Intel Eager to Avoid Clashes With Apple As It Looks to Secure TSMC's 3nm Chip Supply

[collin_]

Wait a minute. There was an article months ago saying that Intel had booked all of TSMC’s 3N capacity. Whatever happened to that…?

Also… an Intel GPU… what a waste of bleeding-edge silicon ?

 

[bodonnell202]

Where do we go we are just about of nano meters 3,2,1,0

Process nodes used to be measured in micro meters, but they made it down to 1 um in 1984, they can just start using pico meters instead.

 

[deconstruct60]

I'm always surprised at the level of hating on Intel whenever these stories pop up. Intel's only problem is that they made the near fatal mistake that almost all deep technology companies make at least once during their existence: Hiring an MBA to be the CEO (Bob Swan), as opposed to someone with a hard-core engineering background. Fortunately Intel has rectified that with Pat Gelsinger...

Errr. Intel started to stumble around 14nm. Brian Krzanich was in charge of fabs when 14nm got rolled out got named CEO. 10nm was in pathfinding mode when he got promoted.

The clusterf**k that is 10nm was on Kraznich's watch. He has a chemistry major in undergrad and was a process engineer. There were lots of folks blowing smoke and doing butt covering instead of solution finding going on.

Swan got the job in part because Kraznich also got into some HR trouble and got bounced. The board put Swan in to "manage" the company while they worked through the problems. Intel's board had already stumbled on being too focused on maintaining the stock price and dividend flow and blowing it on asking solid questions about what they were doing down in the fabs.

However, one of Swans screw ups was not ordering enough ASML EUV fab equipment as a "Scrooge McDuck" move to goose margins. Even if they didn't have their EUV fab process fully worked out , the machines are just plain hard to get ( have to put hard money down to be in line to get one). Fixing the fab clusterf**K was going to be expensive and making Wall St folks happy shouldn't have been a priority ( take the hit on stock price because it was coming anyway).

Intel doesn't even have a shot of catching up until get to the next gen EUV equipment inflection point. Until then will need to outsource a heft chunk of the GPU work and portions of others.

He probably didn't filter out the technical smoke blowers any faster than Kraznich did, but that wasn't really the main problem at that point.


There were multiple folks who messed up Intel and will take multiple folks to dig them out of the hole.

 

[cmaier]

Process nodes used to be measured in micro meters, but they made it down to 1 um in 1984, they can just start using pico meters instead.

No, they will use angstroms

 

[deconstruct60]

Wait a minute. There was an article months ago saying that Intel had booked all of TSMC’s 3N capacity. Whatever happened to that…?

One thing that happened since then is that TSMC has gotten more strict about "soft projections" ( reservations for wafers ) and put-your-money-where-you-mouth is order flow. with the chip shortage too many customers were trying to game the system to latch onto more capacity then they'd use. TSMC is in a better bargaining position now. Raised prices. Scheduling is more critical in a shortage so need more trusted coordination on orders.

Second, that story was spun as though Intel had bought up all of the N3 capacity for the entire service lifetime of process. That was nonsense. More likely was more so about who had how much of the very early ( late phases of "at risk" " and early volume) capacity. Given N3 is rolling out somewhat out of phase with the A-series lifecycle it was probably up in the air as to who else would be a major buyer early on.

Intel has a long game on GPUs on leading edge TSMC processes. And they have probably know for a relatively long time. Intel isn't stuck on a relatively arbitrary date ( 2-3 week of September) for product launches. Moore's law if about 18 months long... It isn't going to always sync up with a 12 month cycle over the long term. Intel probably didn't commit to early "placeholders" for N3. That wasn't necessarily indicative of actually buying it all up. Just that they were not going to sit back and wait for scraps to fall off the table.

Intel probably has both N3 GPU tiles for SoCs , compute tiles for discrete Xe-HPG 'gaming'/'mainstream' GPUs, and AI/ML/HPC compute tiles lined up for N3. A family of N3 GPUs which will be in different volumes at different times. Fitting that into the jigsaw of other companies demand bubbles is probably what Intel needs to work out, but the need placeholders in early because at least one of those GPU family products needed to roll out first.

 

[OnetyOne]

Most news sites:
"Intel keen on closing next-gen chip deal with TSMC"
"Intel To Finalize 3nm Deal With TSMC This Month As Pilot Production Kicks Off – Report"

MacRumors:
"Intel Eager to Avoid Clashes With Apple As It Looks to Secure TSMC's 3nm Chip Supply"

Most MacRumors comments:
"Apple good, Intel bad."

