Home / Component / APU / AMD: first products made using 14nm/16nm FinFET are ‘on track’

AMD: first products made using 14nm/16nm FinFET are ‘on track’

Although Advanced Micro Devices has not revealed any details about its 2015 roadmap this year, the company continues to unveil pieces of the puzzle. Last week the firm implied that it will use 20nm fabrication process for select products, but on Tuesday AMD hinted that products made using 14nm and 16nm FinFET manufacturing technologies could emerge only in 2016.

“We will continue to transition and we have our FinFET designs well underway, but we will not be the first user, the bleeding edge of any new technology node,” said Mark Papermaster, chief technology officer of AMD, at Barclays Global Technology Conference. “You will see us be a very, very fast follower, so we are right on track with our FinFET designs and what you will see next year is a really 28nm and 20nm products from AMD.”

According to publicly available information from AMD, all of the company’s upcoming accelerated processing units aimed at high-volume market – code-named “Carrizo”, “Carrizo-L”, “Kaveri”, “Seattle” and other – will be made using 28nm process technologies. AMD also said that the first system-on-chips that belong to the Skybridge product family of ARM and x86 chips will also be made using 20nm fabrication process. Finally, there is 20nm Xbox One SoC in plans as well as 20nm graphics processing units.

amd_carrizo_apu_fusion

Keeping in mind that AMD has not announced any 14nm or 16nm products, it is possible that the first commercial chips made using the aforementioned process technologies are on track to be released sometimes in 2016.

TSMC plans to start volume production of chips using 16nm FinFET process technology in late Q2 or early Q3, 2015. If AMD introduces its 20nm graphics processing units early in the year, it will have a competitive GPU lineup for several quarters. However, if the company is late with its introduction, it will not be able to enjoy any advantages of TSMC’s 20nm fabrication process. AMD's arch-rival Nvidia, who reportedly plans to skip 20nm and use 16nm FinFET instead, will rely on 28nm GPUs until the end of the third quarter of 2015. AMD should release 20nm GPUs in the Q1 2015 in order to have performance and power consumption advantage over its competitor.

GlobalFoundries plans regarding 14nm FinFET technology it licensed from Samsung are not completely clear. Samsung should be ready with 14nm manufacturing capacity even now, but the situation with GlobalFoundries is uncertain.

Discuss on our Facebook page, HERE.

KitGuru Says: Outdated process technologies may not be a big problem for AMD (even if it offers 20nm GPUs while Nvidia sells 16nm chips). A really important issue could be products that are not competitive against offerings from AMD's rivals…

Become a Patron!

Check Also

Qualcomm details the Adreno X1 GPU powering the Snapdragon X series

In the lead-up to the official launch of the Qualcomm Snapdragon X Elite and Plus …

41 comments

  1. Talk less do more…

  2. Kristijan Vragović

    We’ll see. They have to deliver brutal products or they are toasted. No one will buy their CPUs if they dissapoint again, maybe for really low pricing but that is doom anyway. So let’s wait and see, there is another year in front of them. They will deliver GPU for sure, but CPUs, well i will wait to see. Really hope for AMD comeback

  3. NOT good. This means that in 2015 ALL AMD offerings will be behind competition. ALL. They will just try to stay not far behind, but they will be behind.
    There will be no 20nm GPUs. The process is just not good enough for GPUs. So AMD will have to keep improving their 28nm GPU architecture even more in 2015 and lower the prices even more to be able to offer something against Nvidia’s products. I am afraid that the 4th quarter in 2015 will be even worst for them than what this quarter is because of the introduction of GTX 970(GTX 980 is pointless at that price compared to 970).

  4. When did AMD last disappoint on the GPU side? I agree that they need to get it together on the CPU side, though. (of course)

  5. Kristijan Vragović

    Yes, i see now. 🙂 my bad

  6. Kristijan Vragović

    Yes, i see now. 🙂 My bad. Wanted to say CPUs. Will correct

  7. You response is not good. AMD GPU’s current line-up is not that far behind in performance now,so you think the new products will be worse? The have many ways to improve,from HBM to better power consumption through improved processes. AMD has now 3 suppliers for 14/16nm FinFet ready for 2016.Much better outlook than the green side,which has to fight with Apple for wafers.

