They sell the full article which is why they aren't naming it in the public article. I suspect what they're saying is that the company has invested so heavily in getting to tapeout on Intel that it's too late to move their design to TSMC. To do that they would basically have to tape out again which would take the best part of a year even if they could get TSMC to agree to produce the chip, by which point they'd have missed the market.
The fact that the article teases an industry going through change makes me think it's probably a company that produces things for 5G. My money would be on Nokia.
When a chip design is stable enough for manufacturing them. But chips usually have post si bugs and so the tape out is usually a back and forth process depending on how complex thechip is and how competent the design team was.
Tapeout refers to when the photomask is done, basically when you have completed the design, rendered it graphically and are ready to send to fabrication.
tldr this article is basically saying the 10nm is so screwed up they can't just adjust a few things, they need to fix a lot and the more changes you start having to make the more verification that needs to be done before things are signed off and ready for tapeout.
There might be some ARM customer who selected Intel as a foundry for their ARM design; the question is how much ARM would have to invest on such a customer.
I suppose the worst case would be an expensive R&D initiative (with ARM paying Intel for prototypes) about how to make ARM cores with Intel's 10nm process, with no usable result. Not likely to sink ARM.
Rolling things over in my head. Warning mere speculation.
ARM doesn't sell silicon, they license IP. However might be they're licensing IP blocks. Essentially a tested CPU layout you can more or less paste into a design. Think Lego's.
Spinning up a chip isn't as easy is uploading your Eagle CAD file to a Fab and punching in unit quantities. Each Fab's manufacturing process has certain design constrains that will most likely require some amount of custom design and layout. Certain designs might not even be possible on certain processes.
They certainly can go to TMSC or GF with their designs but it will take time (months or years) to validate the designs, and money to make it happen. And this doesn't even take into account Fab capacities. TMSC and GF might be committed to other firms and not have capacity for this new work.
The 5G rollout has already started and vendors should have been shipping their products by now. Not having anything for the first wave means you're going to lose most of the market.
Assuming they need $100K / year to sustain this business, that is two full time Writer, hosting, going to events etc. You will still need 100 paying subscribers.
I know Charlie has an record of breaking all sort of secrets from Intel ( and Nvidia ). But are there really 100 people paying for this?
> breaking all sort of secrets from Intel ( and Nvidia )
Seems like something hedge funds/private equity would pay for. The biggest edge those companies have over retail investors is access to knowledge that's not considered MNPI but is also context the general public doesn't really have, through news sources like this and also (perhaps most importantly) expert network companies.
Sidenote: if you could improve compliance or cheapen the cost of it for such firms, you could make oodles of money. Compliance is a massive, expensive risk center that the industry hates managing.
They do actually have a student subscription, which is a $100 per year. I bought it for a couple years. Well worth it, but since I don't invest in stocks the $1000 per year not really worth it just to satiate my curiosity.
I wouldn't be surprised if it were Nokia - just over 20B, desparately reliant on 5G. If they're using Intel silicon for their early 5G stuff and miss the boat they're screwed.
> The push for speed could keep Qualcomm ahead of Intel, its rival in handset baseband chips. Intel announced in November plans to support CDMA and gigabit speeds in its 2018 cellular modems and up to 1.6-Gbit/s downlinks in 2019.
Nokia sure makes a lot of sense. There's nothing left but the infrastructure biz, and they need a working 5G base system to succeed. If they blow the rollout, they're toast.
Oh God. Yes Nokia were deep in bed with Intel on 5G. How did I miss that? And that have a $20B+ Market Cap. I don't know about some places of EU and US. But Ericsson has been winning more 5G infrastructure project then Nokia in most places I know. Of course by winning I mean that is anything left from Huawei which may be looking at 50% market share in the next few years.
Ericsson were having a hard time competing, if Nokia screwed up may be a M&A with Ericsson makes sense.
Godsh this really is a big story if this turn out to be true.
Edit: Having some more thoughts, the delay still doesn't make Nokia bankrupt, because the initial 5G roll out are small in numbers. Compared to 3G / 4G recurring infrastructure and patents revenue. So I guess it is going to make an impact on Nokia, but nothing as bad as Charles has wrote. He is likely right that his sources suggest Nokia has nothing to deliver or sell in terms of 5G, but again if you know anything about business I doubt this will bring Nokia down.
What advantage would the 10nm process have for base stations?
Possibly in the micro-cell stuff for 5G, but none of this runs on battery. Established 16 or 28nm processes are cheap and run smoothly.
I read an interview with GlobalFoundries which talked about different process node characteristics. [1]
Who would really benefit from the 10nm process? Mobile SoCs, basebands/modems and AI chips. There has to be a high performance and low power requirement to justify being an early adopter of a new process.
