[31] Dr. Ann Kelleher, EVP Technology Development, Intel
HTML-код
- Опубликовано: 19 июн 2024
- Intel recently opened Fab 34 in Ireland for mass production of the Intel 4 process - that's the one with EUV in it. As part of the opening event, I had a chance to sit down with Dr. Ann Kelleher, the EVP of Technology Development, who is in charge of the research and deployment of all of Intel's new foundry technologies.
[00:00] Intro
[01:06] Has EUV development been as expected?
[01:49] Is Intel market-leading in EUV?
[02:14] Do cycle-time and cost come into node development?
[02:57] Intel 4 in Fab 34 - also Intel 3?
[03:09] What is the timeline for node transfer?
[04:10] How does that process transfer from Oregon?
[04:40] Was it seamless for Intel 4?
[05:11] Any changes in procedure for this transfer?
[05:50] How much of a tech transfer is moving the equipment?
[06:30] What makes Ireland a good place for Intel?
[07:31] How is Intel ensuring getting to market on time?
[08:41] Are the jumps from Intel 4 to Intel 3, or 20A to 18A, full nodes?
[09:29] Is Intel on track for 18A?
[10:32] What does a $1b pre-payment get you?
[10:55] Will Intel offer customized nodes?
[12:00] How are node focus points for GPUs different to CPUs?
[13:55] PowerVias - more than just frequency?
[14:41] What's the key takeaway from the Fab 34 opening?
[15:17] What if Pat wants another 5 nodes in 4 years?
-----------------------
Need POTATO merch? There's a chip for that!
merch.techtechpotato.com
more-moore.com : Sign up to the More Than Moore Newsletter
/ techtechpotato : Patreon gets you access to the TTP Discord server!
Follow Ian on Twitter at / iancutress
Follow TechTechPotato on Twitter at / techtechpotato
If you're in the market for something from Amazon, please use the following links. TTP may receive a commission if you purchase anything through these links.
Amazon USA : geni.us/AmazonUS-TTP
Amazon UK : geni.us/AmazonUK-TTP
Amazon CAN : geni.us/AmazonCAN-TTP
Amazon GER : geni.us/AmazonDE-TTP
Amazon Other : geni.us/TTPAmazonOther
Ending music: • An Jone - Night Run Away
-----------------------
Welcome to the TechTechPotato (c) Dr. Ian Cutress
Ramblings about things related to Technology from an analyst for More Than Moore
#techtechpotato #intel #interview
------------
More Than Moore, as with other research and analyst firms, provides or has provided paid research, analysis, advising, or consulting to many high-tech companies in the industry, which may include advertising on TTP. The companies that fall under this banner include AMD, Armari, Baidu, Facebook, IBM, Infineon, Intel, Lattice Semi, Linode, MediaTek, NordPass, ProteanTecs, Qualcomm, SiFive, Tenstorrent, TSMC. 
Наука
![[18] Dr. Ann Kelleher, Intel Process Nodes](http://i.ytimg.com/vi/QbjIylf4yao/mqdefault.jpg)
![[18] Dr. Ann Kelleher, Intel Process Nodes](/img/tr.png)
She is such a different upper managment person, so direct and structured. No corporate/marketing bs. Great to see engineers in a leadership positions.
You can say the same of Pat. You can tell the guy lives and breathes engineering excellence.
Ask an Irish person a question and you'll get a straight up zero bs answer. You may not like the answer so be careful of the question.
@@kojack57I was raised more or less the same, not in Ireland, but in Romania. I wanted to ask a restaurant manager in France something and I phrased it as I wanted to be asked, but my friend said it is a bit too direct and rephrased my question :)) Some get upsaid also if you are too direct when asking something.
You guys must be watching some other video as corporatespeak is the only thing coming out of her mouth. You can even see him getting flustered and trying to find some kind of actual answer out of her.
@@motionstruckbecause she also doesn't want to get severely reprimanded or fired.
When giving an interview to a journalist, she is "functionally" Intel. If she said something accidentally that wasn't publicly disclosed yet, media would run with "Intel confirms X!". She also doesn't want to name competitors by name either.
