Intel is going to spin out their FPGA group

[blueone]

TSMC has many customers who are competing against each other. Looking back, why didn't Altera stay with TSMC?

Altera switched to Intel fabrication when Intel was ahead of TSMC.

 

[nghanayem]

One thing you should also consider is that AMD/Xilinx is wiping the mat with Intel/Altera. Even Lattice and Achronix are in the competitive mix. AMD really did a nice job on the Xilinx acquisition keeping the team separate and motivated. Intel seems to be focused on the Intel specific FPGA market and the rest was up for grabs. Maybe spinning it out will change that but it may be too late.

It would seem Altera agreed with you Dan. My understanding was Xilinx and Altera hadn't launched any of the cheaper stuff in years allowing lattice to sweep in to the underserved parts of the market. Given the launch of the low end Agilex stuff, the like 15 product launches in a year at the DCAI thing earlier this year, and the like 5 record quarters in a row it seems like the lesson has been learned. If this actually took marketshare on the high end from Xilinx or the low end from lattice is anyone's guess though.

I spent quite a bit of time in the FPGA business and it was a fierce battle in the trenches. Altera did well until Xilinx switched to TSMC at 28nm then Xilinx owned it. Altera switching to Intel was rough going and with AMD/Xilinx in the TSMC inner circle that is a serious advantage. For the FPGA business the first one to the new node has a big market share advantage and that has been Xilinx since 28nm, absolutely.

Do the FPGA guys still move to the newest nodes? I thought they were more conservative on the node front these days and just pushed packaging? Doing a quick look on the websites Intel's highend stuff (and I guess now the midrange stuff) seems to be on intel 7 and AMD's seems to be on 16FF and older. Not really sure why at least the high end parts aren't on N4P for the lower cost per FET/power in addition to their packaging technology of choice.

Edit did a little more digging Xilinx does have their top chip on N7.

TSMC has many customers who are competing against each other. Looking back, why didn't Altera stay with TSMC?

My understanding was Altera used to be TSMC's Apple and then they gave Xilinx equal treatment. If I was Altera at the time I also would have tried to hitch my wagon to the guy way ahead allowing me to still have the transistor count advantage and stick it to TSMC. Of course we all know how that worked out.

 

[blueone]

One thing you should also consider is that AMD/Xilinx is wiping the mat with Intel/Altera. Even Lattice and Achronix are in the competitive mix. AMD really did a nice job on the Xilinx acquisition keeping the team separate and motivated. Intel seems to be focused on the Intel specific FPGA market and the rest was up for grabs. Maybe spinning it out will change that but it may be too late.

IMO, and from personal experience, Intel has never absorbed acquisitions well. Frankly, I'm pleasantly surprised Mobileye and Altera survived as well as they did.

I spent quite a bit of time in the FPGA business and it was a fierce battle in the trenches. Altera did well until Xilinx switched to TSMC at 28nm then Xilinx owned it. Altera switching to Intel was rough going and with AMD/Xilinx in the TSMC inner circle that is a serious advantage. For the FPGA business the first one to the new node has a big market share advantage and that has been Xilinx since 28nm, absolutely.

I was surprised when Altera switched to Intel process. How could they not know Intel was not prepared to be a fab for anyone but CPUs and server and client chipsets.

 

[hist78]

Altera switched to Intel fabrication when Intel was ahead of TSMC.

If Intel is better in the FPGA manufacturing than TSMC then Xilinx should have switched to Intel instead of TSMC. I think something is missing there.

 

[blueone]

If Intel is better in the FPGA manufacturing than TSMC then Xilinx should have switched to Intel instead of TSMC. I think something is missing there.

Intel was not trying to be a foundry at the time. Altera's use of Intel's fabs was a special deal. Altera wanted the Intel 14nm process.

Intel to Fab Altera FPGAs with ARM IP

www.anandtech.com

 

[nghanayem]

If Intel is better in the FPGA manufacturing than TSMC then Xilinx should have switched to Intel instead of TSMC. I think something is missing there.

