Coming in 2014: Intel Core M
Intel spent some time with us explaining its move to Broadwell and the 14nm process technology.
The era of Broadwell begins in late 2014 and based on what Intel has disclosed to us today, the processor architecture appears to be impressive in nearly every aspect. Coming off the success of the Haswell design in 2013 built on 22nm, the Broadwell-Y architecture will not only be the first to market with a new microarchitecture, but will be the flagship product on Intel’s new 14nm tri-gate process technology.
The Intel Core M processor, as Broadwell-Y has been dubbed, includes impressive technological improvements over previous low power Intel processors that result in lower power, thinner form factors, and longer battery life designs. Broadwell-Y will stretch into even lower TDPs enabling 9mm or small fanless designs that maintain current battery lifespans. A new 2nd generation FIVR with modified power delivery design allows for even thinner packaging and a wider range of dynamic frequencies than before. And of course, along with the shift comes an updated converged core design and improved graphics performance.
All of these changes are in service to what Intel claims is a re-invention of the notebook. Compared to 2010 when the company introduced the original Intel Core processor, thus redirecting Intel’s direction almost completely, Intel Core M and the Broadwell-Y changes will allow for some dramatic platform changes.
Notebook thickness will go from 26mm (~1.02 inches) down to a small as 7mm (~0.27 inches) as Intel has proven with its Llama Mountain reference platform. Reductions in total thermal dissipation of 4x while improving core performance by 2x and graphics performance by 7x are something no other company has been able to do over the same time span. And in the end, one of the most important features for the consumer, is getting double the useful battery life with a smaller (and lighter) battery required for it.
But these kinds of advancements just don’t happen by chance – ask any other semiconductor company that is either trying to keep ahead of or catch up to Intel. It takes countless engineers and endless hours to build a platform like this. Today Intel is sharing some key details on how it was able to make this jump including the move to a 14nm FinFET / tri-gate transistor technology and impressive packaging and core design changes to the Broadwell architecture.
Intel 14nm Technology Advancement
Intel consistently creates and builds the most impressive manufacturing and production processes in the world and it has helped it maintain a market leadership over rivals in the CPU space. It is also one of the key tenants that Intel hopes will help them deliver on the world of mobile including tablets and smartphones. At the 22nm node Intel was the first offer 3D transistors, what they called tri-gate and others refer to as FinFET. By focusing on power consumption rather than top level performance Intel was able to build the Haswell design (as well as Silvermont for the Atom line) with impressive performance and power scaling, allowing thinner and less power hungry designs than with previous generations. Some enthusiasts might think that Intel has done this at the expense of high performance components, and there is some truth to that. But Intel believes that by committing to this space it builds the best future for the company.
As explained to us by Sanjay Natarajan, VP and Director of 14nm Technology Development, the jump from 22nm to 14nm is not as simple as it might first appear. There are several measurements of transistor construction that can be modified when moving to a 3D design. The fin pitch is the measurement between peaks and has been lowered from 60nm to 42nm, a scaling rate of 0.7x. Intel was able to constrict the fins a bit more to allow for more current performance while reducing the number of fins for improvements in density and lower capacitance.
Interestingly, Intel commented on the issue of the 14nm naming for this process in that no single pitch measurement shown to us was at 14nm. Instead, starting with the 90nm process design, no silicon manufacturer uses a node name that accurate reflects a specific pitch density and instead we are simply seeing a naming convention at work. Intel is basing the 14nm name on the 0.65x scaling rate seen with the combination of pitch adjustments on the tri-gate design, thus going from 22nm to 14nm is a 0.65x scaling rate. When pushed on if this represents the minimum drawable size of the technology, Intel declined to claim that was the case.
As an example implementation with this process change Intel showed an SRAM memory cell that shrunk from 0.108 um^2 at 22nm down to 0.058 um^2 on 14nm technology. That results in an area scaling rate of 0.54x.
The 14nm process technology provides improved performance and leakage and based on this graphic allows Intel to continue pushing forward with existing markets while addressing new ones. While maintaining the same leakage levels as previous process nodes Intel should be able to improve clock speeds for products in server computing and enthusiast designs. Client and mobile computing will see dual advantage of some slightly higher clock speeds (or maintained clock speeds with higher IPC) but also drop in leakage power allowing for longer battery life and new system designs. Most importantly, down into the 0.001x leakage levels Intel continues to push for 14nm to pave the way into the smartphone markets.
