Return to

The Bleeding Edge [Semiconductor Foundry Thread]



None taken, I should have linked the article.


Looking forward to see the anandtech review of the new iphone, their new testing methods are great for seeing real life performance and power efficiency gains.


Intel struggling with capacity on 14nm, and have to invest more in it, this is probably gonna hurt them in the long run.

In a bid to increase production of its 14-nm chips the company is investing an additional $1 billion in its manufacturing sites in Oregon, Arizona, Ireland and Israel that product chips using the said technology. Intel originally planned to spend $14 billion in CapEx this year, but then allocated another billion to boost production capacity of its 14 nm fabs. Intel naturally does not detail how it plans to upgrade the facilities, but $1 billion can buy you many step-and-scan systems to process wafers.


The anandtech iphone review is out with more details on the A12 chip.


What low power dual core risc chips exist in the magical land of SOC’s that would run as a good HTPC environmet?

Would rockchip be a good solution? The idea is hardware based 264 and 265 solutions.


TSMC has taped out their first 7nm EUV customer chip.


Intel is splitting its manufacturing group in three.

Technology Development, to be led by CTO Mike Mayberry
Manufacturing and Operations, led by Ann Kelleher
Supply Chain, led by Randhir Thakur


Samsung started their 7nm production with EUVL for select layers.

Intel just dropped to 3rd and last place in the foundry race by the looks of it.


Hadn’t seen this posted yet:

Samsung has announced that it expects to start NGAA chip fabrication sometime around 2020 or 2021.

They also mention IBM working on using chips for biological sample analysis via “nanoscale deterministic lateral displacement array” (nano-DLD). Sounds interesting to me; any idea if this is practical or just a pipe dream?


It is interesting. Most companies look for multi sourcing. But with this run down to 5nm. And the cost in Fab.

We are getting stuck again with less than a hands fingers full of supplies. Now Intel failing, TSMC, Samsung. Going hard for 7nm and 5nm…and looking to 3nm

On the whole planet !!! Only 3-4 fabs can make what we all want in our PC’s, Phones, the cloud and we are always if you listen to the media at war with each other.

I would prefer an Australian 7nm fab along with a US, China and Russia Fab…Toss in the UK. But We have binary Fab’s in single locations servicing us all.

Why :slight_smile: Profit. Humans would rather die than lose money. We talk about making a mars colony to save the earth but we cant even make a 7nm spare fab to save the earth ? Unless you call 4-5 in asia backup.


Unfortunately our society demands profit in order to prosper.

I’d love to see a CPU fab here in Australia too (rather than just dig dirt out of the ground and ship offshore for processing), but that’s a whole other problem.

If you want to think about something really depressing, compare the cost of a single CPU fab, or stealth bomber or aircraft carrier or whatever to the amount of money set aside for cancer/aids/whatever research.

But that’s a whole other tangent.

Back on topic. I really don’t see how intel are going to get out of this without going fab-less unless they open their fab to others. And not as a premium “we have the best fab in the world” vendor. As a Global Foundries like “second tier” cheap option for stuff that doesn’t need to be cutting edge.

Problem is, most of the “others” are all about power consumption/heat and small physical size (think IoT devices, mobile, etc.). Intel really screwed the pooch on this, and their hold-ups with 10nm are just another nail in the coffin.

How the mighty have fallen; assuming TSMC’s 7nm is as good as intel 10nm or thereabouts, AMD vs. Intel in 2019-2020 is going to be a blood-bath (because TSMC/Samsung actually look to be on track). I suspect intel’s market forecast for AMD gaining 15% market share are best-case, shareholder-friendly projections in order to avoid shareholder panic.

Ryzen and Threadripper are already making huge inroads and they’re curently on an arguably significantly inferior (or at absolute best, equivalent to intel 14nm+ or 14nm++) Global Foundries process right now.

I find it massively ironic that Intel look like they’re going to be cut down via fab inferiority, after being the leader for so many decades.


This sentence is just there to bait me, lol. It’s really hard to not fall for it :grin:.

I mostly agree with the rest.

I think this situation is about to change whithin the next, say 15-20 years or so, as it is increasingly difficult to reach smaller nodes with current silicon based technology. Even EUV can only do so much, it will mostly improve yield. I think when everyone starts to get EUV working cost effectively, we will be stuck on mostly similar nodes where everyone “only” does in-node improvements. This is the time for the various fabs to catch up until the next big breakthrough is production ready and that has to be some kind of material/technology change. (my opinion)


I’d somewhat agree with this, but i think the concern is: if everybody else dies off because of inability to compete, could anyone ever start a new fab and go direct to say 7, 5 or 3nm (or wherever we end up) without ramping up their own tech via the larger process nodes on the way up?

Or put another way, has the window for starting a semiconductor foundry now closed and the barrier to entry is now (and forever after) too high to climb?

I mean, you’d need to not only be able to build the fab with tech good enough to compete, but without the reputation or customer base, you’d need to undercut the existing players in order to gain market share.

I fear that this market will consolidate to a monopoly; maybe government intervention may result in a break up at that point (in order to force competition and second-sourcing), without that i fear that we’ll eventually have one remaining foundry. Maybe two, with collusion between them to keep anyone else out (like the DRAM market at the moment).


If past technology developement is an indicator, then yes it will absolutely be possible to directly go to small nodes, probably not to the cutting edge but close enough.
There is a lot of techology from the earlier semiconductor industry that is now so common place that you can buy equipment for a small(ish) lab. Yes, it is expensive but not the multi billion dollar investment it once was.

There are a couple of ways for a new competitor to gain market share, yes it’s hard, but possible. One way is kind of what GlobalFoundries is doing now, don’t be on the cutting edge but make what you have as cost effective as possible. Arguably that is easier for a new competitor because then you avoid a lot of the baggage of decades of developement, but it has disadvantages, true.
Another possibility is to create your own niche market at the beginning where you can build trust, this is traditionally how new competitors gained foothold in other markets.
I am wary of pointing out specific examples because it is too easy to poke holes into an analogy and get off track quickly, but I am sure you can think of a couple yourself.

I share your fear, however in the long term we need a lot of new technology to further advance the semiconductor industry. It is unlikely that only one competitor will hold all the cards. Even then, there is research at universities around the globe. A lot of the ground work is done there, even if it is not very glamorous.


Treating it as a rumour for now but there are reports that Intel has given up on 10nm.

Given this is now a 3 year delay, I would be inclined to agree with them to stop focusing on it and find other ideas.


Semi Accurate first posted the story.

I also suspect that this is highly plausible.
Intel is moving on to an adaptation of the P1276 process, 10nm was P1274 and 14nm is P1272.

P1276 (7nm) was initially planned for 2020.

EDIT: posted old link.


Oh my…



Well then. I believe Intel because their propaganda department told me so.


If it were true that they had canceled 10nm, wouldn’t this statement have legal repercussions?


Well, I’ve got a feeling that they’re not just cancelling it, but moving their high performance stuff on to something that’s going to work better.

They already have 10NM stuff, so technically they did deliver on 10NM.