Comments Locked

111 Comments

Back to Article

  • essai - Monday, February 16, 2015 - link

    And intel doesn't have any process advantage anymore!
  • psyq321 - Monday, February 16, 2015 - link

    While different manufacturers have different measurements of nodes, so comparisons are not really "apples to apples", Intel has 14nm parts in commercial production for some time already, so they still can be said to have some advantage, albeit smaller than before.
  • essai - Monday, February 16, 2015 - link

    While intel has more experience with FinFETs and probably has a more opimised process, the days of them being at leat one year ahead of the rest are gone. In europe broadwell parts are still hard to come by, while samsung will probably have the S6 shipping everywhere by june, this really reduces intels advantage to just 1quarter, which is pretty negligible. And am not even talking about their atom processors which are still on 22nm. I dont see how they will be able to gain a footing in the mobile market after loosing their process advantage...
  • smilingcrow - Monday, February 16, 2015 - link

    I think you have missed the point that you were replying to. Intel's 14nm process is a full shrink so has ALL the gains of a true 14nm process. Whereas the 14 and 16 mm processes of other foundries appear to not be full shrinks so they lack some of the benefits such as comparable size so will be significantly more expensive per square mm.
  • mali_07 - Monday, February 16, 2015 - link

    What if told you all these are marketing. Even Intel's 14nm cmos real gate length is more than 20nm!!!! Don't ask me how or else i will be thrown out of my job.
  • smilingcrow - Monday, February 16, 2015 - link

    No proof makes you a random person on the web with zero credibility so don't waste your time.
  • LordanSS - Monday, February 16, 2015 - link

    Here, since you asked for some proof. What he's saying actually has substance.

    https://www.semiwiki.com/forum/content/3884-who-wi...
  • Guest8 - Monday, February 16, 2015 - link

    @LordanSS Yeah Semiwiki is a shill for the fabless coundries and is paid for TSMC so there's that. Intel is the only company with a full shrink APU all others are half shrinks that's why their costs are going up while Intels is going down.
  • levizx - Saturday, April 18, 2015 - link

    Where's your "proof" now? Exynos 7420 is physically much smaller than 5433 which has less GPU cores.
  • extide - Tuesday, February 17, 2015 - link

    It's pretty well known that the xxnm process nodes are just marketing names these days, and the numbers come from a relative shrink ratio. They havent been actual measurements for a long time! This isnt news!!
  • ddriver - Tuesday, February 17, 2015 - link

    It is only logical that other manufacturers will catch up, chip process is not something that can scale indefinitely, soon it will hit a brick wall and will require an entirely different technology to continue scaling things down.

    And of all the potential replacements none I can think of is really applicable on a a large scale, so chances are chip process will only get to a point and stay there, you know, like it happened with rocket engines in the end of the 60s - barely anything has been improved upon since then, nor any more promising technology has emerged in those 40+ years.

    My hope for the future is printed electronics becoming cheap and efficient - naturally, they will never be as efficient as current etched chips, but they will be chip to make custom designs in small quantities, and dedicated designs vastly exceed the performance and efficiency of general purpose processors running the workload as a program. And you don't need a factory to make it - all you need is a small printer, it will be like FPGAs but much cheaper efficient, and I assume also very environmental friendly - barely any environmental impact and 99% recyclability.
  • Guest8 - Tuesday, February 17, 2015 - link

    It is logical but improbable. Easiest example Intel buys their tools from the same companies as TSM, Samsung etc and yet they have been and still are ahead in process technology. Logically speaking every one should be on the same process node but they are not.

    "The foundry industry is spending about $25 billion per year to supply Apple and QCOM with about a total of $12 billion per year in product ($7 billion to Apple and $5 billion to QCOM) That can't work. ..... Then to make matters worse, Apple bounces their business, (or plans to) from TSMC to Samsung and back again. How can that work?" - Russ Fisher

    This will be the last stand for TSM and Samsung to come close to Intel as there just isn't enough business to sustain the R&D required no other mfg has a ridiculous amount of cash coming in like Intel does to finance their R&D to continue their node shrinks. It will get even worse when Intel finally produces a competitive SoC at 14nm and below as that will further displace the available money the other foundries would get. It will take several years to play out but this is what will happen.
  • ddriver - Tuesday, February 17, 2015 - link

    Intel makes more money and can afford to burn more money on tooling and process improvements.

    Kids get their clothes from the same store, but kids with rich parents by clothes more often and stay trendy, kids with poorer parents buy new clothes less often, sometimes have to skip a fad or two...

    Traditional chip manufacturing will get stranded in the years to come, I mean in less than 10 years. There are ways to scale down, but none of them are commercially feasible and will likely never be. Chips will get stuck in process, and that's normal - a bell shaped curve has one single peak. And besides normal - it is good enough. So it is not a bad thing that in 20 years we will not have supercomputers in our pockets, because such are not needed, even at this point enough is enough. The industry might be able to scale another 3-4 steps down, but that will be it. The rest is empty promises much like people were supposed to have hover cars and deep space travel by the 2000...
  • mkozakewich - Tuesday, February 17, 2015 - link

    We *do* have supercomputers in our pockets.