Or intel aggressively misled fab customers about the progress of their 10nm efforts.
I'm not sold on it being Nokia, but from their 5g chipset page:
> ReefShark packs more functionality into 50 percent smaller hardware compared to products that use discrete components
> ReefShark chipset decreases mMIMO antenna size by half
>ReefShark cuts energy use by 85 percent, increasing deployment options
>Boosts the intelligence and performance of MIMO antennas
So density and power. This chipset would be around for a long time, it makes sense to be on the latest process.
TI isn't in the application processor business like before. They don't do cutting-edge like 10nm parts, and they are also well-diversified in other sectors. They'll sell 10,000 op-amps for every Sitara that goes out.
Well it can't be Motorola because they don't fab much of their own chip. The List of $20+B Market cap of possiblities, and excluding ones like Qualcomm, TI, Broadcom, Xilinx, AMD, STMicroelectronics as they are either too big or don't do any business with Intel, the last two are known Global Foundry's key customers, and Xilinx are TSMC based. What we have are:
Maxim
Skyworks
NXP
Analog Devices
Microchip
The problem is none of these companies require latest node for their product portfolio, and even if they do the product is likely not be substantial of their revenue, i.e it won't bring the company down as the article has suggested.
This is a tone argument, obviously. You'll have more access to a broader range of opinions if you eliminate fallacies like this from your thinking.
Consider, an expert becomes an expert through their passion and dedication to a topic. If you eliminate emotion from your resources, you eliminate many knowledgeable people.
Tone does convey valuable information though. SemiAccurate has been what can best be described as "anti INTEL" for a very long time. And while they've been right a lot about 10nm, it has also lead them astray in the past. He does not have this tone with articles about other industry players.
The Custom Foundry Roadmap [1] shown last November, they were suppose to be on GA now.
If any of these are true and we assume a company really placed the wrong bet on Intel, then the only conclusion I can come too, as I have previously thought; Intel lied, they lied to everyone, investors, partners, and shareholders about their 10nm progress. They let their pride swallow them.
And then may be everything else made sense. BK's departure, rumours of Apple ditching both Intel x86 and Modem.
Once you caught someone lying, and it does damage to your interest, it will take a long time and hard work to earn the trust back. Something that is hard to quantified in numbers for analyst.
I still have faith in Intel fixing 10nm, this is speaking as someone who has been labelled as TSMC cheer leader for the past few years. Whether they could fix it in time is a different story though.
He used to be an editor in The inquirer, a spin off from The Register ( UK ) focusing purely on rumours. It was very popular in the early 2000s, and I don't remember the details but he broke a few Intel stories that got him pushed out of the site, or The inquirer were punished and not letting them joining any Intel events. Something similar happen to Nvidia too.
He is very much pro AMD, ATI and Anti Intel, Nvidia.
Although he has a comparatively accurate track record for both Intel and Nvidia leaks.
He also tends to be over dramatic in writing, that is even before the paywall and subscription model. So his style were not the results of trying to sell you $1000/Year subscription. But as long as your know his perspective and history, then his article are much easier to read.
Cool, thanks for the history, that all makes a lot of sense. I didn't find anything that hard to read about the article, it's just blatantly anti-intel.
Well I am just doing the summary for you to save you time searching, and more time working on FreeBSD :P
Edit: Now I remember, it was the story about Intel Pentium 4 scaling problems. He broke the story about Intel will never reach whatever silly Ghz they initially promised. And Intel were not happy. ( That was nearly 20 years ago... I am getting old)
Since this was posted it came out that Apple's ditching the Intel 5G stuff and as a result Intel killed off the product and the team that was working on it. I suspect nokia was betting on that product because it was 10nm, just because it was a good fit for their product plans.
The team is supposedly working on other 5G chips now, not killed. Nokia wouldn't have been using that particular product anyway. Its interesting but I'm not sure it's that related. Apple are always taking more and more stuff in-house.
49 comments
[ 3.0 ms ] story [ 121 ms ] threadThe fact that the article teases an industry going through change makes me think it's probably a company that produces things for 5G. My money would be on Nokia.
When a chip design is stable enough for manufacturing them. But chips usually have post si bugs and so the tape out is usually a back and forth process depending on how complex thechip is and how competent the design team was.
tldr this article is basically saying the 10nm is so screwed up they can't just adjust a few things, they need to fix a lot and the more changes you start having to make the more verification that needs to be done before things are signed off and ready for tapeout.
https://www.theregister.co.uk/2016/08/16/intel_foundry_arm/
https://www.arm.com/products/physical-ip
https://www.arm.com/products/physical-ip
I suppose the worst case would be an expensive R&D initiative (with ARM paying Intel for prototypes) about how to make ARM cores with Intel's 10nm process, with no usable result. Not likely to sink ARM.
