Silicon Shortages & Intel's Future with Daniel Nenni - Telegrams for Guests

Hey Tom and Daniel Do you believe that HSMC will ever catch up to TSMC? And what do you know about the inner workings of HSMC and how closely they work with the CHINESE government?

yoda king

Hello Tom and Daniel, In a recent interview, Jim Keller said that ISA doesn't really matter anymore. And given the hoops companies have to jump through to migrate from intel to AMD platform, why would anyone choose RISC-V over x86 or even ARM? Is RISC-V really that big of a deal? We've been hyping it up before Nvidia announced the ARM merger.

Will Global foundries will eventually make their 7nm wafers? Thoughts on the rumor Intel buying Global foundries?

Benjamin Cannon

Thank you for coming on Daniel Nenni! Love your book. Do you think any of the other Fab makers will catch up to ASML with the ability of creating EUV machines? Or is ASML a monopoly in bleeding edge Semiconductor lithography?

Benjamin Cannon

Hello Tom and Daniel, Do you think we will see more high-performance chip designs adopting fan-out packaging? Traditionally, we use 2.5D packaging for multi-die design, but the cost of the interposer is a big burden for implementing such design in consumer products. Recently, we saw Tesla's dojo DPU using TSMC's InFO-SoW instead of active interposer for combining multiple chips on the same package. Can fan-out packaging (partially) replace 2.5D in the future?

Hello. Could you explain to the wider audience, the Arm-China and Mr Wu shenanigans? I've read an article explaining it on SemiAnalysis but I was wondering if you provide any additional colour on this topic Thanks, as always.

AC_666__

Hi Tom and Daniel, thanks for coming on again! It looks like Intel mostly fixed their abhorrent power usage with alder lake actually via uarch changes and not through a new process. This is a factor that gives me confidence in AL being a solid comeback product. Still they only really make it by using smaller cores so imho it seems AMD still has a better designed core+infrastructure regarding energy efficiency for high performance cores at TSMC 7N or equivalent node. And with Zen4 and Raptor lake I would expect this be still be the case, would you agree? Daniel, how do you expect Alder Lake will stack up against the M1 and the upcoming high performance laptop chip („M1X“) from Apple? Not hating on Intel, just saying their competitors are really good atm and have a leg up. Thanks Tom!

dj5k

In 2015, Apple bought a fab in San Jose. Do we know if they are still using it? And just what can they do there? If they are using it to prototype their chips, how useful is such prototyping when it means they need to design and tape out that chip for both the process in this San Jose fab, and for the TSMC node for HVM?

KarbinCry

Hello Tom and Dan! Umm, I mean Daniel. I realized the other day that we are not too far off from TSMC's upgraded 3nm with GAA FET. We generally know what to expect from foundries when they introduce a new node, but not so for new FETs, and I'm quite curious. My question is: what can we expect out of these improvements that will differ from a normal node shrink?

Advanced packaging was the name of the game at hot chips this year so I have a question about it: While advanced packaging is enabling more complex designs and going beyond reticle limit, it appears to do little in terms of improving power efficiency. What routes do you see being taken in the next decade or so to drive efficiency, if any, and do you think that a paradigm shift like processing in/near memory will play a role? On a side note, wanted to thank you guys for getting the semiwiki podcast on spotify, it's made keeping up on the great interviews you guys have much easier.

qhfreddy

Welcome back to the show Daniel. Recently I was doing some deep thinking on what has been going on with the whole supply/demand issue. As I was doing so, the interview of AMD's CFO Devinder Kumar with Ross Seymore of Deutsche Bank came up in my feed a few days ago at the time of me writing this question. I found it curious how out of all of the major semi-conductor manufactures AMD is the one who has been the most confident about getting supply issues resolved. I am aware that they have been doing a ton of behind-the-scenes work to improve things like ATMP (Something like 50-60% growth), helping TSMC and Global Foundries secure more silicon substrate supplier agreements and so on. My question is: AMD's role in the industry, is it growing more than originally expected? How far is their influence expanding?

MelodicWarrior

Hey Daniel, there have been theories that AMD is looking to be bought by another company, possibly TSMC. Do you think this would be a good idea for TSMC and how do you think this would effect both AMD's and TSMC's line up. Are there any companies you think would be better to aquire AMD? Also do you think AMD getting acquired has any chance of happening?