It's not PR speak. It's just someone who's camera shy, very technically savvy, and trying to only speak on matters that can be spoken about.
Always great to see interviews with Ann, she's so frank and doesnt overhype or try to spin things. Looks like Intel is still on track with their foundry, which is a great sign. This whole push for accelerated node development is exactly what Intel needed, both for the design/chip side and to make IFS work.
Never has a company been so excited to say that they are "on track". Nice words, let's see the chips.
Why do you think they are so confident? Seems like they have a secret sauce they know is a winner.
Could they be pulling wool over eyes?
This is an amazing interview! She is so down-to-earth and I really appreciate that. Thanks.
She's brilliant, great representative for Ireland in tech.
I really wish Intel makes it in time with their 4, 3, 1.8 nodes. We need competition on all fronts.
Products manufactured on Intel 4 are hitting store shelves in 3 weeks
Best spokesperson intel has on staff. Good interview. Cheers
I saw the thumbnail and immediately knew I have to watch, love dr. Kelleher, so much fun to listen to.
Great interview! You were able to ask interesting and pertinent questions, which is so much better than the typical fluff, but she actually responded with concise and informative details. She and Gelsinger make a great team.
Great interview! She answers straight to the point always
she is great
Another outstanding interview. Thank you to Dr. Kelleher!
I love the interviews you've done with Dr Kelleher and really hope you can do another after Granite Rapids/Sierra Forest launch as it seems like there would be more she'd be able to talk about then.
Brilliant interview, thanks!
Thanks Ian for these excellent interviews!
Thanks for the very classy interview and video.
I worked for Intel in Oregon in the early 1980s. We had those same wall colors, and at that time, even on the process equipment.
Back at Aloha, right? Were you guys doing DRAM while the original D1 in SC did logic? Or did you guys both do dram but staggered (for example you do 5 micron and I do 250nm)?
@@ab-lymphocite5464 Thanks for the response. I was in fab 4 doing production engineering on cvd and pecvd oxides and nitrides. It was quite a product mix but basically anything that could be made with single or double layer poly and single layer aluminum.
I am not sure if we were at 5 micron yet but there was a lot of 7 micron, 4 mhz in fab 4 using perkin Elmer scanners and not the advanced ones.
The product mix included 8086, 8087, 8049 / 8051, plus we were making mask roms for atari and govt, 16k drams, 16k eprom, and other stuff. I was fresh out of college as a chemical / materials engineer so I barely understood what the parts even did that we were making.
Yes you are right that fab 5 led the development and eventually parts were transferred to fab 4 and other sites to support more production.
Fab 5 was trying to bring up the 64k dram on steppers, but the design failed. Thanks to IBM for coming in and saving Intel from collapse. Their paternal support of the suppliers is the main reason that the industry exists today.
It wasn't the easiest place to work, but most of my career has been related to thin films, deposition, pre clean, metrology, curing, equipment, sales and management roles. So a good starting point.
@@ab-lymphocite5464 Were you at Intel as well in the past? Or the other suppliers?
@@harryniedecken5321 I currently work at RA
Ann is awesome, I really like how Intel's foundry approach is shaping up.
Fascinating interview, good to see a dedicated smart woman at the helm of the manufacturing. I hope to see good things from the upcoming nodes and products. Their competition is strong. Threadripper is strong. Apple Silicon is strong. Qualcomm Oryon is coming. We need strong Lakes and we need them NOW!
Great interview, Ian!
awesome interview!
So it this(FAB 34) Intels first or second EUV production facility??
First dedicated. Technically Oregon could produce volume, but it's still a testing facility to design the process.
I've heard her speak and introduced myself. She is very down to earth and kind. ♥I didn't realize she was so high on our food chain!
It's a shame that Intel had to drop ASML's high-NA scanners for 18A, I guess they will be included for 14A.
It is exactly this kind of decision making that will save Intel. High-NA would've caused more delays for Intel 18A.
@@EnochGitongaKimathi Yeah for sure, it's better to get something out than delay it until they could get high-NA working.