Presumably because Xilinx didn't care about process tech. TSMC was for most of it's existence leading UMC. Yet Xilinx only switched to TSMC at 28nm because UMC 28nm went gate first and didn't yield, making TSMC literally the only game in town. As for the Altera side, in spite of having like a year of TTM advantage from having TSMC as their foundry Altera was still number 2. Once TSMC gave Xilinx early access to their nodes Altera lost their one crutch for not losing more marketshare to Xilinx. Then you have BK spin a line to the tune of "with intel custom foundry you can design for 22nm right now and be a good year or two ahead of Xilinx getting to 16FF or you can design for the the big kahuna in our 14nm where we expect to widen the technology canyon that already separates the best and the rest.".

Long story short Xilinx was under no pressure to get on a more advanced node than Xilinx as they were doing just fine while being behind. Meanwhile Altera was desperate to not become irrelevant and hitched their wagon to intel as they probably thought that would be the only way they could see themselves ever out competing Xilinx

 

[blueone]

Presumably because Xilinx didn't care about process tech. TSMC was for most of it's existence leading UMC. Yet Xilinx only switched to TSMC at 28nm because UMC 28nm went gate first and didn't yield, making TSMC literally the only game in town. As for the Altera side, in spite of having like a year of TTM advantage from having TSMC as their foundry Altera was still number 2. Once TSMC gave Xilinx early access to their nodes Altera lost their one crutch for not losing more marketshare to Xilinx. Then you have BK spin a line to the tune of "with intel custom foundry you can design for 22nm right now and be a good year or two ahead of Xilinx getting to 16FF or you can design for the the big kahuna in our 14nm where we expect to widen the technology canyon that already separates the best and the rest.".

Long story short Xilinx was under no pressure to get on a more advanced node than Xilinx as they were doing just fine while being behind. Meanwhile Altera was desperate to not become irrelevant and hitched their wagon to intel as they probably thought that would be the only way they could see themselves ever out competing Xilinx

This was 2012-13, and Altera wanted the new Intel 14nm process (the so-called CPU "n" process) to give them a leg up. That was the deal made. There was considerable interest from Microsoft (comms protocol offload) and the fintech guys (algorithm offload for high-frequency trading) that ignited some excitement in FPGAs in Intel. Altera was invited to the party; I don't remember ever hearing that Xilinx was.

 

[MKWVentures]

Intel needs a lot of cash/capital to finance the new Intel that Pat Gelsinger is trying to build. Unfortunately Intel's revenue is shrinking along with the unfavorable semiconductor demand condition.

If product sales can't bring in cash that's large enough and fast enough, then Intel must borrow money and/or sell assets (spinoff and IPO). Government subsidy, timing, and size are not certain and they come with lots of requirements and obligations. Compare to government subsidies and borrowing money (in various forms), IPO is probably better for Intel to get more cash with relatively low cost and less negative impacts.

Pat Gelsinger may have those reasons you mentioned in mind. But placing Mrs. Sandra Rivera as the CEO of the new independent FPGA company shows Intel wants to build strong credibility into this new company at the very beginning. Consequently it can potentially help Intel to gain lots of cash through the IPO.

What I'm having trouble imagining is why Sandra Rivera, who currently has a huge position at Intel, would agree to take this far lesser role. I can only think of three potential reasons:

1. She wants to be CEO of a public company so much she's willing to sacrifice one of the chip industry's best jobs.
2. She can't stand working for Gelsinger and this is her ticket out with dignity.
3. She was in trouble at Intel, and this is a win-win for them to replace her and save face for her and Intel. (Firing one of the industry's most senior women wouldn't reflect well on a company that makes so many LinkedIn posts about the importance of diversity.)

it wasn't her choice. lets wait til we see what "stand alone business" means. Its not clear it is much different but lets see how bonuses, salary, financing work. If they do a IPO and Intel keeps 90% of the stock and the price goes up.... it is just a huge value adder for the company (like Mobileye).

 

[blueone]

it wasn't her choice.

You state that like you know for sure. Do you?

 

[MKWVentures]

You state that like you know for sure. Do you?

I dont know anything for sure LOL. Lets see how this all plays out. The idea of doing a spin off, then keeping control of the stock is a brilliant idea.... Intel is then holding all the cards.