    And no matter how powerful CPUs become (even a few more orders of magnitude) people like you will never refer to the ones in our pockets as supercomputers because they won't stand up to the supercomputers of the times.
  • ddriver - Wednesday, February 18, 2015 - link

    No you don't. Just because a cellphone today has more horsepower than a "supercomputer" from back in the days they were made of tubes. It does without saying that when I say "supercomputer" today, I mean one by contemporary standards, not something abstract and relative.

    Just like when you say "tomorrow" you mean the day after today, not that "always the next day" which will never arrive.

    To put it in an absolute metric - by a "supercomputer" I mean a petaflop scale device. It will not happen in 10, nor in 100 nor in a million years.

    As for the fact we use devices which are supercomputers relative to stuff that was used in the past for achieving great tasks and reaching important milestones, and we use that for making duckface "selfies" and waste our lives playing stupid games - that just goes to show better technology doesn't really result in better people, and can often be on the contrary...
  • Guest8 - Wednesday, February 18, 2015 - link

    @ddriver the problem with your analogy is in the APU mfg business you can't "skip" a node or two and come back roaring as you will lose the sales that come with bleeding and leading edge processes. When you lose the sales you lose R&D funds and without funds you can't exactly go into the APU thrift store and pick up new methods. If Apple doesn't come back to TSM how will TSM justify the expense of going down to 10nm and below? They don't have the cash, they don't have the sales and they don't have the technical knowledge acquired by producing the previous nodes. It will be down to Intel and maybe samsung depending on how many employees samsung can poach from Intel. That is the only reason samsung can stay competitive with Intel as they just poached TSM employees on the cheap. You can already see it now everyone is moving to Samsung's 14nm process which chip vendor has announced TSM 16nm FF+? Qualcomm? Nope. Apple? Nope. That's the bulk if not all of your volume at the bleeding edge. Qualcomm may come back to TSM but why would they after the problems they have at 20nm and losing Samsung APU business? There is way more than just can it be done in the business economics play a huge role. The only business TSM will have in a few years is going to be the low margin legacy node business they won't be able to command a premium at the bleeding edge as they will no longer produce at the bleeding edge. 16nm FF+ is their last stand against Samsung and Intel.
  • jimjamjamie - Monday, February 23, 2015 - link

    I feel like software still has a lot of catching up to do compared to hardware at this point anyway - see the current state of the GPU industry for an example. All those billions of transistors wasted until the likes of Mantle/DX12 and multi-threaded drivers have shown that there is a huge amount of performance tweaking that can be done at the low level. Obviously, this can only go so far. But from what I can see, we're a long way off fully exploiting the existing hardware available.
  • Kristian Vättö - Tuesday, February 17, 2015 - link

    The lithography refers to the smallest pitch in the silicon and always has. The actual density doesn't necessarily correlate with that, so Intel's 14nm can't be compared with others' 14nm until the first die shots are out.
  • PC Perv - Monday, February 16, 2015 - link

    "No True Scotsman," I see
  • Yojimbo - Monday, February 16, 2015 - link

    I read an article showing that by at least one metric, the Intel 14nm process is much more dense than either the Samsung 14nm or the TSMC 16nm, and previously, at times when Intel was far ahead in the nodes, the opposite was true. So perhaps Intel's lead is pretty much the same as it was, less than it seemingly was before and more than it seemingly is now.
  • witeken - Monday, February 16, 2015 - link

    Let's forget that Intel has been shipping *hundreds* of millions of FinFETs since early 2012, which puts Samsung 3 years behind Intel.

    The only company who you can buy FinFETs from, is still Intel. Take a moment to realize this; Intel is now in its 2nd gen FinFET, while other companies have yet to ship their first generation! TSMC will only start HVM in H2 of this year.

    And let's also not forget that this node shouldn't have been called 14nm in the first place by Samsung, since it uses the same 20nm feature sizes. The accurate name would be 20+FF. But marketing didn't do that because then it would seem to people that Intel will be 2 nodes ahead when it launches 10nm next year (which will indeed be the case, but now people won't know because of the FUD marketing is spreading).
  • toyotabedzrock - Monday, February 16, 2015 - link

    Third generation, remember they had to improve things after the first 22nm parts, over clocking was improved after. And they seem set to release skylake on time or ahead of time.

    Would be great if anand could do an article on mpx and what it will do in skylake.
  • Guest8 - Monday, February 16, 2015 - link

    Yeah the faux 14/16 nm parts are really 20nm with FinFETS. So the Intel lead is very much real and still intact.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    Um, no. Samsungs "14nm FF" process is actually 20nm in disguise. The only company at 14nm is Intel, and it's by a huge lead. https://www.semiwiki.com/forum/content/3884-who-wi...

    Samsung's process at 14nm FF is 33% larger than Intel's. Sorry but Intel's > 1 year lead will remain until TSMC's 16nm FF+, and then Intel drops to true 10nm.
  • psychobriggsy - Monday, February 16, 2015 - link

    Certainly Intel doesn't have the 2 to 3 year lead it used to have.

    And as this SoC will be produced in the millions (tens of millions even) - and I'm sure Apple are bringing up the A9 on the same process too (tens of millions), the yield on the process at the sizes of these mobile SoCs (100mm^2 - 150mm^2) must be good.