The key benefit across the board is performance per watt and Intel claims that the 14nm tri-gate process allows Broadwell-Y to see a 2x improvement over Haswell-Y in that field. The 2nd generation tri-gate transistors with better scaling at lower voltages along with the better than normal area scaling provided by this process shift help make this happen. But a lot of work was also done by the architecture design team working hand in hand with the process team to optimize the core for dynamic capacitance reductions (clock cycle switches).
Logic area is the combined area of gate pitch and metal pitch and this has been scaling at the rate of 0.53x since the 45nm process used on the Core 2 processor, allowing for more densely packed transistors. Interestingly, according to Intel, competing foundries have actually been a bit better over time in the density measurement but at the expense of being noticeably later in release. Just think back to the 40nm/45nm days as well as the current state of 22nm/20nm technologies out in the market to see this as being the case.
This time though it looks like Intel might be bypassing the competition in both timing and density, based on papers published by TSMC and the IBM Alliance. While GlobalFoundries and TSMC continue to attempt shipping their first FinFET designs in the market Intel has nearly ready to put products on the shelves based on its second generation of FinFET design. This is good news for Intel and associated products but is bad news for all other competing solutions that are dependent on third-part manufacturing.
Though the company clearly wasn’t going to be quoted with yield levels of 14nm wafers, Intel says that 14nm product is in a “healthy range” and has improvements coming down the line, just as they did with the 22nm process. The simple graphic above indicates that 14nm is slightly below where 22nm was this point, but that isn’t a surprise to those working on the platform. Intel is confident that they will be able to bring 14nm in line with the results seen with 22nm production which is actually Intel’s highest ever yield after tweaks and changes after implementation.
Both the process and the lead product (Broadwell-Y) have been qualified and are in volume production, shipping product to customers with a target release of holiday 2014.
For the x86 tablet space, the
For the x86 tablet space, the Broadwell is an improvement, but look what Nvidia is doing with ARM (Custom Denver ARMv8 ISA, and Reference ARM quad cores)at 28 nm, and Kepler graphics, the graphics on the Broadwell better bench within range, and who knows how many Kepler GPU cores Nvidia could add with a die shrink. Apple’s A8 is just around the corner. For sure Broadwell will make it into some High End High priced OEM SKUs, but the Nvidia K1, and its graphics will lead in the low cost Tablet Market, and if the K1’s graphics and GPGPU(Via OpenCL, and CUDA) can/and are using more applications that are able to take advantage of GPGPU, then Intel will not be able to make a dent in the mainstream tablet market. AMD has its x86 variants, with AMD graphics, and a custom ARMv8 ISA based variant in development. The application ecosystem in the mobile/tablet market does not have the legacy code that made x86 as necessary for the desktop market, and the custom ARMv8 ISA SOCs from Apple, and Nvidia, just wait until Samsung and GlobalFoundries get that 14nm process going, and even before there will be 20nm ARMv8 ISA (custom wide order superscalar) products out, with their already low power needs. Nvidia is going to be out front of the mainstream tablet/chromebook market with low cost SKUs, with graphics that no one currently can beat, Apple will definitely have to Pull a wizard, and get some GPU improvement, Intel graphics is just not there yet, and those price points are going to have to be low, on any SKU if Intel wants in on the mainstream mobile/tablet market, contra revenue will not work in the mobile market, low prices will have to remain low.
Edit: To Add,
Hats off to
Edit: To Add,
Hats off to Intel on their 14nm process, but this is only available in 2 cores, and 4 cores are not for a while, Samsung’s and GlobalFoundries'(licensed from Samsung) 14nm process is not as good with the pitch, and traces scaling, and Intel has apparently reduced the need from 3 finfets per circuit down to 2, so will Intel pass the savings on, or be able to compete more in the mobile mainstream market, Apple is sure pissed off with the delay, and more rumors are circulating about ARM based MacBooks. Bigger and Taller Fins, sounds like the big fin era is about to begin for Chips, just like it did for cars way back in time. AMD, now is the time to get your new x86 microarchitecture rework out the door, and take advantage of the Intel delay, to get some better graphics, and x86 into Macbooks before the inevitable, Apple A8s, or A9s, give Apple enough power to go all ARM, except for mac pros! But even for the Mac Pro there will be competition for the Mac Pro’s server/workstation chip business beginning in 2015, from a non ARM, non x86, based competitor.