@@EnochGitongaKimathi18A is planned for H2 2024 Volume production
High NA scanners won't arrive until Q1 2024 and it takes to to install and calibrate equipment.
edit: it takes time to install equipment.
She is the best.
Love your channel Dr Cuttress, I'm happily subbed; but respectffully would like to beg you to decrease use of "sounds good" to the replies of these fantastic interviewees. It sounds like you are declaring they are just speaking PR talk instead of making genuine claims.
Sorry, I only do that when I'm tired. Went back to back to back to events, and was quite anxious that if something doesn't work, then the whole opportunity is lost. I'm more aware of it than you think
@@TechTechPotato I'm so honoured you replied! Thank you!! Made my day...
Anywho, your mannerisms are enjoyed by all of us, and bring a great positive energy and fun spirit to emerging technology (as opposed to the vapid MBA side of the tech-world). No apology needed; only wished for you to be respected as much as possible...
So it's interesting to see this after meteor lake has been released. Since it doesn't really live up to the hype, I wonder if it's a core design problem? Like, the packaging is great and it's really efficient when the CPU tile gets turned off... but the performance under load is not great compared to Phoenix, which is TSMC 4N. So if the node is really better or comparable, where does the disadvantage come from?
How were all the issues with 10nm resolved?
TSMC (half joking)
@@bmurph24 they were pretty smart then, and managed the ego, that's a indication that things will move smother then
@@tonynaranjo2982They didnt want to do it when Jim Keller suggested it though and thats why he left (at least thats what the news said)
She's everything she made great steps forward inside Intel
2024 is going to be pretty amazing.
She looks so sympathetic, big fan of Intel, they just work
She's very smart, I bet Engineers love having her (another Engineer) as a manager! She's also very sweet and kinda cute :D
That would be an ecumenical matter.
Hahaha. Father ted
💥
Is there a good write up of what went wrong at Intel for the last decade?
It was run by bean counters for too long. Sme thing happened previously at AMD. You might call it it "Corporation Syndrome". The original (failed) 10nm ask was too big a jump a decade back and they were stuck in development hell with 14nm for years. By that time AMD's foundry partner (Ironically which would have been AMD if they hadn't spun it out) had chucked in the towel and AMD got success with TSMC.
Intel aimed for the moon with it's advancements going into 10nm, and ran into countless issues. Meanwhile TSMC and Samsung moved to EUV at the time, which ended up being the best thing to do. AMD moved from GloFo to TSMC and rode that wave, while Intel was insistent on making 10nm work.
@@__aceofspades Thank you for explaining it better than I !
Pat better talk to me, now that's an Irish Lady's way of say you better not 🤣🤣
I have a question: How come everybody seems ok with ASML being a monopole for so many years?
All the major players were investors in ASML, almost to the same degree and quite substantially (10-15%+ each). It made ASML more like a collaborative conglomerate, than it's own thing. I think Intel, TSMC, and Samsung now all own less than 5% each of ASML, but they're still highly involved and invested.
Very informative, ty.@@TechTechPotato
@@TechTechPotato Sounds like ARM
I'm telling you they are working with Atomera. Just wait and see.
The end got me, lol
An upgrade to the equipment in Oregon, okay sounds good! WHAT red flag!!! new upgraded equipment is NOT the same as the Equipment in Oregon. potential for problems is clear.
Why i believe intel is so confident and why their customers are starting to line up for 18a?
First off look at Samsung and 3nm GAA. What did they use? FDsoi smart cut.
Very expensive process and their yield isnt great. Notice they dont have the customers lined up and pretty much at the point of ditching for 2nm.
What's very very important to control in these GAA architectures is RDF. If you cannot control that you're gonna have a ton of problems because one atom gets into the channel and its money flushed down the drain.
I believe i know what IP Intel will be using for RDF.
Its something i believe they have been working with for several years. Perfecting the process.
In the advanced nodes they use more epi.
So this is a way to control rdf in suitu vs fd soi which is a very expensive process.
This other process is much less.
"Oh, I know who that it is!" God. I am such a nerd.