Side note: what percentage of PSG silicon comes from Intel?

 

[blueone]

Side note: what percentage of PSG silicon comes from Intel?

In the future the roadmap presented last year said the chips would be mostly fab'd by Intel. I don't know the percentage for current sales.

Intel To Broaden FPGA Lineup And Make Them At Home - The Next Platform

Back in 2015, when Intel was flush with cash thanks to a near-monopoly from X86 datacenter compute, it shelled out an incredible $16.7 billion to acquire FPGA maker Altera because a few hyperscalers and cloud builders were monkeying around with offloading whole chunks of CPU compute to FPGAs to...

www.nextplatform.com

 

[hist78]

Presumably because Xilinx didn't care about process tech. TSMC was for most of it's existence leading UMC. Yet Xilinx only switched to TSMC at 28nm because UMC 28nm went gate first and didn't yield, making TSMC literally the only game in town. As for the Altera side, in spite of having like a year of TTM advantage from having TSMC as their foundry Altera was still number 2. Once TSMC gave Xilinx early access to their nodes Altera lost their one crutch for not losing more marketshare to Xilinx. Then you have BK spin a line to the tune of "with intel custom foundry you can design for 22nm right now and be a good year or two ahead of Xilinx getting to 16FF or you can design for the the big kahuna in our 14nm where we expect to widen the technology canyon that already separates the best and the rest.".

Long story short Xilinx was under no pressure to get on a more advanced node than Xilinx as they were doing just fine while being behind. Meanwhile Altera was desperate to not become irrelevant and hitched their wagon to intel as they probably thought that would be the only way they could see themselves ever out competing Xilinx

Can you please explain it a little bit more? Thanks.

"Long story short Xilinx was under no pressure to get on a more advanced node than Xilinx as they were doing just fine while being behind."

 

[Daniel Nenni]

Can you please explain it a little bit more? Thanks.

"Long story short Xilinx was under no pressure to get on a more advanced node than Xilinx as they were doing just fine while being behind."

Here is the real story:

Altera was like family to TSMC, they started together when TSMC was founded. TSMC used Altera to pipe clean new processes and it was a very intimate relationship. Xilinx was partnered with UMC (Xilinx had a whole floor at UMC HQ) but UMC had a horrible time with 40nm and Altera jumped ahead.

Xilinx moved to TSMC at 28nm and Morris Chang gave Xilinx an equal deal as with Altera. The Altera CEO took it personally and jumped to Intel 14nm. I was in Fab 12 when Morris did the announcement that Altera jumped. He said it was an experience that they (he) would learn from but they (he) did not.

TSMC signed with Apple at 20nm due to the nonsense with Samsung (IP theft, bad PDKs, etc...). Apple got a "most favored nation" deal with TSMC which pissed of QCOM, NVDA and others. At the time QCOM was TSMC's biggest customer so QCOM responded by moving more product to Samsung.

TSMC is the trusted foundry but Morris had a hard time with customer relationships. CC Wei is excellent with professional relationships (no BS) so TSMC is in a much stronger customer position now than ever. QCOM, NDVA, and all of the "whales" are closely aligned with TSMC.

I was still working with Altera when they jumped to Intel 14nm. It was funny. The PDK was incomplete and the DRC deck redacted. It was unusable. Intel was still learning how to be a foundry.

And remember, Altera had a close relationship with Intel early on and bought the Intel PLD business so they were not strangers.

 

[blueone]

I was still working with Altera when they jumped to Intel 14nm. It was funny. The PDK was incomplete and the DRC deck redacted. It was unusable. Intel was still learning how to be a foundry.

And remember, Altera had a close relationship with Intel early on and bought the Intel PLD business so they were not strangers.

I'm sure you're being sarcastic. Intel manufacturing clearly had no plan in place at the time to become anything like a foundry. The Altera deal was apparently thrust upon them, and they were unprepared, to say the least. But now at least I know why Altera attempted to use Intel process. Interesting background. But feels more like they shot themselves in the foot to spite TSMC.