The SOC low bidder wins in mobile, and the 14nm circuit cramming contest may still have the same winner, but the price and lack of GPU power on those SOCs may make that process node victory hollow.
The broadwell SoC picture in
The broadwell SoC picture in the last slide looks strange. If the cpu cores take up the upper right quadrant, then it looks like there is only 3 structures there. There is something in the middle that looks similar to the other 3 structures but definitely different. I would expect that portion to be part of the gpu though, given the size ratios. Is this a 3 cpu core chip?
No I think it’s Intel trying
No I think it’s Intel trying to make the best of the process yields, and moving things around to fit in a space constrained tablet form factor, that bottom hole in the motherboard requirement looks to be a good idea for extra air flow, and making the motherboard’s thickness more of a non factor in getting Broadwell into tight spaces, while getting some components back of off the CPU die. Its more like package on package, with some more package space underneath, and others may do the same, although the entire industry may be moving to a Mezzanine module type of packaging with stacked RAM, CPU, and GPU sharing the module’s wide bus/high bandwidth interconnect fabric. For Mobile(and PC also, though not as necessary), the module(thin tablet form factor) will rest in a cutout from the main board allowing the module to take advantage at least a motherboard’s thickness for extra space, and ventilation/heat dissipation from both sides of the package/module. The cutout lined with mezzanine connector pins/sockets placed around the entire circumference and taking up less vertical space. Intel still needs to get its GPUs competing with Nvidia, and AMD graphics, and Intel’s pricing will keep it out of the lower cost products.
Probably the voltage
Probably the voltage regulators.
Broadwell M…finally a replacement for my Pentium M laptop. 😉
Not kidding. Still using it to this day.
Pentium M was ahead of its
Pentium M was ahead of its time, but it’s pretty limited in today’s world without a dGPU. I assume your laptop also has an NVIDIA (6000-series) or ATI (8000/9000-series) GPU to complement it?
Oh yeah, it’s pretty darn
Oh yeah, it’s pretty darn slow by today’s standards. But just fine for browsing the web or editing the occasional office document.
It does have intel onboard graphics and it can just BARELY play a 1080p .TS file. But I know what it’s limits are.
BTW, it’s a HP nx5000 that came with a Celeron processor. It’s gotten a PentiumM CPU upgrade, RAM upgrade, and I upgraded to a 1440×1050 display. I tried adding a mSATA SSD using a 2.5″ IDE adapter from Addonics but it didn’t work. I’ve polished this turd as much as possible. 😉
I believe you,
It’ll be
I believe you,
It’ll be replacing my 2008 Core2Duo t5500 1.66Ghz laptop
With WinXP.
Intel is moving away from the
Intel is moving away from the high end (enthusiast) with good reason.
The thing is even though Intel is managing these die shrinks, they are resulting in more and more unused silicon on the die. Particularly noted on the extreme high end such as the 5960X with 4 (1/3 of total) disabled cores compared to 2 (1/4 of total) for it’s predecessor.
This also means that mobile components (which is clearly the focus) with smaller die sizes that can more easily cut portions from a wafer will be more able to see these improvements without the wasted silicon.
It’s going to be weird having ultra fast chips that can only be made as big those currently in phones.
Intel certainly still needs to alter it’s architecture for performance in the mobile market (particularly with their GPU), but they are still way ahead when concerning process node. As focus shifts more and more we are sure see their architecture become more specialized for mobile.
At this point, I’m not sure if architecture or process node will be more important, but Moore’s law is as good as done.