LOL.

 

[cmaier]

Most news sites:
"Intel keen on closing next-gen chip deal with TSMC"
"Intel To Finalize 3nm Deal With TSMC This Month As Pilot Production Kicks Off – Report"

MacRumors:
"Intel Eager to Avoid Clashes With Apple As It Looks to Secure TSMC's 3nm Chip Supply"

Most MacRumors comments:
"Apple good, Intel bad."

LOL.

Intel:
Jesus jesus blah blah convert or burn blah

 

[deconstruct60]

Intel needs to spin off its fabs if it wants a level playing field against Apple, Nvidia, AMD etc.

Right now , That would be an even bigger screw up that Global Foundries was for several years. Intel's fab business isn't really viable as a separate company until their Foundry Services business builds up to something like at least 25% workload coming from outside jobs.

This isn't like Coca Cola where can sell off the bottling companies and just sell syrup to them at high prices. The foundry has to be able to make money from other customers than Intel to be long term viable. AMD got locked into. "gotta buy from GF" contracts for far too long because otherwise GF didn't work. Probably could find some buyer with more money than sense but if they ran Intel fabs into the ground in several years then Intel would be between a rock and hard place.

There is no excess capacity in the system for Intel to move all of their production to TMSC and/or Samsung.

What Intel needs to do is stop half-stepping at the Foundry Services business and just do it. Wean things down from just having just "one customer" and some minor, lower priority side shows. They also need a wider family of fab processes with different tweak/objectives. Needs to be siloed enough to bring in substantive customer order flow ( possibly into a subsidiary )


Intel also needs to unwind a bit of trying to do everything for everybody in their own product spaces. They have spent money like drunken sailors on AI/ML processor solutions. Bought McAfee. MobileEye . Billions in stock buy backs to goose the price higher . A modest fraction of all that money could have been plowed into better, wider (incremental + "big bang" evolution ) fab process pathfinding and they wouldn't be in as big of a mess now.


If Intel Processor/SoC/FPGA sales shrink low enough ... then yeah they could spin out the fabs but as of now they have enough of a large/volume product kernel to try to do both. If sucessfully jumpstart as a 3rd viable dGPU vendor then they can offset some CPU volume for new product over the long term.

 

[DeepIn2U]

If they want a good relationship maybe they should halt the crass adverts.

Agreed.

But maybe Apple will feel nostalgic and once again burn Intel's bunny

 

[CarlJ]

This article reads very much to me like, "Intel has declared that it's asking for a real live unicorn for Christmas." TSMC has a fixed amount of production capacity, particularly for the more leading edge stuff. Apple has a lot of this sewed up tight, and is part of the reason that TSMC has that capacity in the first place. Intel is going to ask for a larger share of fixed production capacity. Who, precisely, is Intel suggesting that TSMC should break contracts with, to free up this capacity? Or is Intel suggesting that they'll funnel a fleet of boatloads of cash into TSMC to fund building some more fabs?

 

[CarlJ]

No, they will use angstroms

Hoping they eventually reach my favorite unit of measurement, the atto-angstrom.

 

[Unregistered 4U]

No, they will use angstroms

Do the numbers actually align to anything, though? Like, physically, is there anything you can examine and say “Yup, this is 5nm instead of 7nm for sure!”

 

[darngooddesign]

Why would TSMC prioritize Intel over Apple when Apple has a symbiotic relationship with them and is funding the new nodes, while Intel has openly stated their intentions to compete with TSMC and is only a short-term customer?

Intel wants to make sure that TMSC can produce for them in addition to Apple, not prioritized above Apple.

 

[cmaier]

Do the numbers actually align to anything, though? Like, physically, is there anything you can examine and say “Yup, this is 5nm instead of 7nm for sure!”

Supposed to roughly correspond to minimum feature dimension. Usually does. Nobody uses transistors with minimum feature length, though.

 

[rukia]

Before Apple, the critical partners for advanced nodes were the gate array companies. That's because gate arrays (although quite complex) are more regular and easier to characterize than SOCs and bleeding-edge processors.

Altera uses TSMC for advanced fab. And Intel purchased Altera a few years back. So Intel is already (via gate array production) a bleeding edge customer for TSMC.