 

[hist78]

Here is the real story:

Altera was like family to TSMC, they started together when TSMC was founded. TSMC used Altera to pipe clean new processes and it was a very intimate relationship. Xilinx was partnered with UMC (Xilinx had a whole floor at UMC HQ) but UMC had a horrible time with 40nm and Altera jumped ahead.

Xilinx moved to TSMC at 28nm and Morris Chang gave Xilinx an equal deal as with Altera. The Altera CEO took it personally and jumped to Intel 14nm. I was in Fab 12 when Morris did the announcement that Altera jumped. He said it was an experience that they (he) would learn from but they (he) did not.

TSMC signed with Apple at 20nm due to the nonsense with Samsung (IP theft, bad PDKs, etc...). Apple got a "most favored nation" deal with TSMC which pissed of QCOM, NVDA and others. At the time QCOM was TSMC's biggest customer so QCOM responded by moving more product to Samsung.

TSMC is the trusted foundry but Morris had a hard time with customer relationships. CC Wei is excellent with professional relationships (no BS) so TSMC is in a much stronger customer position now than ever. QCOM, NDVA, and all of the "whales" are closely aligned with TSMC.

I was still working with Altera when they jumped to Intel 14nm. It was funny. The PDK was incomplete and the DRC deck redacted. It was unusable. Intel was still learning how to be a foundry.

And remember, Altera had a close relationship with Intel early on and bought the Intel PLD business so they were not strangers.

It's a interesting story. TSMC started developing close relationships with Xilinx after this Xilinx switch in to TSMC/Altera switch out to Intel. Mr. Moshe N. Gavrielov, former President and CEO of Xilinx, is now a member of TSMC Board of Directors.

I'm wondering if Altera tried to keep a dual sourcing approach between TSMC and Intel to manage the risk. Was it because the limited volume or R&D resource that prevented Altera to do that?

 

[hist78]

I'm sure you're being sarcastic. Intel manufacturing clearly had no plan in place at the time to become anything like a foundry. The Altera deal was apparently thrust upon them, and they were unprepared, to say the least. But now at least I know why Altera attempted to use Intel process. Interesting background. But feels more like they shot themselves in the foot to spite TSMC.

I think about two years after Altera jumped from TSMC to Intel, Altera was acquired by Intel for $16.7 billion in 2015. Did the acquisition have any thing to do with the Altera switching from TSMC to Intel? Did then Altera's CEO have some strategy in mind when Altera decided to jump that can lead to the eventual Intel acquisition?

In 2017 Intel acquired Mobileye but let Mobileye stay with TSMC. It was a very different approach.

 

[Daniel Nenni]

I'm sure you're being sarcastic. Intel manufacturing clearly had no plan in place at the time to become anything like a foundry. The Altera deal was apparently thrust upon them, and they were unprepared, to say the least. But now at least I know why Altera attempted to use Intel process. Interesting background. But feels more like they shot themselves in the foot to spite TSMC.

Altera was acquired by Intel so there was a happy ending. Had Altera not been acquired they would have died a slow death because they could not compete with the TSMC Xilinx combo, my opinion.

Even though we are surrounded with brilliant people and amazing technology there are a lot of egos and things do get emotional so the laws of physics do not always apply.

 

[blueone]

Even though we are surrounded with brilliant people and amazing technology there are a lot of egos and things do get emotional so the laws of physics do not always apply.

Based on my experience, truer words have never been typed.

 

[Daniel Nenni]

It's a interesting story. TSMC started developing close relationships with Xilinx after this Xilinx switch in to TSMC/Altera switch out to Intel. Mr. Moshe N. Gavrielov, former President and CEO of Xilinx, is now a member of TSMC Board of Directors.

I'm wondering if Altera tried to keep a dual sourcing approach between TSMC and Intel to manage the risk. Was it because the limited volume or R&D resource that prevented Altera to do that?

No dual sourcing per say. Altera moved to Intel at 14nm with only legacy products at TSMC. FPGAs have a long life so there is probably still business at TSMC, especially at 28nm, 28 was a great node for FPGAs. Last I heard Xilinx is in production on TSMC N7 and Intel FPGAs are on Intel 10, a node behind. Maybe with the spin out Intel will catch up.

 

[VCT]

Looks like a hint for Intel future spin off Intel Foundry.