He's spoken to her before.
Wasn't she a former deputy for Clearwater County?
But is 18a really 18a??? You know Intel 7 7nm.
Not how nodes work. What does "really" 7nm mean?
If the leading manufacturer (TSMC) calls a node with X characterists "7nm", then Intel is simply following the naming convention. What was previously planned to be called 10nm ESF had density and ppw characteristics on par with TSMC N7
@@moist_ointment Its so irritating... some reviewer learned something new when intel was stuck with the 10nm situation and now everyone thinks they found the holy grail on how to classify a chip...
@@moist_ointmentIMO Intel should have tacked on 'ne' or 'nanometer-equivalent' to their node names. So "Intel 7-ne" and "Intel 4-ne" etc... That might have helped the dumdums see beyond their ignorance of node sizes.
20A would have to be on par with N3P in PPA or Intel would have serious problems with TSMC's transition to gaafet
Id ratter say that TSMC is the one who has to worry ;)
These are some bogus and baseless claims by TSMC, TSMC has no idea how dense Intel 20a or 18a are, TSMC is entering a new era where it could lose a significant portion of its future revenue to Intel, if Intel meets those timelines, then TSMC would be a couple of years behind.
@@Capeau TSMC roadmap and execution have been flawless lately TSMC won the finfet era by being slow and steady and node improvements every thanks to Apple, if all goes to plan N2 would be competitive with 18A even without backside power delivery.
It's getting harder and harder to trust Intel but it seems like manufacturing is the most honest and on-track. I hope that's true!
A product using Intel 7 released in 2021, so 5 nodes in 4 years means a product using Intel 18A releases no later than 2025. I remain a bit doubtful
it might be available, but it might also be so expensive that it starts out only on server and mobile lines
@@sinephase If that happens I would consider that they achieved the goal. What I wouldn't say achieves the goal is saying 18A is "manufacturing ready" but no product releases in 2025
@@adonisds it will start mass production in H2 2024
Intel might says 5 nodes in 4 years. But as a consumer (intel classifies me as a consumer, customer is something else) - I am getting three years of the same node and similar architecture in a row. workstation seems to be really slow as well, since there is apparently another SPR revision with what is essentially Golden Cove with AVX512 and AMX etc?
Sure Meteor Lake is exciting but the NPU seems not at all powerful, just efficient. So it wouldn't work as a big accelerator for workstation, its GNA3.5 or a large VPUx. The real acceleration is from MTL tGPU.
Intel had Gaudi2 and PVC, but dont sell that for workstation.
So my question ends up being - if this was such a great development, when do consumers get the rewards of it? 3-5 years in the future? My CPU from 2015 is really end of life and I have a majoe disadvantage when competiting in CS2. Is right now (14th gen) a bad time to buy Intel? The only real positive note I heard is that's it's the most mature node they ever had with Intel7+++
why don't you make your company instead? It is not like Intel is borrowing readymade stuff like AMD does from TSMC. You guys are notoriously dumb
Meteor Lake seems to be the turning point node to buy in at for client CPUs. I think Intel 3 and 20A are a few years out before production still in terms of client.
ill split it and say 4 yrs
they said 5 nodes in 4 years, and they're serious: they have the nodes available. Problem is that the design teams are unable to get their designs working in short order, leading to this outcome.
@@Eromasin MTL is only mobile products. That's irrelevant for high performance, desktop and workstation.
Doesn't matter , meteor lake is barely better than raptor lake
Reference Dr Kelleher says everything has/is running smoothly and on time, in your own article Ian you had Intel 4 'ready' from middle of 2022 yet we are just, and i mean just, getting Intel 4 products at the very end of 2023!
But ok that's fine, nodes coming online and retail products hitting shelves are very different things but if you follow those timescales and Intel executed it's planned roadmap of Intel 20A being ready for the middle of 2023 like it says in your article then that means products like Arrow Lake won't be on the shelves until December 2025 and Intel 18A products would still be over 3 years away!!! So either something very much went wrong with Intel 4 or it's going to be a tough few years for the Intel family!