  8. They are pretty much behind in efficiency. Also in performance they are competitive only because Nvidia came out with mainstream models, priced as hi end. AMD needs a 512bit data bus, lower prices and GM200 locked in a drawer to considered “not that far behind”. They are behind. Look, I am an AMD fan, but I also see reality. And I don’t like it. AMD’s GPU business is starting to look like the CPU business, when Bulldozer came out.

  9. GloFo is on track with their with their 14LPE… It’s in qualification now and 14LPP should come by Q2… It’s an exact copy of Samsung’s and they are in full production now…

    I believe they will release a ZenK12 APU as the 2016 one… They’ve been working on them for 3 years… FX WILL RETURN…

  10. who are the 3 suppliers?

  11. CPU for desktop on a low power node??? If that will happen that this will have to be not great architecture but monstruos perfomance wise. But, we will see. Let’s hope that AMD will ready CPUs with Zen for holydays 2015 or very early 2016.

  12. Kristijan Vragović

    CPU for desktop on a low power node??? If that will happen that this will have to be not great architecture

    but monstruos perfomance wise. But, we will see. Let’s hope that AMD will ready CPUs with Zen for holydays 2015 or very early 2016.

  13. Global Foundries, Samsung and TSMC are all AMD partners. nVidia only uses TSMC.

    Samsung and GoFlo both have 14nm on track as of now, TSMC is lagging behind with 16nm(14nm) just getting up to speed with volume production by end of 2015. GoFlo and Samsung 20nm process is already up and running in volume.

  14. Samsung have even 14nm for Apple A9. The G.F. and Samsung 14nm finfet it’s a real 14nm or have internal component at 16/20nm (i read it time ago)?

  15. I see you’ve already delivered a verdict on a new architecture that we know nothing about aside from the people working on it.

  16. Ok, but why are you assuming they’re not going to release any new cards until 14/16nmFF?

  17. Kristijan Vragović

    I see… Well that was a miss spelling… I really want them to bring a major performance upgrade. My bad. I will correct last sentence

  18. Kristijan Vragović

    Corrected…

  19. Kristijan Vragović

    Mistakes happen. English is not my native language and sometimes i do mistakes when typing. Honestly, because of the flop with Bulldozer AMD is in though position and few people trust them

  20. Yeah, they definitely are, but they’ve got some very successful people working on the new architecture, so I would expect it to put up excellent competition against Broadwell-E, and maybe decent competition against Skylake, depending on how well it performs.

  21. Kristijan Vragović

    According to many articles on the web and small amount statements of AMD people they are planning to compete on all fronts… From low to enthusiast level… In my opinion Zen will replace all of the products from low to high end as i think that it will be very scalable and versatile, but that is just my opinion. That could make things easier to AMD. But i’ve learned that some things one just need to wait and see

  22. Kristijan Vragović

    What’s Skylake and Broadwell-E is concerned you just have to admit that this products will be hard to beat… I’m not saying that AMD is not capable of it, but it will be though…

  23. I mean seriously, you don’t really think they’re going to keep the exact same GPU lineup until the second half of 2015, do you? You’re also assuming that they haven’t made any efficiency improvements, and have been working on literally nothing for the past year since Hawaii was released. What, do you think they just sat back and twiddled their thumbs while Nvidia was working on Maxwell?

  24. I haven’t said that they will not produce anything. I said that if they stay at 28nm for too long, they will be constantly behind Nvidia, at least until 14nm FinFET products come out. Looking at when the first Maxwell products come out and what Tonga had to offer months latter, it does show that at 28nm Nvidia does have an advantage. If Nvidia moves to 16nm FinFET at the end of 2015, and AMD will have to wait until 2016, then in the 4th quarter of 2015 we will have 16nm FinFET products from Nvidia and 28nm from AMD. In case AMD moves to 20nm in 2nd quarter of 2015, with some performance products and not just some low end mobile GPUs, APUs or SoCs, things could be a little better.

