for EDA companies because TSMC's customers always ask what tools TSMC is using etc...

TSMC was using a Taiwanese company's layout and simulation tools (Springsoft) internally. Lip-Bu went in and gave TSMC a great deal on all of Cadence tools and the rest as they say is history.

It also helped that Synopsys bought Springsoft making an opening for Cadence with TMSC but Lip-Bu is the one who jumped in and made it a win-win partnership that has lasted for 20+ years. I was helping Tanner EDA at the time with their foundry relationships and I had hoped to replace Springsoft with Tanner EDA. Lip-Bu beat me like a drum on that one. Tanner EDA was then sold to Mentor Graphics so we did okay on that one.

 

[swka]

One experience I had Lip-Bu and foundries was when he first joined Cadence. TSMC did not have a great relationship with Cadence at that time but Lip-Bu turned that around in a hurry as a customer and a partner. I saw Lip-Bu in Taiwan more than Silicon Valley. Having TSMC as a customer is big for EDA companies because TSMC's customers always ask what tools TSMC is using etc...

TSMC was using a Taiwanese company's layout and simulation tools (Springsoft) internally. Lip-Bu went in and gave TSMC a great deal on all of Cadence tools and the rest as they say is history.

It also helped that Synopsys bought Springsoft making an opening for Cadence with TMSC but Lip-Bu is the one who jumped in and made it a win-win partnership that has lasted for 20+ years. I was helping Tanner EDA at the time with their foundry relationships and I had hoped to replace Springsoft with Tanner EDA. Lip-Bu beat me like a drum on that one. Tanner EDA was then sold to Mentor Graphics so we did okay on that one.

That was Laker from SpringSoft that gave cadence a lot of heart burn. Unfortunately, 6 months after Synopsys acquired Springsoft, cadence management felt a lot better about icfb/Virtuoso as there were a lot of challenges for Synopsys to consolidate the technologies and continue Laker momentum, to say the least

As for tsmc relation, I believe it was ChiPing at Cadence who drove the change, starting from handling skill PDK and OA. He is also one of the key personnel to bring Lip-Bu up to speed on EDA and come up with strategies. The other one is Charlie. But as they say, history is written by winners. LBT is given all the credits now

 

[KevinK]

Believe it ran from around 2010 to 2018. In its later/final stage it was using Intel 10nm (probably the most delayed process in Intel history). Believe they had a small number (likely less than 10) of large account engagements - but nothing of huge volume.

ICF customers started with 22nm and ran to 14nm, but only Achronix and Altera made to real production volume AFAIK. Only Altera taped out on 10nm andI believe that was only after Intel bought them.

Achronix Semiconductor became the very first ICF customer. It used Intel’s 22 nm Tri‑Gate process to fabricate its Speedster22i FPGAs — the first commercial chips ever built on Intel’s FinFET technology by an external company. Achronix cited Intel’s “end‑to‑end foundry services” for dramatically accelerating its product development, reporting finished silicon shipments by early 2013 and later moving to 14 nm designs .

Altera was by far the most prominent ICF partner. In 2013 it signed an exclusive agreement for Intel to manufacture its flagship Stratix 10 FPGA family on Intel’s 14 nm Tri‑Gate node. The two firms extended this partnership to include multi‑die packaging and heterogeneous integration in 2014. Intel later acquired Altera in 2015 for $16.7 billion, effectively internalizing its own marquee foundry customer .

Tabula used Intel’s 22 nm technology for its time‑folded programmable logic chips. Its CEO praised Intel’s “outstanding design‑for‑manufacturability collaboration” after achieving first‑silicon success in 2013. Although Tabula collapsed financially two years later, it demonstrated that Intel could successfully deliver complex fabless designs through external tool flows .

Netronome leveraged Intel’s 22 nm process for its network flow processors, becoming one of the first ASIC‑style ICF customers. Intel provided backend design services to produce fully functional 200 Gb/s silicon, illustrating flexibility beyond pure FPGA clients .

Panasonic announced a 14 nm Tri‑Gate process collaboration in 2014 to develop next‑generation system‑on‑chips, marking Intel’s entry into consumer SoC fabrication outside North America .

 

[swka]

Tabula used Intel’s 22 nm technology for its time‑folded programmable logic chips. Its CEO praised Intel’s “outstanding design‑for‑manufacturability collaboration” after achieving first‑silicon success in 2013. Although Tabula collapsed financially two years later, it demonstrated that Intel could successfully deliver complex fabless designs through external tool flows .

I had the "luck" to be at front row working out ICF process at Tabula. We knew we were one of the 1st ICF engagement, but it is only after the 1st meeting with technical team, we found out we were the 1st engagement, as they had little process or collaterals that are ready for external customers.

Tabula CEO's comment was a very political correct one. Truth is it was a very difficult process to work on. in addition to Intel's lack of experience supporting external customers, this is the 1st finfet process. coming up with all kind of design rules to "hide" finfet nature makes layout several time tougher.

good news is after so many years, intel foundry is much better

 

[Daniel Nenni]

That was Laker from SpringSoft that gave cadence a lot of heart burn. Unfortunately, 6 months after Synopsys acquired Springsoft, cadence management felt a lot better about icfb/Virtuoso as there were a lot of challenges for Synopsys to consolidate the technologies and continue Laker momentum, to say the least

As for tsmc relation, I believe it was ChiPing at Cadence who drove the change, starting from handling skill PDK and OA. He is also one of the key personnel to bring Lip-Bu up to speed on EDA and come up with strategies. The other one is Charlie. But as they say, history is written by winners. LBT is given all the credits now

You are right about Chiping, I worked with him at Avanti. Chiping did not negotiate the deal. Charlie Huang was involved as well. I know Charlie too. Let's say it was a team effort. Anything that goes wrong the CEO is blamed so the CEO should get credit too.

 

[blueone]

I learned a lot here about ICF. Thanks everyone who shared their knowledge.

 

[Xebec]

Personally, I can't imagine taking a Waymo. Trusting self-driving software? I don't think so. Yeah, I'm a dinosaur. I seldom engage my smart cruise control either.

FWIW -- Self-driving software (supervised) + driver paying attention is safer than one of these alone. I also find it very relaxing to use (both) when you're driving in an unknown area (new town or whatever). The car won't miss turns, allowing you to focus on the road and hazards exclusively. It's also just relaxing on highway drives to be able to look around a bit more than usual.

Even if 'unsupervised' isn't for you - if you ever get a chance to test drive a new Tesla for fun, try the FSD and work with it (rather than against it) - you'll find it pretty freeing . Kinda like the first time you drove a car instead of riding a horse..