    And 14nm yield is one of the rumoured issues of Intel's 14nm, putting it's mass release back by a year, and having the initial products be small die low-wattage parts.

    However the actual real world density of the processes is still to be compared - I believe that Samsung's 14nm process uses a 20nm BEOL for a start. Things like SRAM cell size, logic transistors per mm^2, and so on.
  • toyotabedzrock - Monday, February 16, 2015 - link

    It was a 3 month delay and it has been fixed.
  • Guest8 - Monday, February 16, 2015 - link

    More likely the yields are terrible. It's my belief that's the reason for the tech sharing with GloFo. Crappy yields but we need millions so we will just run more wafers. I will bet money there is a clause in that deal that allows Sammie to buy the output at cost if needed from GloFo in exchange for the 14nm tech to be shared. Not a bad deal for both since GloFo was in third place.
  • TheJian - Tuesday, February 17, 2015 - link

    Yields are fixed and good, that is why Apple, Qcom AND Nvidia all are going there Q2 according to many sources (samsung claims Q2 mass products from all these). In that order supposedly too, so NV stuff last, but all should make it into xmas products as it only takes 4.5-6mo to put something out with google being the best so far AFAIK at 4.5mo for nexus 7. I would be pretty happy with a 14nm shield device with maxwell and denver both with boosted clocks (which NV confirmed is on future chips - for anyone thinking Denver is gone). They explained they are on a TICK/TOCK like Intel now so went with off the shelf for time to market on X1 (and it was aimed at cars, where off the shelf IP means nothing power wise, huge car batteries).

    GF will benefit the most from TSMC screwup though, if sammy can't do enough for all. There deal goes both ways however. Whoever gets something fixed, the other gets it ASAP. So if Finfet+ or whatever they call it has great yields at GF, samsung gets it immediately too. I doubt there is a AT COST clause. It's a combined effort for the good of both to catch Intel (with IBM doing the groundwork first, then the two take it from there).
  • Guest8 - Tuesday, February 17, 2015 - link

    And you know these yield numbers how? That is one of the most guarded secrets in the industry. It took me weeks to get a snippet of info on how low TSM 20nm was yielding. Yields are not fixed. They are running a ton of wafers. Only time will tell who's right by seeing much product ends up being out there. Thus far Apple A8 is the only TSM 20 nm product out there on the shelves.
  • frenchy_2001 - Tuesday, February 17, 2015 - link

    You are mixing up his answer. He talked about Samsung's yields. TSMC is having problems with their 20nm (hence why Apple pretty much has the only mass produced chip from TSMC at that size). Samsung seems to have the volume and now switched to "14nm".
    There is no "at cost" clause in the co-development from IBM/GloFo and Samsung. It is just *SO* expansive to keep developing foundry processes that most companies cannot do it alone. Companies owning fabs has been a shrinking list for years, even IBM got out recently. It costs BILLIONS to develop a new shrunk process and BILLIONS more to tool a fab line. This is why we can count companies fabbing at 20nm and lower on the fingers of ONE hand (Intel, TSMC, GloFo, Samsung).
  • patrickjp93 - Tuesday, February 17, 2015 - link

    But Samsung has only published claims about yields for its own processors. No one has 20nm chips out yet made by them, and Qualcomm and Apple only just jumped to Samsung a couple months ago, Nvidia much more recently. We have no proof of yields yet.
  • JoshHo - Monday, February 16, 2015 - link

    It's important to note that this process uses 20nm BEOL so it isn't quite comparable to Intel's 14nm process.
  • krumme - Monday, February 16, 2015 - link

    Density is probably more or less the same, due to interconnect, as evident from other comparisons. If this goes to S6 - and it looks like - yield is not an issue either.
    This process is tuned for mobile devices and not high perf as Intel - thats the difference - so its comparing apples to oranges. But for low power mobile devices Samsung now have a proces lead to Intel - and it looks like the arch is nice and small.

    As a sidenote the LPDDR4 will make sure the GPU is fast as well as evident from prior 20nm 810 AT review.

    What is questionable is DSP perf. , modem perf/watt, everything outside the GPU/CPU - that
  • Yojimbo - Monday, February 16, 2015 - link

    Density does not seem to be the same:

    https://www.semiwiki.com/forum/content/3884-who-wi...
  • Krysto - Monday, February 16, 2015 - link

    Only the interconnects are 20nm and the transistors are 14nm. I think Intel did something similar with its "22nm Trigate" process for Ivy Bridge and Haswell. I think they had 22nm interconnects with 26nm transistors. So Intel's "22nm" wasn't a pure 22nm either.
  • Yojimbo - Monday, February 16, 2015 - link

    Yes, Intel's 22nm node was less dense than Samsung's and TSMC's. But what does that have to do with the 14nm node?
  • patrickjp93 - Tuesday, February 17, 2015 - link

    And to be fair, Samsung and TSMC arrived at "28 nm" much later, and with AMD's HDL research, they pushed 28nm farther down than its original design plans. Now with FinFET added the rest of the industry seems to have taken a backstep in density the way Intel did, but it saves them dealing with massive leakage currents that the planar process was making far too problematic.
  • III-V - Monday, February 16, 2015 - link

    Node names have nothing to do with the process dimensions. It's been like this for decades:

    http://spectrum.ieee.org/semiconductors/devices/th...
  • witeken - Monday, February 16, 2015 - link

    There's no such thing as a "14", "16", "22", "26", etc. *transistor*. The easiest way to (roughly) compare transistors is to look at their features: does it have HKMG, FinFET, etc. and at which generation?

