News Intel launches new 18A website, highlights milestones and specifications

[Admin]

Intel tries to improve 18A image ahead of volume launch later this year.

Intel launches new 18A website, highlights milestones and specifications : Read more

 

[bit_user]

The new production node will also be the company's first leading-edge process technology that is compatible with industry-standard electronic design automation (EDA) tools as well as IPs from third-party providers and will be available for Intel Foundry external customers.

Was that not already true of Intel 3? They definitely had announced external customers on Intel 3.

When Intel's fabrication technologies were years ahead of the industry — which was just about a decade ago — Intel would always release details about its next-generation manufacturing technologies together with actual product launches.

No, the distinction you're missing is that they didn't have a foundry business, back then. When you do contract manufacturing, the manufacturing process becomes a product in its own right. This is why TSMC and Samsung publish their process roadmaps & updates: in order to attract customers. A part of it is investor relations, but it's mainly about pitching their technology in a bid to win new manufacturing contracts.

 

[usertests]

LOl they are betting the farm on their N2 competitor.

 

[bit_user]

LOl they are betting the farm on their N2 competitor.

They have to. It's make-or-break, for them. If 18A is a dud or simply fails to attract enough customers, there might not be another chance for them to prove that IFS can be a viable TSMC competitor.

 

[Alvar "Miles" Udell]

Even if the tech is superior, they're going to have to show they're both as reliable as TSMC as well as have the capacity to fill orders as fast as TSMC.

 

[jheithaus]

They also have to be more reasonable when you call IFS. The pricing can’t be 2-3x TSMC.

 

[paviko]

Why so much pressure to serve other customers? Intel is producing so many it's own CPU the capacity for 18A will not be enough for themselves?
We also need to remember that TSMC 2 will be exclusive almost for 2 years for Apple. Who is now using "old" TSMC 3 - only Apple and Intel? Intel will have 2 years after introduction of TSMC 2 to catch up. 18A needs only to be as good as TSMC 3.

 

[bit_user]

Why so much pressure to serve other customers? Intel is producing so many it's own CPU the capacity for 18A will not be enough for themselves?

Semiconductor R&D and fab buildout costs have outstripped what Intel can bankroll from its own products. They need more volume, or else they'd have to get out of the manufacturing business altogether.

We also need to remember that TSMC 2 will be exclusive almost for 2 years for Apple.

Source?

Who is now using "old" TSMC 3 - only Apple and Intel?

AMD is using it for Zen 5C chiplets, since last year. I'm not sure who else is using it, but I'm sure they have other customers. Qualcomm is going to use their N3P node:

https://www.notebookcheck.net/TSMC-...ontract-as-Samsung-loses-ground.938739.0.html
​

 

[phead128]

They sent engineering samples to dozens of potential customers over 6 months ago. This website is mostly for PR.

 

[Gururu]

For the life of me could not find the link. Here
They will have or put up (and by that I mean knock our socks off) or shut up

 

[thestryker]

Why so much pressure to serve other customers? Intel is producing so many it's own CPU the capacity for 18A will not be enough for themselves?

Intel's own products will likely fill up most of the initial capacity, but they're bringing an entire fab online in AZ for 18A. I don't think it's possible to overstate how much damage the delay in acquiring EUV machines and problems with 10nm cost Intel.

The IFS pivot was a smart reaction not only to the cost of new nodes, but it allows Intel to shift the way they do fabs. In the past they would retire old nodes as a way to maximize fab efficiency. This way they weren't constantly building fabs and/or having idle fabs. As we've seen with TSMC there's a lot of benefit to having multiple EUV nodes available since this allows for more optimal cost/benefit node choices.

Intel's fab expansion if executed correctly (as we'll find out by the end of the year) should allow them to open up 18A and then later 3 while developing more advanced versions of their nodes. This is also what should allow them to pay for the High-NA gamble and move forward to 14A.

We also need to remember that TSMC 2 will be exclusive almost for 2 years for Apple.

Apple has not actually bought the first run of N2 unlike prior generations. I don't think anyone knows what their exact strategy is this time, but seeing as they bought the entire first wafer runs of N7/5/3 it's a pretty big shift.

Intel will have 2 years after introduction of TSMC 2 to catch up. 18A needs only to be as good as TSMC 3.

This is a very bad take. If 18A with GAAFET and BSPDN is only as good as any of the N3 nodes then Intel will have failed. 18A absolutely has to compete with/be better than N2 in at the very least high performance. They have to be delivering something that nobody else can to get the big contracts that they need.

 

[jackt]

I will believe it when ill see a chip on the market.

 

[Mr Majestyk]

I will believe it when ill see a chip on the market.

So you think Panther Lake will be cancelled despite being almost ready for Q3 launch? Panther Lake has as much chance of being cancelled as Arrow Lake had, and it's too late to redesign the chip using TSMC n3e(p).

 

[endocine]

Intel 18A, now featuring 2 times the power draw, with +3/-13% performance "bump" and guaranteed to degrade in a year.

 

[bit_user]

So you think Panther Lake will be cancelled despite being almost ready for Q3 launch?