    Are you still confused about what I said?

  25. GloFo will have their 14nm LPP node ready in Q3 2015, so AMD should hit 14nm at the same time or a little sooner than Nvidia, depending on if TSMC can catch up. Also, the 285 was released before Maxwell, unless you’re considering the 750 Ti, which was essentially a prototype. The 285 was AMD making efficiency improvements to their architecture and then testing it on a mid-range card. The fact that it was a mid-range card means they obviously didn’t intend for it to be anything groundbreaking, just showing what improvements they made to GCN. The 20nm node is completely unsuitable for high powered products like GPUs, so that’s not even going to happen. We’ll only be seeing SoCs and ARM products using that node.

    Also, your previous paragraph compared the efficiency of Maxwell to AMD’s year old Hawaii as if they were competing architectures, which they’re not. Maxwell, as far as efficiency and design goes, can only be compared to AMD’s offerings from the same generation, which will be Fiji and Bermuda. I mean really, would you compare a 7970 to a 780 to try and compare architectures?

    Can you see why I found your statements a little absurd?

  26. Your post is full of excuses and best case scenarios.

    Let’s just comment only the last about efficiency. IT DOESN’T MATTER TODAY if Fiji is as efficient as Maxwell in 6 months from now. If you think that it does, then just go on your favorite online shop and buy your Fiji card NOW.

    You know something? I am an AMD fun, but the day I will post something as -how you call it, ah yes- absurd as that you just posted, I will throw my PC -and it is a full desktop- out the window. You write a huge post to say that you happily replaced logic with wishes.

  27. You realize there’s a 2-4 month interval between the release of Nvidia’s and AMD’s new cards, right? The 680 was released 3 1/2 months after the 7970, and it was the main competitor for it. Did that mean Nvidia was behind? No, just that they needed to finish their final production stages before launching the card. Did it impact them in the long run? No, not at all, because it was powerful enough to provide good competition for the 7970.

    The 14nm is based on statements from GloFo and AMD, and the fact that TSMC is a fair bit behind GloFo on producing FinFETs, nothing to do with speculation. Also, considering AMD has a card that has been in testing for over a year and has shown to be 100% stable using GDDR5 for the past several months, I would think that they’re waiting for Jan/Feb 2015 so that they can use HBM in their cards, as HBM production is entering large scale in January. I would also expect them to be using GloFo’s 14nm LPP node in Q3 2015 for their flagship, as that is the only way to ensure that that 4096SP monster doesn’t consume ridiculous amounts of power, or overheat because of an extremely dense die.

    My speculation is based on educated guesses after looking at statements from companies, Sisoft benchmarks, and the state of production nodes, not wishful thinking.

  28. You do realize that your wishes, expectations and educated guesses, are NOT universal law, right?

  29. Obviously, but they’re extremely reasonable, and nobody has any better ideas right now. There’s also the possibility that they finished up performance optimizations in the architecture and are just barely putting the card into final production, and that there won’t be any HBM involved, but besides that there’s no reasonable explanation for why they haven’t released a fully stable GPU to counter little Maxwell yet. Also, I don’t think we’ll need to worry about Nvidia using 16nm until Pascal in 2016, as there is a card that uses Maxwell architecture with 3072SP in Sisoft, which indicates that it’s being made on 28nm. There’s also the fact that Nvidia wouldn’t be willing to sacrifice yields and efficiency in exchange for being on the absolute bleeding edge of a node. Also, there were no wishes in my statements, as those would be entirely unreasonable.

  30. If that’s your best argument, I think I’m done now. If the cards launch and I’m painfully wrong on all counts, then feel free to come back and rub it in my face as much as you’d like, but until then I’d need to hear an actual reason for me to be 100% wrong in my assumptions.

  31. I have realized much much earlier, like a week ago, that with you a dialog would be going forever, except if I was accepting your thoughts. So yes, that’s my best argument if you totally ignore my first posts and I am glad you are done here. And no, I don’t care to come back and rub it in your face.