    Right now, Intel is in its 2nd gen FinFET (14nm) while Samsung and TSMC only have 2nd gen HKMGs (20nm). This puts Intel roughly 2 nodes ahead in terms of transistor characteristics.

    The easiest way to compare density/interconnect is to multiply 2 feature sizes (which can be obtained by measuring them like Chipworks does or using publicized data), after which you can get such a graph for density: http://image.ofweek.com/uploadfile/newsimg/b/2014/... This is only a rough comparison too because a process node always has multiple flavors like high density for SoCs and high performance for CPUs etc.
  • phoenix_rizzen - Monday, February 16, 2015 - link

    When clicking that image link:

    403 Forbidden

    nginx
  • teiglin - Monday, February 16, 2015 - link

    just remove the trailing period.
  • phoenix_rizzen - Wednesday, February 18, 2015 - link

    Ah, didn't notice that. Nifty graph!
  • PC Perv - Monday, February 16, 2015 - link

    Why is that "important?" Out of curiosity. Because that makes you feel like your xxxx is bigger?

    This is a serious question. No one is innocent in these marketing talk including Intel, yet only when Intel is perceived to be under threat, now it matters? Why? Because Samsung (or anyone else) will make x86 CPUs tomorrow? Or are there phones with Intel's 14nm SOCs?

    What is "important?" There may be differences between each OEM's tech and nomenclature, but I do not see how that is "important" outside the marketing.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    There are now phones with Intel's SOCs, including the new Zen Phones and a few motorolas.
  • r3loaded - Monday, February 16, 2015 - link

    Intel has a "true" 14nm process though and it's already manufacturing chips for devices that are in consumers' hands right now. Samsung have simply announced 14nm, there'll be a long lead time from this announcement until those chips make their way into a shipping product. Intel's also had plenty of experience with FinFET designs from 22nm.

    The gap is certainly closing but Intel are still ahead. That said, there doesn't seem to be much time left until silicon itself runs out of steam somewhere in the 5nm range.
  • Krysto - Monday, February 16, 2015 - link

    Not a long time. If the 7420 is also 14nm, then the device should appear in March, or April at most. Also keep in mind most 14nm Broadwell machines are just NOW coming out. Not to mention there's no 14nm Atom yet, and in case people around here have forgot Atom is the real competitor to ARM's chips, not Core i5 or even Core M (which costs much more and is still not suited for mobile use).

    And another thing. Intel could barely compete with ARM when it had half a node + FinFET ahead of ARM (22nm FinFET vs 28nm). Now Intel only has that half a node, at best (14 nm FinFET vs Samsung's 14nm FinFET).
  • kwrzesien - Monday, February 16, 2015 - link

    I think Core M *will* be a competitor to ARM in the high-performance tablet market which is just blooming. It's where Intel needs to win vs ARM because they sure haven't managed to penetrate the cheap tablet or any kind of mobile phone market share.
  • Krysto - Monday, February 16, 2015 - link

    Not really. Here's why:

    1) ARM was never in the $1,000 "tablet" market.

    2) Those "hybrids" or whatever you want to call them will just end up replacing laptops. So there's no real "gain" for Intel over ARM here.
  • sorten - Tuesday, February 17, 2015 - link

    Apple (ARM) owns the $900 tablet market. Core M devices will compete with the iPad for consumer dollars.
  • smilingcrow - Monday, February 16, 2015 - link

    Have Samsung actually commented on whether these chips are in production and if so when we can expect to see them in devices or are you just speculating?
  • Kvaern2 - Monday, February 16, 2015 - link

    Should, would could.
    Over the years I've heard too many overly optimistic node shrink proclamations from competing foundries which turned out to be vapor or got hit by huge delays so I'll believe there's parity when I see it in a shipping device.
  • anactoraaron - Monday, February 16, 2015 - link

    Why do folks here keep saying that there's no 14nm atom yet? It's called Cherry Trail, and started shipping to OEMS at the beginning of the year... http://www.anandtech.com/show/8831/intel-shipping-...

    This means that there will be cherry trail devices soon. So enough with that noise
  • Speedfriend - Monday, February 16, 2015 - link

    "And another thing. Intel could barely compete with ARM when it had half a node + FinFET ahead of ARM (22nm FinFET vs 28nm)."