Screw Dropper

Greetings gentlemen, my question is about die sizes and it appears that silicon as a thing may become extinct as nodes get smaller. What are the future materials that will be utilized instead of silicon. Will those materials be as reliable and honestly what about efficiency and heat as obviously traces will be so close together.

Dr Forbin

Hello! My question is regarding Apple's current performance lead on phone/tablet market and soon the desktop/laptop too...do you guys think any cpu maker has a chance to actually outperform Apple in any market? It seems almost impossible as they'll always be on leading edge node and have all the money to make the best cpu possible.

Falto

Hi guys, what a treat to have Mr. Nenni back! As the M1 came out since he was last on, I'd love to hear his thoughts on how he currently sees Apple and their place in the market going forward, as well as if he makes anything of the N3 love triangle with Intel and TSMC that's allegedly on the horizon. Thanks for your time, and take care!

Tick Dickler

Look up Sam Zeloof. He's done his own um transistor fab work in his garage.

AC_666__

I have global market question: we know demand has spiked overall, but what relative trends each continent has vs each other? Is some area stagnant, still stable and is some continent, geographics are skyrocketing even vs US/europe? Do these have effect in various node markets catering different product segments, industries and price tiers (ie could south america mostly be demanding something other than TSMC 5/7nm)? How about mobile phone centric areas like central africa, south east asia?

Timo H

HI Daniel Nenni, let's say that I want to make a fab in my garage based on EBL (electron beam lithography). And let's say that I can make a electron beam 5nm in diameter and with 1400w of power .Is it possible to make my own silicone and if yes then how low I can go(in terms of nm-s not quality ) and what other machinery will I need (like spin coaters , magnetron sputtering machine etc etc).I can produce my own machinery and equipment so this question is serious and not a theoretical one .

Valko Milev

Hi Tom & Daniel! Miners are obviously here to stay: with billions invested and an entire ecosystem of enterprise and home users amassed, it's safe to say that mining demand is not going anywhere. In fact, it's so big a market now that perhaps even Ethereum switching off the mining model is not enough to cause the extinction of crypto-coin miners. We're also witnessing never-before-seen recovery from bear markets now with the pricing swing over the last couple of months, and though no one in the supply chain (barring the helpless gaming consumer) likes to admit how closely tied to the crypto markets GPUs have become, they clearly have as the crash we witnessed a few months ago led to almost instant discounts and sales with the 3060Ti suddenly hitting sub $500 price-points in even my region, which is one of the smaller ones. Granted, they were the crappiest models and the pricing wasn't great, but hey they were that low in pricing for the first time since launch here, and they haven't been that low since. Given this, how does the industry intend to satisfy this new found demand going forward? Does it perhaps not intend to satisfy it at all, perhaps profiting from this new found GPU gold rush-esque thirst for as long as market forces allow for while refusing the admit it all along? Or will production start ramping up to eventually normalize prices and supply to at least sane levels despite heightened demand? Even the silicon firms initially denied admitting the massive impact crypto had to their bottom-lines, but do you see them becoming more accepting of it now and going forward? Lastly, do you believe that a single crypto coin, even one as influencial as Ethereum, can cause the mining market to crumble or will there always be a new profitable coin to mine? Thank you for all the amazing content and cheers!

Abheek Gulati

Excellent guest, very excited!

Abheek Gulati

Hi Mr Nenni. It's really awesome to have you back! In your last visit, you mention growing up as a Wafer Jockey. Would you care to elaborate more on that journey in your career as you went from fab worker to EDA monkey? :D This EDA field is something I know very little of, other than going through basic VDHL/Verilog and having seen the recent Synopsis announcements on their tools just blew me away.

AC_666__

Hi Tom and Dan! On SemiWiki, you often mention the “silicon shortage” in air quotes, normally in the context of questioning the narrative from Intel, TSMC, Samsung, etc., that they must build more fabs (and thus need more government money) — do you still hold that the silicon shortage is overblown and that your predicted “silicon glut” will come to fruition in 2023/2024? Is it really just substrate and component shortages at this point, and that new fabs are just chasing after timely “incentives”?

Daniel, could you explain how the new substrate contracts AMD and possibly others are making to prevent the current substrate bottleneck to reccur? Is it a significant departure from how business used to be done? Reference from DigiTimes: https://www.digitimes.com/news/a20210503PD216.html

KarbinCry

How would you evaluate the current state of the two big pending mergers/acquisitions - Nvidia/ARM and AMD/Xilinx?