Well, if Intel can make a
Well, if Intel can make a Broadwell laptop CPU with 6, or more cores then, maybe I will give it a look, I like a standard form factor laptop with a Quad core i7, but if Intel is too obsessed with Thin And Light( read under powered and poorly preforming) then Haswell will be the last laptop CPU for me to consider, and Haswell i7 CPU based laptops will be a good deal, whenever the 4 core Broadwells arrive, If Intel gets more Cores into a standard laptop SKU designed for the non thin and light market, that may make me reconsider. Intel’s thin and light UltraBook obsession, was an unwelcome distraction to the Desktop replacement laptops, that were touted as being just around the corner, and I had hoped that Intel would get more mainstream laptop SKUs, with a few 6 core laptop CPU variants, and more acceptance of thunderbolt on laptops, other than Apples costly products. If AMD could get a future ARMv8(custom ARMv8 ISA based core, not ARM reference core) server/portable workstation SKU I’ll seriously look at a portable workstation ARM based product, but the Xeon based portable workstations are out of my price range.
For sure AMD could take its IP/know how and create a very good custom 8, or more, CPU cores, ARMv8 portable workstation SKU, and Firepro graphics at an affordable price point.
I am not sure what you are
I am not sure what you are talking about with wasted silicon. The extreme edition parts are essentially salvaged Xeons. If they have some defects in the cpu cores or in cache area, they can just disable the components and sell them as extreme edition parts rather than fully functional Xeons. If you have an 8 core chip with 30 MB of cache, you are certainly going to get some with defects that can not be worked around by lower level redundancy. The extreme edition parts are such a small market share, that they may be able to supply it completely with salvaged parts.
For higher performance segments, the die area will not be as small as broadwell-y. The integrated GPU will be made much larger. A current high-end gpu is more than 400 mm2 (I have seen 551 mm2 for the Nvidia gtx titan). Broadwell-y is only 82 mm2, with an integrated gpu. Probably around half of the 82 mm2 is the IGP. Other large segments will be external IO. The actual cpu cores are tiny. Since GPUs scale almost linearly with the amount of hardware, a high-end gpu will not get smaller, they will just include more hardware for more performance.
You seemed to have missed my
You seemed to have missed my point entirely. I know that they are salvaged Xeons, but that doesn’t change that fact that there is a higher percentage of the die being disabled at approximately the same price.
Since you want to bring GPUs into the discussion they further my argument, but your comparing apples and oranges as far as die size the titan is a 28nm node gpu, this is a 14 nm node CPU using finFET.
Considering we still aren’t even expecting the next GPU release to even be 20nm it seems they are pretty clearly having production problems getting such large areas in production on the finFET nodes, which makes sense based on my implications particularly since (as you happened to note) GPUs scale differently.
Consequently, Intel will gain
Consequently, Intel will gain power saving information, and apply it in the enthusiasts CPUs.
An enthusiasts’ chip is too complex to apply power saving modules, It makes more sense to start with a small chip first.
Power saving in an enthousiast chip, could lead to cooler running chips, that could be overclocked even more.
Broadwell seems nice if you
Broadwell seems nice if you want a laptop, but for desktop it looks as a waste of time and money if one already has sandy bridge or newer, maybe even nehelem or newer
One would think if they can’t improve IPC they would ad least use higher transistor density and reduced power output to make 6 core CPU for $300 i.e. mid range market. But alas 6+ cores are to remain enthusiast only at ~$1000 price range. Greedy bastards.
actually Haswell e will have
actually Haswell e will have 2 6 core processors 1 will be the 5930k around $650 and the other 6 core 12 thread 5820k expected to be around $350. This time around the 8 core processor 5960x will be at the $1000 mark.
It wouldn’t call it greed , more like lack of competition to bring prices down. intel has no competitors. Its just business…if people are willing to pay $300 – $1000 for a good cpu then why not charge it. I would love prices to be cheaper but until AMD or anyone else brings a serious effort and can compete directly with intel..thats not going to happen.
WinTel 4 life !
WinTel 4 life !
So 14 nm is 42 nm or it could
So 14 nm is 42 nm or it could be be 90 nm for some chip maker?hahaha.so in the end they haven’t shrunk at all compared to pre 90 nm era.wow so basicly this means and isn’t as far as user think.no wonder ms xbox one use amd. Nm are irelevent.so it all boil down to programmer!I love it,this means if and focus on cooling solution?and isn’t has far as people were led to believe.OK now I need an Xbox one