  32. Considering your first post was speculation with no supporting information that only considered the worst case scenario… I think I’m good. Also, if you point out something that makes my statements blatantly wrong then I’ll certainly apologize and stop spouting BS, but I did several hours worth of research to figure out what I should expect from AMD next year. I have no idea how competitive it will actually be, as that depends on where Big Maxwell lands in the 30-50% area of improvement over Little Maxwell, but I have a good idea of what to expect as far as specifications, process nodes, and release dates go. You seem determined to believe that AMD has no hope of competing whatsoever… which isn’t very characteristic of an AMD fan…

  33. Being an AMD fan doesn’t mean that I have abandoned logic. As for your research, have you calculated also their financial position? Because someone could, for example, after hours of research conclude that AMD will come out in the first half of 2015 with AM3+ Excavator processors and FM2+ APUs with more than 2 modules considering for example the smaller size of Excavator cores compared with Steamroller cores. Why is AM3+ abandoned? Why is Excavator confined to Carizzo mobile chips? I could make a dozen more questions of this kind, and then maybe think a few more.

    Anyway, things are much more complicated than you might think. The only hope of AMD, is their ties with GlobalFoundries. If they can get a good deal on 20nm and 14nm FinFET from day 1 those precesses are ready for mass production, something that the financial department will not oppose, then we could have a much better 15 than what I am expecting.

  34. AMD had one of their top people with extensive chip manufacturing experience jump ship to GF, so that probably indicates something, although I’m not sure what. We’re definitely getting a 28nm GPU, probably with HBM, in late January or early February. The flagship is either going to be 20nm FD-SOI or 14nm FD-SOI, with a small chance of 14nm LPP depending on how suitable it is, and it’ll be released sometime in Q2 or Q3. I mean, there’s a chance it’ll be 28nm, but it’s not very likely.

    Adopting a new node for an architecture isn’t ridiculously costly unless you have to redesign the architecture because it was planned on a larger node, and I doubt AMD lacks that much foresight. I doubt they’ll lose too much in yields either, as they said they would adopt new nodes when it made financial sense, which means they’ll wait until yields are reasonably high to enter mass production.

    Gonna have to say that the first paragraph really doesn’t make sense, as anyone who did their research would conclude that AMD has no reason to use more than two modules in their APUs, as they provide perfectly fine performance already and the increased die size would probably produce too much heat with that third module, not to mention they’d see greater improvements from adding more GCN units. Excavator wouldn’t be on AM3+ because AMD would have to produce independent CPU chips, which wouldn’t sell very well and would cost quite a bit to produce, and it makes no sense for them to do that when they’re prepping a new CPU architecture with a new socket and new chipsets for 2016, so making Excavator on AM3+ would be a waste of time and resources. Not having any new architectures for the socket gives them no reason to update chipsets, thus it’s abandoned.

    Sorry, just felt I should point out that last bit, research without knowledge in the area leads to stupid assumptions, so your first paragraph would’ve applied to people that didn’t know very much about how the industry worked.

  35. I read it and I regret it. Ridiculous post and that last paragraph. Stop smiling at the mirror while writing a huge pile of nothing.

  36. Not smiling at the mirror, cringing in front of my laptop screen trying to figure out why my fingers won’t stop moving. It’s getting just as tiresome for me as it is for you.

  37. AMD and Nvidia is not launching any 20nm graphic cards come 2015. They are thinking about GlobalFoundries instead of TSMC, that only went ahead with the 20nm because of Apple and Qualcomm.

    PhoneyVirus
    https://twitter.com/PhoneyVirus
    https://phoneyvirus.wordpress.com/

  38. TSMC,GF and Samsung.

  39. AMD gives a 512bit data bus,the competition-196. 😉

  40. But but but but…. it’s only -3% lol

  41. EternalCrimsonSoul .

    in the power efficiency market they are struggling on both cpu and gpu now that maxwell reared its ugly head im wondering why though nvidia hasnt brought a successor to the 750 ti power consumption on a grander scale e.g 192 bit card 3gb with a higher clock using the more fine tuned 204 chip