KarbinCry

Do you have any information on the timeline for Foveros Direct - hybrid bonding with 9um pitch? So far, TSMC is ahead in this area with V-cache Zen3 products that use hybrid bonding expected to reach HVM before the end of the year. We however don't know how far ahead TSMC is because we don't know the timeline for Intel's and Samsung's hybrid bonding. What timetables do you expect there?

KarbinCry

Hello Tom and Daniel My question is about the future of silicon and scaling performance in the next 5 to 10 years (after 2025). We`ve seen clever solutions like chiplets, Big.Little, 3d stacking cache and other types of dies and even Cerebras Wafer Scale. But beyond that and with the physical limits of silicon inching ever closer, what do see as being a long term way of continuing to scale performance? Maybe a refined version of the methods cited above or is it something revolutionary far out on the horizon like new materials?

Matheus Duque

Seeing Intel use TSMC more and for product categories that need the best performance, do you think Intel hard design teams have as much access and cooperation from TSMC as AMD, Nvidia or Xilinx would have? Given that Intel is an IDM, and is seeking to be one a direct competitor with IFS. I'd it possible Intel will always have this "handicap" when using TSMC nodes?

KarbinCry

Most people don't really have a need for drastically more cores/threads than we're getting now (although the associated IPC gains are always nice), moving up to denser nodes, what do you think is the most interesting way to use the additional die space? More cache? Machine learning accelerators? GPU cores? Do you see architectures splitting off more drastically for laptop vs desktop (i.e instead of being monolithic vs chiplet, laptop looks COMPLETELY different from it's desktop conterpart)

B. Fish

As a EE in Automotive Style Electronics do you have any idea when shortages in automotive will let up? Seems like multiple MFGs are having supply chain issues with not only Silicon, but mold compound, lead frames, and just back end packaging throughput. Do you think the unprecedented demand is hoarding and price gouging or real? Its been suggested to me from multiple MFGs that there is a lot of hoarding going around.

Do you see signs of Intel retooling their older nodes for foundry services? Or does it seem only new nodes will be designed with 3rd party users in mind? If Intel does not retool their old nodes, can they truly succeed as a foundry? And how likely is it that Intel buys an established legacy foundry (GloFo, Tower, Vanguard, UMC...) or some fabs and node IP from one?

KarbinCry

Daniel, could you talk about the differences between stuff like IBM 2nm and actual HVM ready nodes? When IBM showcased the 2nm wafer, a lot of people got confused why IBM is so far "ahead" of other fabs (not to even mention IBM isn't a foundry anymore).

KarbinCry

Thanks Daniel for coming back for more abuse! A question has been bouncing around in my head for a while now. How can you look at specs for upcoming silicon and "see" that it "should" do "4.3 ghz with ipc of xxx?" for example. Probably getting into the weeds with some engineering stuff but as a layman it always interests me when I see someone look at an upcoming product, sometimes years away, and be able to predict what kind of performance we can expect

Hey Tom and Dan (see what I did there? Lol) I am absolutely excited for the future of the I5 and I7 for both Alder Lake and Raptor Lake. I know AMD top end cpu will compete with the I9 but I can't see the Ryzen 7 or Ryzen 5 competing with the I7 and I5 for short term future. How will 8 and 6 lone cores compete for 8+4, 6+4 on Alder Lake and 8+8 and 6+8 on Raptor Lake. Do you see AMD rebranding the future Zen 4 12 core to a ryzen 7 and the 8 core to ryzen 5 to compete or slash prices. I have a i9 9900k in my system now and don't see myself upgrading for Alder Lake but I'm definitely keeping an eye on the 13900k, if that thing has the performance and cores you rumored it to have I'm definitely buy it. Thanks again for all that you guys do. 2022 is going to be insane.

Kinihun25

Hey Daniel I would like to ask about ASML. The industry is increasingly dependant on EUV machines. How is ASML going to scale up in upcoming years to meet the demand from its customers and is it even possible? Also will Intel really be able to outcompete TSMC with new generation of EUV machines?

QuickJumper

Good day Tom and Daniel! This is a real technical question so I apologise if we lose some listeners here but, silicon shortages, when can my brah get back to PWN'n n00bz? Do your best. Thank you.

Go-Kart