The question isn't necessarily one of whether it launches, but how good it is (and also how profitable for Intel). Meteor Lake still launched, but its performance was rather underwhelming.

 

[Stomx]

Intel 18A, now featuring 2 times the power draw, with +3/-13% performance "bump" and guaranteed to degrade in a year.

Crash landing of fake nanometers everybody expecting eventually will happen?

Imagine you just got the first sample of 18A with these parameters and walking to report it to CEO...
Imagine what kind of night sleep CEO will have...

 

[TerryLaze]

They have to be delivering something that nobody else can to get the big contracts that they need.

Not really, realistically they only need to achieve parity with what tsmc has, and not even 100% parity but just close enough.
Intel already proved that they can translate intel products to tsmc manufacturing with arrow lake which means they can do the other way around as well.
Companies going to intel as a 'second source' supplier to lower the danger of delays if not outright losing shipments or just to get more wafers quicker is all intel needs to at least make good business.
Every big release in the last years has been extremely limited because they just can't get enough wafers.
Being even better would just allow them to charge more so of course they would want that but it's not really needed.

 

[thestryker]

Not really, realistically they only need to achieve parity with what tsmc has, and not even 100% parity but just close enough.

This is flat out wrong for leading edge because none of the foundries can afford to undercut enough. You can see how this has played out with Samsung vs TSMC and how many big orders Samsung has gotten on their EUV nodes (it's zero outside of in house).

Intel already proved that they can translate intel products to tsmc manufacturing with arrow lake which means they can do the other way around as well.

Lion Cove and Skymont were already on N3B with LNL which means a lot of the work was already done when they switched ARL from 20A to N3B.

Companies going to intel as a 'second source' supplier to lower the danger of delays if not outright losing shipments or just to get more wafers quicker is all intel needs to at least make good business.

I'd love to know how you think the economics of this works out with a leading edge node. Can you find a single example of a company doing this on EUV nodes?

Also BSPDN is an optional feature of 18A, but I don't believe GAAFET is which means extra design work on top of the existing to go between nodes.

Every big release in the last years has been extremely limited because they just can't get enough wafers.

This isn't really accurate* outside of Intel's MTL launch, but that is directly related to amount of available EUV machines. Leading edge nodes cost a lot and in general time between releases are going up so it seems pretty apparent that several companies are being conservative with wafer buys.

*The crypto/COVID buying spikes shouldn't be considered in a long term business conversation in my opinion.

Being even better would just allow them to charge more so of course they would want that but it's not really needed.

It's not about charging more it's about getting big enough contracts to keep the fabs running. If they're not offering something the companies cannot get somewhere else then there's little reason to go with Intel yet. If they nail the execution of 18A then they can absolutely leverage execution and don't have to worry as much about competition comparison.

Intel 3 appears to be a good node, but it's also only been used for in house Intel parts (GNR/SRF/MTL refresh). So while this says Intel can execute with an EUV node it says nothing about their ability to execute an external design at volume.

 

[johnnydollar]

"is is" ??? In the age of AI, we still can't handle grammer : (

 

[TerryLaze]

This isn't really accurate* outside of Intel's MTL launch, but that is directly related to amount of available EUV machines. Leading edge nodes cost a lot and in general time between releases are going up so it seems pretty apparent that several companies are being conservative with wafer buys.

Everything else I might believe since I don't know enough about it, but nvidia or sony not buying enough wafer because they are afraid that their GPUs or consoles won't sell?!?!? Come on now.

 

[spongiemaster]

This is flat out wrong for leading edge because none of the foundries can afford to undercut enough. You can see how this has played out with Samsung vs TSMC and how many big orders Samsung has gotten on their EUV nodes (it's zero outside of in house).

It might not be zero for much longer. It isn't all rainbows and unicorns for TSMC right now with their 2nm process. Nvidia sticking with 4n(5nm) instead of TSMC's 3nm for their current generation was pretty telling. TSMC is getting too expensive.

[News] TSMC to Lose 2nm Orders? NVIDIA and Qualcomm Reportedly Mull to Team up with Samsung | TrendForce News

With rumors indicating that Apple, one of TSMC’s most critical customers, may delay the adoption of the foundry giant’s 2nm process until 2026, a Comm...

www.trendforce.com

Intel should be able to beat Samsung, which is all they really need to do in the short term.

 

[thestryker]

Everything else I might believe since I don't know enough about it, but nvidia or sony not buying enough wafer because they are afraid that their GPUs or consoles won't sell?!?!? Come on now.

The PS5 is using N7 and it launched after N5 went volume. It also launched into the pandemic which drastically changed buying habits (who knows what effect the crypto boom also had by way of people not being able to buy video cards). That means it undoubtedly blew past their sales expectations which means they'd have to add in a new wafer buy which wouldn't be cheap since everyone was buying a maxing out fab utilization rates. Given the margins that Sony makes on the consoles having an elevated cost wafer buy could wipe them out entirely and require selling at a loss.