    Geekbench
    Atom Z3795 multi core 3166.
    Galaxy Note with Snapdrgon 805 multi core 2975
    iPhone 6 multi core 2885

    You are right, it could barely compete....
  • Krysto - Monday, February 16, 2015 - link

    Now show me the GPU numbers. I was talking about the SoCs in general. Atom was usually at least a generation behind in GPU performance.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    Who cares about the GPU score? Do you intend to do intense gaming on your phone now?! If the GPU does what it needs to functionally without causing you lag, it's fine. Intel and PowerVR know this all too well.
  • PC Perv - Monday, February 16, 2015 - link

    Where the hell is Atom Z3795? Never heard of it. Sounds like overclocked Z3750/Z3770 for the sake of.. you-know-what.
  • IntelUser2000 - Tuesday, February 17, 2015 - link

    Great, compare chips that go into phone that doesn't need to be heavily subsidized versus one that needs to be subsidized and can't go into a phone. Fair comparison. And the lead for Atom even accounting for that is only 9% at the best. That is BARELY competing.

    ARM chips on the "behind" process were being neck-and-neck with Atom chips, at least when 20nm ARM chips came to completely kick Atom chips to the curb, Atom was on 22nm for quite a while. Now 14nm Atoms are even later than 14nm ARM chips.

    Cherry Trail I remember to be mere 2x in GPU and 5-10% in CPU compared to Bay Trail. It will still lose massively.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    No, Intel's are the only 14nm chips out there and shipping in devices. Jeez you people hate the only company leading the pack...
  • AnakinG - Monday, February 16, 2015 - link

    Can you stop posting if you are wrong about the prediction of Samsung rolling out 7420 at 14nm node in March or April? I'm wondering how confident you are in your claims.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    You do realize Intel had to go from designing for performance to designing for perf/watt, right? That's a massive paradigm shift and it took a while. Now ARM is hitting a brick wall in performance whereas Intel is dropping in power while maintaining the same performance or gaining just a little. ARM has to go a bit more CISC every generation, including out of order superscalar processing (very CISC), to keep up the performance, but, as it turns out, that performance circuitry is electrically and thermally expensive, and ARM's branch predictors are barely more than 40% accurate vs. Intel's 93% accurate BP.
  • levizx - Monday, February 23, 2015 - link

    That's plain stupid. What does parallelism have anything to do with ISA? As a matter of fact RISC is the one with OoO superscalar capabilities early on. Intel x86 had to go (internal) RISC just to compete.
  • III-V - Monday, February 16, 2015 - link

    Congratulations, you have fallen victim to Samsung's and TSMC's marketing.

    Process node names have no relation to physical dimensions, and are purely marketing names. Here's a good article to inform you of the matter:
    http://spectrum.ieee.org/semiconductors/devices/th...

    Intel does in fact have a process advantage. Their transistor performance is a good 3 years ahead of their competition, and they're significantly denser.
  • PC Perv - Monday, February 16, 2015 - link

    Total hearsay and unverifiable assertion. And you talk as if Intel's 14nm (or 22nm, 32nm, 45nm,...) had any meaning other than marketing.. haha.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    not speculation. Undeniable fact: Intel is massively ahead at 14nm. https://www.semiwiki.com/forum/content/3884-who-wi...
  • levizx - Monday, February 23, 2015 - link

    And 14nm is THE ONLY one Intel had a "lead" then. Intel's 22/32/45nm are all LIES by the same standard from your undeniable fact.
  • Jon Tseng - Monday, February 16, 2015 - link

    It's complicated but essentially Samsung 14nm FF is actually 20nm FF (underlying interconnect still at 20nm). So this means its basically somewhere between Intel 22nm FF (2012) and 14nm FF (late-2014), although perhaps skewed a bit closer towards to the former.

    However you also need to factor in that Intel's SoC process (used for mobile) lags its main CPU process by 6-12 months so 14nm hits SoC middle for 2015 with Cherry Trail.

    So depending on how you cut it Intel can still have anything from a 3 year (14FF vs. Ivy Bridge to a 0 year lead (14FF vs. Cherry Trail). All sort of subjective really!
  • smilingcrow - Monday, February 16, 2015 - link

    Cherry Trail is already shipping to OEMs so they do still have a lead there on actual shipping products plus with a denser chip.
    It's hard to make direct comparisons but overall Intel's lead seems to have slipped somewhat.
  • smilingcrow - Monday, February 16, 2015 - link

    http://arstechnica.com/gadgets/2015/01/intel-begin...
  • name99 - Monday, February 16, 2015 - link

    And for all we know A9s are already shipping to Apple... (Apple has to build up MASSIVE inventory to hit their September launches without too much product shortage.)
    That's the problem with "shipping to OEMs" --- it doesn't actually mean anything.

    *Selling* (as in I can order it on Amazon and get it shipped) is a little more meaningful but still not great. God knows there've been more than a few "selling" products where the yield appears to be one device per day.
  • Speedfriend - Monday, February 16, 2015 - link

    A9 is certainly not shipping to Apple yet. It only ships the quarter before launch of the phone as you could see in TSMC's numbers last year. Samsung only likely to hit Apple size production by end of Q2.
  • GC2:CS - Tuesday, February 17, 2015 - link

    What about the iPad Pro ?

    Reportely on sale in April, it could either have an 20nm A9(X) that will be improved to 16nm FF+ for smaller devices, or an 16nm A8X or just the regullar A8X overclocked for a bigger chassis.... Or an 16nm A9X to absolutelly blow away the competition.
  • smilingcrow - Monday, February 16, 2015 - link

    "That's the problem with "shipping to OEMs" --- it doesn't actually mean anything."