Sadly, I think the reality is that Intel needs to magically pull in 24 months of product development from nowhere and obviously that is just not possible. The Intel 4 (7nm) node is very late to the party, meaning Meteor Lake is much later than anyone ever expected it to be and to rub salt into the wound it's performance isn't going to make up for that either!
Coupled with the AXG team not being able to include Battlemage, let alone Celestial which was the very original plan, it's not looking good. So all Intel has now is to try to hope that the AI engine in Meteor Lake will be enough to carry this product over the line.
You should understand that Intel 4 node is not 7nm, If its is like you say, where does it come from, do you have process specifics? More then 10 years ago intel node did follow in nm dimension but after TSMC started to grow bigger in market share they have did already what intel is doing now, TMSC called there nodes 16nm, 14nm, 12nm, 10nm without the node actually being that dimension per se. So it was TSMC who started first with a market name to a node and since we have INTEL 7, 4, 3 18A nodes Intel's node are also name only by marketing name. You should take a look at lets say gate pitch length for each node today on the market and compare!
@@menger88p Yeah I was purely putting that in there for reference to Intel current 10nm (Intel 7), I totally agree the names are meaningless, could call them capital cities for all it's worth but originally Intel was calling these 10nm and 7nm.
It will be incredibly interesting when we get Arrow Lake (I think it was Arrow Lake) because my understanding is that Intel is using both 20A and TSMC's N3 for the CPU tiles, so we will be able to compare the nodes exactly.
@@Speak_Out_and_Remove_All_Doubtcalling Intel 7 "10nm" and Intel 4 "7nm" is odd.
Do you also refer to TSMC N6 as "7nm+" and TSMC N4 as "5nm+"?
Just use the proper names.
@@moist_ointmentIt's for historical reasons. Intel originally named them 10nm and 7nm, but halfway through development renamed the nodes for marketing "clarity" I guess you could say. Which is important context for those who followed node development before the rename. That's my 1.7331 (inflation adjusted) cents.
@Nanerbeet yeah, key word being "renamed", so referring to Intel 4 as "7nm" is either someone way behind on the latest info, or intentionally using the wrong name to cause confusion.
Maybe Nvidia could just buy Intel for fun, TSMC needs competition, hehehe. Nvidia just killing it with AI at scale.
you don't compare Nvidia to intel their business model is totally different.
@jmtake85 But in terms of valuation, every company is comparable based on its assets, cash flow, and potential earnings. That's why you can diversify uour portfolio across industries in the ends its about asset creation.
I'm pretty sure instead of stacking the pluses
Intel is now just stacking the As xD
As long as Intel runs business driven by their marketing team, intel will lag behind and become like HP
Intel is run by engineers now. Pat is an engineer and Ann is an engineer.
@@__aceofspades the culture begun with Grove, Otellini where they take a new product and ask marketing whether it is viable is embedded into Intel, they will never again experiment like they did with 4004 upto 8085, after that it has been Grove’s mentality of make more money with less, outsource everything. They will never experiment with RiscV just to see where makers take it. So now their only hope is Fabrication. They have given up lead again and again to outsourcing and marketing.
Thats exactly the big chance that everyone is talking about... try to pay attention pls
@@Capeau where do I pay attention? Even in the earnings interview Gehlsinger brushed off ARM saying it will never be mainstream, meaning he is not even attempting to have ARM and Risc V ready for Windows. His only hope seemed to be fabrication. If you want I can attach a link here.
@@climbeverest He is probablyright though... BTW intel owns a big chunk of ARM ;)
I'm sure the fab being in Ireland has nothing to do with any kind of quid pro quo for Ireland having been a nice little tax evasion hub for Intel.
On the other hand moores law is dead always talks shit about intel he is such a amd fan boy.
For Intel to regain its leadership on all fronts they realistically need to release Nova Lake next year on 14A, sadly we won't see this until 2027-2028!!!
AMD has so many different ways to beat Intel at moment, Zen 5 is basically ready to come out whenever AMD wants it to which will wipe the floor with RLR, they can use Zen 5c to destroy Arrow Lake multi-threading, whilst all options being much much cheaper to manufacture so they can win on value too if they want to.