As for nvidia the 30 series launched into a crypto boom (then pandemic) and they shifted focus to an extent making more of the bigger chips. The 40 series didn't really have availability problems aside from launch which has been standard for video cards for a long time. With the 50 series who knows what's driving the availability issues since nobody knows how many chips nvidia is actually having produced. While they don't use the same process it's still possible the Blackwell respin caused issues with consumer chip availability. Keep in mind nvidia also has to sell these cards for 2+ years which means they need to be careful to keep margins/price high.

 

[thestryker]

It might not be zero for much longer. It isn't all rainbows and unicorns for TSMC right now with their 2nm process. Nvidia sticking with 4n(5nm) instead of TSMC's 3nm for their current generation was pretty telling. TSMC is getting too expensive.

[News] TSMC to Lose 2nm Orders? NVIDIA and Qualcomm Reportedly Mull to Team up with Samsung | TrendForce News

With rumors indicating that Apple, one of TSMC’s most critical customers, may delay the adoption of the foundry giant’s 2nm process until 2026, a Comm...

www.trendforce.com

The biggest question is whether or not Intel/Samsung can compete on overall cost. While I'm sure TSMC would rather the big customers move on to the most advanced node so long as they're not losing that customer I'm not sure how concerned they are.

Intel should be able to beat Samsung, which is all they really need to do in the short term.

You can't play short term when it comes to foundry services: see the Intel 10nm debacle. The long game is the entire game, and TSMC being slightly more conservative than Samsung/Intel in recent times is why they're where they are process wise. This doesn't guarantee them victory of course, but Intel needs to be pushing to be at the top of the stack.

 

[TerryLaze]

Given the margins that Sony makes on the consoles having an elevated cost wafer buy could wipe them out entirely and require selling at a loss.

Share the same amount of wafers between two makers and the end cost will be the same while you get the same amount of product in half the time...
That would help sony in every release to get more consoles into peoples hands.
The margins on the console are almost irrelevant since they make money from subscriptions and royalties on the games themselves.

Keep in mind nvidia also has to sell these cards for 2+ years which means they need to be careful to keep margins/price high.

How is that going to change either way?!
Making the GPUs at more than one place wouldn't have any impact on this.
Maybe if they get tsmc and intel into a bidding war they could increase their margins even.

The biggest question is whether or not Intel/Samsung can compete on overall cost. While I'm sure TSMC would rather the big customers move on to the most advanced node so long as they're not losing that customer I'm not sure how concerned they are.

Intel can fall back on the profits they make from their normal business, selling cpus gpus and so on, they can operate their fab business at cost for a long while before they even start having to be concerned.
Or as said often enough before, they can produce their own stuff at the fabs and sell them for normal profit.

You can't play short term when it comes to foundry services: see the Intel 10nm debacle.

The 10nm debacle produced them tens of billions more net income for several years each.
It would have been bad if they where a foundry business back then, at least for their customers that would have wanted to move on, but for intel it was a really good business move.

 

[thestryker]

Intel can fall back on the profits they make from their normal business, selling cpus gpus and so on, they can operate their fab business at cost for a long while before they even start having to be concerned.
Or as said often enough before, they can produce their own stuff at the fabs and sell them for normal profit.

Except that they cannot do that and this is exactly why their stock price is ~$25 right now. As soon as they were transparent about the cost of fab operation investors punished them and wiped out around half their valuation. While I'm sure we'd both agree this is stupid it is the reality.

The 10nm debacle produced them tens of billions more net income for several years each.
It would have been bad if they where a foundry business back then, at least for their customers that would have wanted to move on, but for intel it was a really good business move.

No, it was never a good business move because it was very short sighted and has cost them immense amounts of money. It also undermined future profits and allowed their competition to catch up. Had they executed 10nm (or only been ~1 year behind) then it would have paid off, but that isn't what happened.

Share the same amount of wafers between two makers and the end cost will be the same while you get the same amount of product in half the time...
That would help sony in every release to get more consoles into peoples hands.

You seem to be ignoring the reality that the PS5 launched into and the reason behind the shortages. Most console launches have had short term supply issues due to FOMO, but that's an expected part of the market. Sony sold a lot more consoles than they were expecting up front and their wafer buys are based on sales projections.

The margins on the console are almost irrelevant since they make money from subscriptions and royalties on the games themselves.

Completely false for the modern consoles. In the past they accepted selling at a loss because they could lower the cost of manufacture a lot over the course of a generation. This is no longer true because the more advanced nodes cost enough more to wipe out shrinkage benefits. This is why Sony has only shifted from N7 to N6 for the PS5 shrink whereas the PS5 Pro went with N5. They've also saved money by shrinking the heatsinks in the console.

How is that going to change either way?!
Making the GPUs at more than one place wouldn't have any impact on this.
Maybe if they get tsmc and intel into a bidding war they could increase their margins even.

This is basic supply and demand economics where they need to be mindful to not produce so much that they lower the value of their own products. They benefit from limited supply so long as the limits are not extreme because that keeps the value over time (look at 4090 prices for example). If they were to split their wafer buys between two foundries it's entirely possible they'd get a worse deal than just going with one not to mention the cost of developing for two nodes.

You seem to think that foundry capacity is the primary reason for shortages when there's no evidence this is the case outside of uncontrollable circumstances (ex: crypto boom/pandemic).