    1. Start Production.
    2. Start shipping to OEMs.
    3. OEMs release products.

    We know where we are with Intel with stage 3 due for Q2/15.
    Have Samsung confirmed production or a release date yet?
  • Krysto - Monday, February 16, 2015 - link

    Shipping to OEMs is not the same as "shipping" on the market. Exynos 7420 will likely arrive on the market months before the first Cherry Trail.
  • smilingcrow - Monday, February 16, 2015 - link

    "Shipping to OEMs is not the same as "shipping" on the market."

    Thanks Captain Obvious. ;)
  • PC Perv - Monday, February 16, 2015 - link

    Meaningless until I see products en masse. I bet those tablets will show up later than iPad Air 3 and will be slower. When that time comes, we will hear about Intel shipping 10nmFFF++ chips ahead of competition. The cycle continues..
  • Krysto - Monday, February 16, 2015 - link

    Perfect. This is exactly how it is. At best Intel has kept a 1-year advantage for its notebook CPUs in process technology, and between 0-6 months for its Atom CPUs. But remember Atom, despite being 22nm vs 28nm against ARM (1 year advantage), and also having FinFET (which is basically a 2 year advantage) was merely "competitive" with ARM chips. Atom chips with 0-6 months lead time ONLY will be uncompetitive.

    And that's just in terms of performance/power consumption. Atom probably still need to cost 2x as much as a typical high-end ARM chips to be profitable for Intel, which is why Intel is losing $1 billion per QUARTER subsidizing it.
  • smilingcrow - Monday, February 16, 2015 - link

    Intel's main problem is that x86 is the outsider and that the ARM ecosystem is king so they have an uphill battle.
    Pricing is of course the other main issue but without knowing the exact economics I can't say if Intel's model is sustainable.
  • Michael Bay - Monday, February 23, 2015 - link

    Wishful thinking.

    There were great many proclamations from every other foundry how they`ll catch up to intel, and what? There`s only one brand with 14nm parts technology in the market, and guess who it is.
    No indication of that changing in the short-to-medium term, either.
  • IntelUser2000 - Saturday, February 21, 2015 - link

    Jon,

    the cores are significantly larger on the Intel chips so Intel *needs* that density lead to merely equal it in density. If I remember correctly 22nm Silvermont cores are still larger than 28nm ARM cores. Now, the density advantage for Intel 22nm process versus 28nm ARM is about 20-30% depending on who you talk to.

    Intel needs about 60% lead in density to be on par with competition, so 20-30% advantage that their "14nm" offers is not enough. Cost advantage: ZERO

    Let's assume they do have the cost advantage though. The performance is absolutely behind competition so they'll compete against Chinese manufacturers. Not good for Intel. As I remember though its far worse. Core M isn't that much better than ARM right now.
  • tipoo - Monday, February 16, 2015 - link

    Samsung's 14 nm production is a hybrid design. It uses 14 nm for front end of line, and 20 nm for back end of line. Intel's production is 14 nm throughout.

    http://en.wikipedia.org/wiki/Front_end_of_line
    http://en.wikipedia.org/wiki/Back_end_of_line
  • GC2:CS - Tuesday, February 17, 2015 - link

    They still managed to shrunk down to 14nm interconnects - much higher density.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    Nope, worse density than the 20nm node actually. https://www.semiwiki.com/forum/content/3884-who-wi... It's a consequence of initial FF designs. The leakage current drops like a stone, helping with thermals, but density takes a hit.
  • Klimax - Tuesday, February 17, 2015 - link

    Not shipping yet, just began manufacturing. And there are still qiute differences between 14nm by Samsung and by Intel.
  • trane - Monday, February 16, 2015 - link

    This must mean high performance 14nm at Globalfoundries is not far behind? Given the common platform alliance? I am wondering. Could we see AMD back at 14nm with the new Zen architecture sometime early 2016? And would that not mean process parity with Intel, for the first time - ever? (Yes, I know Intel 14nm != Samsung/GF 14nm, but it sure as hell is a lot better than 28nm!) No, I don't expect them to beat Intel out of nowhere, but I'm optimistic this could put AMD right back in the game.

    7420 looks like a killer SoC. A whole bag of hurt awaits 810 if Samsung sort out LTE and drivers.
  • Krysto - Monday, February 16, 2015 - link

    I sure hope AMD is super aggressive about adopting new process nodes. AMD NEEDS that. It's bad enough that they don't have competitive IPC. When they use older processes, they just become irrelevant in the market. If they can't afford it, then maybe they should consider selling themselves to Samsung, and turn Samsung into a heavy competitor against Intel (wouldn't that be cool?).
  • A5 - Monday, February 16, 2015 - link

    My understanding from past speculation on the topic is that AMD's x86 license is automatically terminated if the company is sold.
  • Krysto - Monday, February 16, 2015 - link

    I know, but surely Samsung will be able to negotiate something with Intel? I don't think world's governments will LET Intel become the only x86 provider. I'm sure they would intervene and start treating Intel as a monopoly. I don't think Intel wants any of that kind of attention.
  • Michael Bay - Monday, February 23, 2015 - link

    Yes, that`s exactly why intel is donoring cash to AMD from time to time. Wait a year, and you`ll see another lawsuit.