ARM doesn't sell silicon, they license IP. However might be they're licensing IP blocks. Essentially a tested CPU layout you can more or less paste into a design. Think Lego's.
"ARM, challenging Samsung, Taiwan Semiconductor Manufacturing Company (TSMC) and GlobalFoundries in the mobile SoC fab space."
https://www.theregister.co.uk/2016/08/16/intel_foundry_arm/
They certainly can go to TMSC or GF with their designs but it will take time (months or years) to validate the designs, and money to make it happen. And this doesn't even take into account Fab capacities. TMSC and GF might be committed to other firms and not have capacity for this new work.
The 5G rollout has already started and vendors should have been shipping their products by now. Not having anything for the first wave means you're going to lose most of the market.
I know Charlie has an record of breaking all sort of secrets from Intel ( and Nvidia ). But are there really 100 people paying for this?
Seems like something hedge funds/private equity would pay for. The biggest edge those companies have over retail investors is access to knowledge that's not considered MNPI but is also context the general public doesn't really have, through news sources like this and also (perhaps most importantly) expert network companies.
Sidenote: if you could improve compliance or cheapen the cost of it for such firms, you could make oodles of money. Compliance is a massive, expensive risk center that the industry hates managing.
For investors $1000/year is a bargain.
My money is on Motorola, even though it's just under $20B. Other options don't look as reasonable. Maybe HP?
Motorola has already changed hands once, as has Nokia
> The push for speed could keep Qualcomm ahead of Intel, its rival in handset baseband chips. Intel announced in November plans to support CDMA and gigabit speeds in its 2018 cellular modems and up to 1.6-Gbit/s downlinks in 2019.
Ericsson were having a hard time competing, if Nokia screwed up may be a M&A with Ericsson makes sense.
Godsh this really is a big story if this turn out to be true.
Edit: Having some more thoughts, the delay still doesn't make Nokia bankrupt, because the initial 5G roll out are small in numbers. Compared to 3G / 4G recurring infrastructure and patents revenue. So I guess it is going to make an impact on Nokia, but nothing as bad as Charles has wrote. He is likely right that his sources suggest Nokia has nothing to deliver or sell in terms of 5G, but again if you know anything about business I doubt this will bring Nokia down.
I read an interview with GlobalFoundries which talked about different process node characteristics. [1]
Who would really benefit from the 10nm process? Mobile SoCs, basebands/modems and AI chips. There has to be a high performance and low power requirement to justify being an early adopter of a new process.
Or intel aggressively misled fab customers about the progress of their 10nm efforts.
[1] https://www.anandtech.com/show/12438/the-future-of-silicon-a...
> ReefShark packs more functionality into 50 percent smaller hardware compared to products that use discrete components > ReefShark chipset decreases mMIMO antenna size by half >ReefShark cuts energy use by 85 percent, increasing deployment options >Boosts the intelligence and performance of MIMO antennas
So density and power. This chipset would be around for a long time, it makes sense to be on the latest process.
A failed chip project will not sink them.
Maxim
Skyworks
NXP
Analog Devices
Microchip
The problem is none of these companies require latest node for their product portfolio, and even if they do the product is likely not be substantial of their revenue, i.e it won't bring the company down as the article has suggested.
Consider, an expert becomes an expert through their passion and dedication to a topic. If you eliminate emotion from your resources, you eliminate many knowledgeable people.
If any of these are true and we assume a company really placed the wrong bet on Intel, then the only conclusion I can come too, as I have previously thought; Intel lied, they lied to everyone, investors, partners, and shareholders about their 10nm progress. They let their pride swallow them.
And then may be everything else made sense. BK's departure, rumours of Apple ditching both Intel x86 and Modem.
Once you caught someone lying, and it does damage to your interest, it will take a long time and hard work to earn the trust back. Something that is hard to quantified in numbers for analyst.
I still have faith in Intel fixing 10nm, this is speaking as someone who has been labelled as TSMC cheer leader for the past few years. Whether they could fix it in time is a different story though.
[1] https://www.eetimes.com/document.asp?doc_id=1332527&page_num...
He is very much pro AMD, ATI and Anti Intel, Nvidia.
Although he has a comparatively accurate track record for both Intel and Nvidia leaks.
He also tends to be over dramatic in writing, that is even before the paywall and subscription model. So his style were not the results of trying to sell you $1000/Year subscription. But as long as your know his perspective and history, then his article are much easier to read.
Edit: Now I remember, it was the story about Intel Pentium 4 scaling problems. He broke the story about Intel will never reach whatever silly Ghz they initially promised. And Intel were not happy. ( That was nearly 20 years ago... I am getting old)