BTW, gate array backlogs are one of big causes of the worldwide semiconductor shortage affecting all products.

FPGA's, along with SRAMs, were favored for pipe cleaning the process to bring down the defect density. They were also viable products due to the ability to use redundancy and repair allowing for some tolerance to the high initial defects. They were also products that could leverage the highest density possible. However, unlike SRAMs, only the largest FPGA's marketed for base stations could capture the margins required to justify the leading edge mask costs (AI co-processing has taken over the high margin category). However, there is a massive difference between the symbiotic relationship between Altera and TSMC compared to Apple. 1) Altera's volume was significantly lower than Apple's. 2) Altera was did not have Apple's deep pockets so tape-outs were much more limited. From TSMC's point of view Altera provided far less value for gathering data on yield improvements for their logic process which has a wide variety of BEOL and FEOL options. Altera's far lower volume than Apple's did not allow TSMC to scale up rapidly. Also note that TSMC is more than just process but also packaging. Altera never leveraged TSMC's packaging technology and used an external vendor for this. Apple not only uses the full stack but also takes an active part in the development process. The Apple / TSMC working relationship probably closer, but not identical, to Intel's CPU team and TMG or AMD with SDC and the Dresden fab (pre GF spinoff).

After Intel acquired Altera, except for the legacy FPGA lines, the new products were rapidly moved to Intel's fabs. Intel's process was cutting edge at that time and the convergence between TMG and Altera was the synergy that was supposed to unlock value. Even when Intel tried the foundry 1.0 model Altera was really the only customer that tried it out and the idea of the merger was already in the works. Altera, currently Intel PSG, no longer has the same working relationship with TSMC as a leading edge pipe cleaner. Apple is the one stop shop for that now.

 

[sideshowuniqueuser]

Great! I hope AMD, Intel, and nVidia blow Apple Silicone out of the water.

It has no effect on my equipment as I decided I would never go back to Windows, and it will only push Apple to work even harder with their chips.

At last, a comment that makes sense! People being Intel haters just because they like Apple products and Apple switched from Intel is just weird. Good competition is good for all of us, even including the Apple share holders, because complacency ends in doom. I really really hope that not just Intel, but all the other chip makers, absolutely kill it, and give us choices. Imagine if there was a computer out there just as good as, or even better, than Apple computers, how wonderful would that be.

 

[koil]

That's fantasy. Apple would not be able to buy Intel no matter how much they try, not to mention such an acquisition would never get approved anyway.
For example Nvidia's acquisition of ARM is practically blocked.

I am obviously talking about TSMC, not Intel.

 

[rukia]

I am obviously talking about TSMC, not Intel.

Trying to take direct control of a company that the world depends on is a non starter. Looking at you Jensen. What Apple could do, if they wanted mess with the industry, is to take control of Synopsys, Cadence, and ASML along with a few packaging and test houses. Most people don't realize how many fragile choke points there are in the semiconductor industry due to the capital requirements and niche expertise required. There's even a significant number of component suppliers and IP vendors that have no viable substitutes. Given how small and unknown some of these companies are they would not attract the scrutiny of the FTC. China is taking an even more stealthy approach and just poaching the talent at these companies.

 

[cmaier]

Trying to take direct control of a company that the world depends on is a non starter. Looking at you Jensen. What Apple could do, if they wanted mess with the industry, is to take control of Synopsys, Cadence, and ASML along with a few packaging and test houses. Most people don't realize how many fragile choke points there are in the semiconductor industry due to the capital requirements and niche expertise required. There's even a significant number of component suppliers and IP vendors that have no viable substitutes. Given how small and unknown some of these companies are they would not attract the scrutiny of the FTC.

Synopsys *and* cadence? No way they can be combined.

Not to mention that Apple probably doesn’t even use most of their tools, if they are using anything like the design methodologies that their team members were using at their former employers. Maybe cadence for routing and synopsys for Primetime, but most other stuff is in-house.

 

[rukia]

Synopsys *and* cadence? No way they can be combined.

Not to mention that Apple probably doesn’t even use most of their tools, if they are using anything like the design methodologies that their team members were using at their former employers. Maybe cadence for routing and synopsys for Primetime, but most other stuff is in-house.

The FE and BE flow is probably quite a bit different than the PA semi days (just a guess).