    This whole state of x86 competition is truly awful.
  • DanNeely - Monday, February 16, 2015 - link

    Originally their license would've been terminated if they sold/outsourced their foundry business; when they spun Global Foundries off a bunch of lawyers made a large fortune renegotiating the contract with Intel. If AMD itself ended up being acquired by someone else; the same thing would happen.
  • Penti - Monday, February 16, 2015 - link

    Somebody could just buy a large share of AMD's stocks or get AMD to issue new stocks and control the company that way rather than through a merger. Abu Dhabi essentially did and is the largest individual share owner plus founded and is backing Global Foundries.

    Centaur's and Cyrix's licenses expired when VIA bought them, but due to tech (IP) Centaur owns they eventually did get a new license deal. That is a firm in Texas with about 100 employees.

    AMD splitting up isn't really on the table, that would really only come from a bankruptcy and for that to happen they would have to run out of cash and so on first. There's really only small cores, big cores x86 (-64) and graphics in AMD nowadays. But I guess they could essentially be split up into three entities. Ones they have custom ARMv8 CPU's they could just sell of that tech to another company if they wish to exit the ARM space and get some cash. Something major like that isn't really in the process of happening, and would probably be a slow process.
  • Michael Bay - Monday, February 23, 2015 - link

    Not really the same thing. intel could afford someone buying AMD production capability since it had then and has now a very substantial lead. But what would happen if, say, Apple decided to buy AMD and produce x86 chips for itself? Worse yet, if they sell those chips to the likes of Dell and HP? With amount of money they have it`s a trivial deal.

    Of course intel can`t have that.
  • CiccioB - Wednesday, February 18, 2015 - link

    I think it is not that simple. Current x86 technology contains lots of AMD IP (all the -64 part at least) so I can't think Intel can terminate anything so easily as it is not more the unique owner of the technology.

    It the terms were the same as in the past, as soon as Intel terminates x86 license to AMD, AMD could immediately termnate x64 support to Intel. Think about what a mess that will be.
  • smilingcrow - Monday, February 16, 2015 - link

    "This must mean high performance 14nm at Globalfoundries is not far behind?"

    Not at all as high performance is a very different beast so don't assume anything.
    Until they officially announce it's in production or actually shipping who knows when it will be ready.
  • Krysto - Monday, February 16, 2015 - link

    Snapdragon 810 can't compete with a 14nm chip. Game over for Qualcomm (at least concerning Samsung's devices, but Samsung COULD sell it to others, too - and it should!). Also Snapdragon 810 seems to throttle heavily, much more than previous Qualcomm chips, so the current benchmarks you see on other sites showing the 810 as being "equal" in heat dissipation to 805 or winning in performance are misleading.
  • sonicmerlin - Monday, February 23, 2015 - link

    No one can compete with qualcomm's modems.
  • SydneyBlue120d - Monday, February 16, 2015 - link

    I'd like to know if only the AP is made with 14nm process or the Exynos modems coupled with the AP will be made at 14nm? What do You think?
  • Penti - Monday, February 16, 2015 - link

    It might as well be used with Intel or Marvell modems.
  • savagemike - Monday, February 16, 2015 - link

    It's great to have the increased competition and to have that you need comparable technologies. So bully for the announcement.I also agree that 14nm has become a bit of a marketing term and so 14nm might not be the same as 14nm.
    All that aside I see several comments talking about how Intel took a year to get 14nm in the market due to reliable production issues, and so lost something of their lead. However with that in view what we have here is Samsung saying it will be doing this. What we don't have is them saying they have good production at scale. This stuff is the most complex manufacturing the world has ever known. It's quite possible that Samsung too will have delays in actually getting it to market.
  • Gaugamela - Monday, February 16, 2015 - link

    Samsung/GF supposedly surpassed the production issues. That's why Apple and other ARM manufacturers such as Qualcomm are looking at them to make their new chips instead of TSMC.

    This is great for everyone. Hopefully next year we will get AMD x86 chips in a 14nm FinFet process thus increasing the possibility of competitive AMD CPUs for the first time in the last 5 or 6 years.
  • smilingcrow - Monday, February 16, 2015 - link

    More like 9 years I'd say.
  • Guest8 - Monday, February 16, 2015 - link

    By surpassed you mean charge their customers per wafer instead of per chip. Only then do I see them having yield issues worked out. I have a feeling these new processors are going to be pricey probably 2 times what the previous generation was assuming the yields can get up to 50% TSM starting shipping A8 with 50%ish yields so yeah that's one way to go about it. Intel would never ship anything with yields this bad.
  • Michael Bay - Monday, February 23, 2015 - link

    Apple is looking at Samsung not because of better process technology, but because TSMC simply couldn`t deliver the volume of chips(what was it, A7 or A8?) they required. Nobody but intel or Samsung could produce something at such volumes, and Apple considered Samsung to be the lesser of two evils.
  • witeken - Monday, February 16, 2015 - link

    "and so lost something of their lead"

