Intel Broadwell Architecture Preview: A Glimpse into Core M
by Ryan Smith on August 11, 2014 12:01 PM ESTPutting It All Together: Small Core M
Next to power constraints, the final element of Intel’s fanless challenge is the size of the SoC itself. Sub-10mm thickness doesn’t just put constraints on the heat capacity of the device but it also constrains just how large an SoC and its supporting circuitry can be. As a result Intel has focused on making Broadwell-Y the smallest Core processor yet, making the entire SoC under 500mm2 in size.
As was the case with power, reducing the size of Broadwell-Y is a multi-faceted effort. The 14nm process plays a big part here, allowing for one of the smallest Core CPU dice yet. At 82mm2 the Broadwell-Y CPU die is some 37% smaller than the Haswell-Y CPU die, none the less packing a dual-core CPU and a full GPU slice.
With such a small die Intel was in turn able to reduce the size of the entire SoC package through the combination of the reduced die area and further optimizations to the packaging itself. Haswell-Y’s already small ball pitch of .65mm was further reduced to just .5mm, producing a package with Intel’s smallest solder ball pads yet. Intel considers the reduction in the ball pitch to be the key change that allowed Broadwell-Y to be so small, as they were already pad-limited on Haswell-Y despite having ample excess packaging even after taking the CPU die’s larger size into account. As a result Broadwell-Y takes up almost 50% less surface area (XY) than Haswell-Y.
Intel has also made a number of changes for Broadwell-Y to reduce the Z-height of the Y SoCs, as even 1.5mm for the SoC starts to become a significant design constraint in a sub-10mm device. Again owing to the 14nm process, the Z-height of the Broadwell-Y die itself is down to 170um. Meanwhile the Z-height of the substrate has been cut in half from 400um to 200um, which accounts for nearly half of the total reduction in SoC Z-height.
The final element in reducing the SoC Z-height, and what’s likely the most unorthodox change for Broadwell-Y’s packaging, is Intel’s 3DL inductors. The 3DLs aren’t just to improve energy efficiency as we discussed before, but they are part of Intel’s efforts to reduce the SoC size. For Broadwell-Y the 3DLs are on their own PCB on the back of the SoC, extending well below the back of the package. To accommodate this, logic boards housing Broadwell-Y will have a hole in them where the 3DL PCB would be in order to allow the complete SoC to fit. Because there are no BGA connections here this change isn’t quite as radical as it first appears, but it’s a very good example of just what lengths Intel was willing to go to reduce the package Z-height.
All told then, the combination of these space optimizations has reduced Broadwell-Y’s Z-height by nearly 30%, from 1.5mm on Haswell-Y to 1.04mm on Broadwell-Y (3DL PCB not included). By bringing Broadwell-Y’s thickness under 1.1mm, the SoC is now no taller than the other common components on a logic board (e.g. RAM), meaning the SoC will no longer stick out above the other components, which is useful both for saving space and for allowing simpler (flatter) heatsinks.
Finally, the smaller size of the Broadwell-Y package will also have a knock-on effect on the size of the logic board, further feeding into Intel’s goals to get Broadwell-Y into smaller devices. Intel tells us that the size of a complete platform (logic) board for Broadwell-Y has been reduced by roughly 25% as compared to Haswell-Y, allowing Broadwell-Y to better fit into not just thinner devices but overall smaller devices too.
Facebook
Twitter
RSS
158 Comments
View All Comments
IntelUser2000 - Wednesday, August 13, 2014 - link
Broadwell Y's CPU portion is on 14nm but the PCH is on 32nm.Not to mention that Intel's 22nm process isn't that small compared to Foundries 28nm. 28nm Foundry is ~30% larger than Intel 22nm not 100%.
20nm Foundry is really a half node jump from 28nm considering density improves but performance much less. 16nm is another half node since density barely changes but perf improves.
Really, it can only be judged as a product.
mkozakewich - Thursday, August 14, 2014 - link
The 'myth' was that x86 was a horrible power-sucking pig. It was shown that it was possible to at least get close to ARM processors.Meanwhile, Intel's chips are NOT as low-performance as Apple's. The chip in Surface Pro 3 is about 2x-3x faster, and these should be about the same. With this year's Apple chip, imagine it to be 2x faster or so. Meanwhile, they'll probably both take the same amount of power. SoCs these days are running multiples of Watts. Even Intel Atoms used to take over 10, but now a full Intel Core computer can run under 6 Watts. It would be really interesting to run all these numbers and give a complete report on the state of mobile vs. notebook processing. They seem matched on power usage, but ARM chips are far cheaper and run at far less performance.
peevee - Monday, August 11, 2014 - link
Finally, a long awaited fanless design. I don't care about thickness that much as about energy waste on the heat and the fan to dissipate the heat.But... I have 2008 MacBook Pro with 2.4GHz Pentium M. If they achieved fanless design by bringing frequency down to, say, 1.8GHz, I am not interested (given that IPCs are not that different for real world applications). For me to upgrade, I want it to reach 3GHz, even if for a second and in a single thread, when starting applications for example. Anything below that is not a noticeable upgrade, and below 2.2GHz or so will be downgrade in practice.
And biggest problem of Intel is not how thick they processors are, it is Microsoft - with Windows 8 (and 8.1) being so unbelievably awful (yes, I do own it for a while). Krzanich should call Nadella immediately and tell him to fire Larson-Green, or they both are going down.
MikhailT - Monday, August 11, 2014 - link
You do realize this is for tablets and low power laptops (Chromebook/netbook style) only, right?It's not coming to MBP for any time soon unless you're talking about something else entirely, like the MBA, which also is not going to get M either because it'll be too big of a regression on the performance.
I don't think we'll see any Macs with Core M.
ZeDestructor - Monday, August 11, 2014 - link
You forget about IPC. Last I checked, compared to a Core 2 CPU at equal clock speeds, a Sandy Bridge CPU is 50+% faster on average, and Haswell is a further 15+% faster on top of that, all the while using less power.Krysto - Tuesday, August 12, 2014 - link
Since this is the FIRST core to be "fanless", they're probably squeezing a lot of stuff to make that work, and it probably still overheats. I wouldn't be too excited about it until we see how it does in actual devices, both from a power consumption point of view, but also a performance one (because if performance didn't matter, then we'd all just use ARM chips, no?).It would be laughable if Denver, which already beats mainstream Haswell Celeron, would be in range of Broadwell Y in performance, but still more energy efficient and with much better GPU performance.
UNCjigga - Monday, August 11, 2014 - link
Time to spin up the 12" iPad Pro rumor mill again...but would Apple really need to build a device that runs 64bit iOS *and* OS X?ilt24 - Monday, August 11, 2014 - link
Apple doesn't seem all that interested in adding touch screen capabilities to OSX.isa - Monday, August 11, 2014 - link
OK, I'm confused, so any help appreciated. I want to replace my old desktop replacement laptop with a Broadwell equivalent. For example, the Broadwell version of an HP Envy 17t. What flavor of Broadwell am I waiting for? The Y flavor? U? H? Something else? thanks.mapesdhs - Monday, August 11, 2014 - link
I would ask the opposite: why do you need a replacement at all? What is it your current
device cannot do, or does poorly? Once the newer products are out, find out which one
solves those issues at the lowest cost. This focus on the jargon side of computing tech
is the absolute worst aspect of how consumer computing has evolved since the 1980s.
Ian.