 

[cmaier]

The FE and BE flow is probably quite a bit different than the PA semi days (just a guess).

Still doubt they use a lot of synthesis and auto placement. Intrinsity and PA Semi, not to mention all the AMD folks now working there, wouldn’t put up with that.

 

[rukia]

Still doubt they use a lot of synthesis and auto placement. Intrinsity and PA Semi, not to mention all the AMD folks now working there, wouldn’t put up with that.

I don't believe they have ever embraced the AMD AXE / BAXE methodology, hand place and autoroute with tedious manual buffering. Placement QOR has improved quite a bit. Gate counts and schedule compression has also decreased the ROI.

 

[cmaier]

I don't believe they have ever embraced the AMD AXE / BAXE methodology, hand place and autoroute with tedious manual buffering. Placement QOR has improved quite a bit. Gate counts and schedule compression has also decreased the ROI.

buffering was automatic when I was there, so they must have backslid somewhere between ace and axe? A few years back, someone there hinted to me that they were using something like .place files, though nobody would ever tell me exactly what they are doing since they are so paranoid about security.

 

[BuddyTronic]

Intel is looking to develop a closer relationship with TSMC, Apple's largest and most important chip supplier, to avoid possible clashes with the Cupertino tech giant over TSMC's 3nm chip production, according to DigiTimes.

Yesterday, we reported on TSMC beginning pilot production of its 3nm process that will eventually be used in future Apple silicon Macs. Apple currently uses a 5nm process in its latest iPhones, Macs, and iPads.

Ahead of Apple's adoption of the smaller chip process in its products, Intel is now trying to establish a clear relationship with TSMC to ensure the Taiwan supplier will fulfill its orders for a rumored upcoming 3nm GPU, without conflicting with Apple's orders.

DigiTimes reports today that high-level executives from Intel plan to visit Taiwan and TSMC in mid-December to discuss 3nm chip production and production capacity. The report says that during their meeting, Intel will be "striving for more available 3nm process capacity at TSMC" and that "Intel is eyeing a closer tie with TSMC to avoid fighting with Apple for the available process capacity."

A report last month said that Intel would be looking to adopt TSMC's 3nm process for its upcoming Meteor Lake processors. Intel currently doesn't utilize any smaller processes, and handing off the job to a third-party, such as TSMC, is a way Intel can possibly catch up to Apple.

Apple is undertaking a two-year-long transition that will phase out Intel chips in its Mac lineup in favor of Apple silicon. The transition officially began in November 2020 with the M1 chip in the 13-inch MacBook Air, Mac mini, and MacBook Pro. Apple furthered its transition this year with the 24-inch iMac, followed more recently by the new 14-inch and 16-inch MacBook Pros.

The 27-inch iMac, the Mac Pro, and a high-end Mac mini remain the only Intel-based Macs in the lineup. Apple has described Intel's chips as "power-hungry processors" and has credited the efficiency and size of Apple silicon for some more radical Mac design changes, such as those seen in the most recent iMac.

Article Link: Intel Eager to Avoid Clashes With Apple As It Looks to Secure TSMC's 3nm Chip Supply

Intel and TSM are both building facilities in Arizona.

Intel gets rained on by tax dollar money from Biden.

Only TSM can make 3nm

Apple has already locked in on future TSM production

Apple and Bitmain are the biggest customers already now for years.

Nvidia, AMD, and Intel’s “best silicon” is ALREADY produced by TSM. But nobody notices?

Anyone wondering how come TSM stock has gone sideways for a year? When will people realize that TSM makes ALL the best silicon.

Yes, you can say that Nvidia and AMD etc does the software, but TSM has been playing way too nice. Jack that TSM stock up please.

75% of TSM output is allocated to Apple last time I checked. Wish I had the pie chart handy to check my memory.

 

[rukia]

buffering was automatic when I was there, so they must have backslid somewhere between ace and axe? A few years back, someone there hinted to me that they were using something like .place files, though nobody would ever tell me exactly what they are doing since they are so paranoid about security.

Sounds like they are still hand placing certain blocks. BP was always a good RLM for that. The CAD group was oversubscribed so the flows were very unstable. They were far more concerned with using hashes so engineers couldn't cheat on the checkpoints. When it comes to performance reviews engineers cheat Ironically, AMD's flow has been stolen by so many ex employees who went to startups. Why anyone would want to steal it is beyond me.