    The jury is still out on that one: Intel might start HVM of 10nm at the end of 2015, less then 2 quarters after TSMC started HVM of their 20+FF node. So the TTM that Intel lost at 14 might be regained at 10 (Intel's 10nm will be comparable to TSMC/Samsung's 7nm). But Intel is very quiet about 10nm, so we'll have to see what happens with Intel's lead in the long run. I'm betting on a status quo of 3-4 years.
  • CiccioB - Wednesday, February 18, 2015 - link

    Technically it appears Intel can start 10nm at the end of the year.
    Problem is how much money that will cost.
    Remember that to be profitable (and winning) on the market you do not have only to offer a good product, but it mainly has to be on par with competition about the price.
    Intel already took a big hit in their finance with 22 and 14nm PP. Added costs, delays and the need to subside 22nm mobile chips because they clearly misses all those "fancy advantages" Intel proclaimed they could offer. We are speaking of something as 6 billions of extra costs for 14nm + 8 billions in two years for try to give a meaning to 22nm in mobile market. Clearly failing in that. Going so soon for 10nm would mean another financial bleed, as 10nm in desktop/server maket is not needed (no competition at all there) and in mobile market Intel is still having big issue with its not so really competitive chips. Agt 10nm they would costs really too much even though could be on par with the competition.

    When Intel claimed the "supremacy" of their 22nm over the competition, the rest of the world was still producing at 40nm. I can remember all tests (here too) where done comparing early samples of Bay Trail with some random A9 (40nm) SoC already on the market for a year at least.
    The time it took Bay Trail to enter in a device, 28nm ARM SoCs were already in production. The time someone could buy a Bay Trail device, 28nm ARM Socs were still into new devices. Intel clearly could not compete with a 22nm chip vs ARM 28nm chips. Not in term of perf/Watt and surely not for cost.
    I can see a worse situation here: Intel has tried (again) to be early on the mabile market with a new Atom (the first chip produced at 14nm) while all others were 28nm. Intel did not probably expected Samsung to go from 28 to 20 and then 20+FinFet (called 14nm or every MyTastyYogurt for what that matters) in less that 2 years.
    This 14nm "MyTastyYogurt" PP is quite a manace for them. It means that Intel won't be competitive again for a single day on the mobile market. Despite the huge quantity of money Intel put in their advanced PP. Be it 3 months, or 6 or 9, soon enough (well before 10nm can be ready) all other ARM SoC producers will saturate TMSC/Samsung 14/16nm capacity with billions (not just few millions x86 mobile chips) of ARM SoCs flooding in the mobile market. A nightmare for a company that knows that it needs to be at least 1 PP ahead to be somewhat competitive on thechnical field but still well behing for costs.

    If 16/14nm production capacity will be big enough at the end of the year, I think it will put an end to Intel hope (and investments) in the mobile market. At least on any device that does not have Windows on it.
    Let we see how long this trailing Wintel simbiosys will last. We have already seen that to be productive one does not really needs Windows and all the associated costs (that means bigger devices, bigger batteries, higher costs). With ARM SoC becoming more and more powerful, it is possible for them to take a quite big piece of the low end x86 market even without (or even because there's not) Windows on those devices.
  • costeakai - Monday, February 16, 2015 - link

    samsung vs the rest of the world; i love it , i love you south coreans.
  • DanNeely - Monday, February 16, 2015 - link

    The real question is how quickly can they scale production up. This is the first second generation double patterning process; working out all the problems with DP has been responsible for Intels 14nm, and everyone else's 20nm processes being badly delayed and having very slow volume ramps. If Samsung is able to quickly ramp this process to high volumes it suggests we should still be able to count on a few more iterations of new processes every two years. If it ends up being a very slow trickle like the first gen DP processes are, it will strongly suggest that the time to market for new processes is going to be permanently stretched going forward.
  • tipoo - Monday, February 16, 2015 - link

    Samsung's 14 nm production is a hybrid design. It uses 14 nm for front end of line, and 20 nm for back end of line. Intel's production is 14 nm throughout.

    http://en.wikipedia.org/wiki/Front_end_of_line
    http://en.wikipedia.org/wiki/Back_end_of_line
  • PC Perv - Monday, February 16, 2015 - link

    You seem desperate, posting same links twice. Haha.

    I understand.
  • IntelUser2000 - Tuesday, February 17, 2015 - link

    Does anyone REALLY believe that Intel is 3.5 years ahead of competitors because they shipped 22nm with FinFETS 3.5 years before others?

    They aren't even 3.5 years ahead in timeframe wise:

    Ivy Bridge: Mid 2012
    Atom 22nm: Q3 2013
    Samsung 14nm FF: Q2 2015

    That's only 3 years. Then again, 22nm Atoms had hard time competing against 28nm ARM parts. At this point I would bet that Intel is fudging numbers about process and leadership more than ARM guys are.
  • patrickjp93 - Tuesday, February 17, 2015 - link

    Too bad Samsung's 14nm process is bigger than its 20nm process... https://www.semiwiki.com/forum/content/3884-who-wi...
  • pSupaNova - Wednesday, February 18, 2015 - link

    Or you could say Samsungs 22/20nm was 50% smaller than Intel's 22/20nm process...

Log in

Don't have an account? Sign up now