AMD will almost certainly have Zen 6 ready for 2025 and Intel will still be on Arrow Lake and I could easily see AMD have the replacement for Zen 6 ready to go before Intel has a replacement for Arrow Lake!!!
Time to wake up dude, intel is back wether you like it or not.
@@Capeau I'm writing this on an Alder Lake system and just bought 14700K, I'm not a blind fan-boy of any company but my Intel systems use a lot more power than my AMD ones and most of the time I would much rather have 16 full cores than only 8 plus some questionable E-Cores.
I don't think anyone can honestly claim the Meteor Lake development has gone smoothly. Arrow Lake sounds good on paper but when will Intel get it out the door?! We've not had Intel 4 yet, let alone Intel 3 and Arrow Lake is the one after that so I'm not holding my breath!
Nope.
What?
@@HDRPC Dont mind him, hes one of those AMD kids that dont like reality
Hope she’s telling the truth. The CPU side has been all lies and delays and disappointments for so long. They are basically snake oil salesmen. Hoping the contract manufacturing side is honest and on time.
Dude, time to change the battery of your clock... its not 2018 anymore
Snake oil? huh?? Intel's products have flaws but they're certainly not snake oil lol, I can run my code just fine
@@kelownatechkid Snake oil salesmen as in their product marketing. If you don't know what I'm talking about, best not reply with such a stupid comment. Any cpu can run code just fine.
@@sflxn wrong! AMD Cpu Can't run Code,. and AMD's marketing term is snake oil salesmen, not intel, ask anyone in the market know-how pushy the marketing term is
@@jmtake85 Yawn. We all know about all the Intel's past fake benchmarking and 10 years of the same Haswell architecture rebadge. I've been following this industry for the last 30 years... back when I worked in it as an engineer in the semiconductor industry. Intel's marketing BS doesn't fool me.
nop nop, Intel lost tech/node/foundry many moons ago and i'm afraid they have a huge battle ahead and im not sure they will be able to turn the boat around... too little too late. This might be an Intel engineer (manager) interview but make no mistake, this was a heavily "redacted" and overviewed by Intel. I'm short
Why do I have the feeling you'd say that regardless of who was being interviewed. You have your mind made up already
@@TechTechPotato You are correct, i've had made my mind before even without seeing this interview. Seeing it just reinforced it more but let's hope im wrong because competition can only be good for consumers... take care
You just cant accept reality... too bad for you.
Everything in on track, but somehow, SOMEHOW, everything is going to be 2 years late. 18A in customer's hands in 2027 basically.
Intel still does not get it, ARM will take over.
People still do not get it, the ISA does not matter
Many people have claimed when M1 came out that now X86-64 is dead and will completely wiped out from market in 2022.
Many people said ARM will take 30 to 35% market share in all computer and laptops combined.
Now in 2023 ARM sales are going down. Arm was 14.5% and now ARM is 10.2% even without Meteor Lake, Arrow Lake, Nova Lake, Panther Lake and etc.
Now Intel process node and new architecture will start from 14th December 2023 to provide both great efficiency, battery and great performance.
The future is RISC-V 😂😂😂 the ISA matters
@@HDRPC Yes, the microarchitecture matters. Intel and especially AMD made a lot of progress since M1 was released.
But looks like Meteor Lake is quite disappointing that a lot of manufacturer reduced orders and shift over more to AMD, but not ARM ;)
@@EnochGitongaKimathi 🤭
This is bad, we WANT Intel to keep losing to AMD, at least until AMD reaches 50% marketshare across ALL x86 market segments...then Intel can fight back. From the view of us as consumers, the best we can hope for in a duopoly is a balance of power, a 50%/50% split in marketshare. This would bring about the most competition and innovatiom and all at the best prices....everyone in the PC community should be an AMD fan at this moment
FYI: If the roles were reversed, Id be cheering for Intel, because my ONLY allegiance is to what's best for me, the consumer.
Don't worry, they will.