Samsung To Ship 14nm FinFET Chips For Apple, Qualcomm And AMD Late Next Year
At a press conference, the president of Samsung's semiconductor arm declared that the company's profits should improve soon once it starts using its next-generation process technology. Samsung is expected to build chips using its 14nm FinFET (3D transistors) process for companies such as Apple, Qualcomm and AMD by the end of next year, which is right around the time Apple is supposed to launch its A9 processor.
The 14nm process should bring 35 percent less power consumption and have 20 percent increased performance with a 15 percent smaller die area. It's getting more expensive to build chips on newer and smaller process nodes, and one way to reduce the cost of the chips is to reduce their size, even if that means a lower increase in performance.
This year, Apple is already using the 20nm process from both Samsung and TSMC, although it's only using Samsung's foundry for 30 percent of the A8 chips. According to ZDNet Korea's sources, it looks like Apple will use Samsung's foundry almost exclusively for the 14nm process next year.
TSMC should have its 16nm FinFET process ready by then too, but for some reason Apple doesn't seem as excited about it. It's doubtful that the difference between 14nm and 16nm processes made Apple decide to go only with Samsung. Knowing Apple, however, it's possible the company thinks that difference can be an impactful one from a marketing point of view.
It's interesting to see that both Qualcomm and AMD will be Samsung's customers, too. Qualcomm will switch to 20nm just half a year before going to 16nm, so it was slightly unlikely the company would switch to a smaller node so soon, but that's exactly what Qualcomm needs to do if it wants to stay competitive. Going with 28nm as early as it did for Snapdragon S4 (back in 2012) helped the company a great deal in the popularity of its chips, which quickly beat Nvidia's 40nm Tegra 3 in energy efficiency and performance.
Over the last few years, AMD has seen a decrease in revenue, market share and even some relevance in the market, with the majority of notebook customers preferring Intel-based machines. While Intel has gotten to that point through a combination of more advanced micro-architecture designs that perform better than AMD's CPU, being roughly a generation and a half (22nm Tri-gate vs 28nm planar) ahead of AMD in process technology has also made it much easier for Intel to keep that advantage. New processes can greatly reduce the power consumption of a chip compared to the previous generations.
Going with 14nm FinFET almost at the same time as Intel (which won't have that many 14nm Broadwell chips in the market until close to mid-2015 anyway) could help AMD recover some of its image in the notebook market by becoming much more competitive with Intel's chips.
Nvidia also promised to move to TSMC's 16nm FinFET process as soon as possible, at least for its next-generation Tegra chip, so late next year we should see some very interesting comparisons between all of these companies' processors.
Follow us @tomshardware, on Facebook and on Google+.

Could be the beginning of TSMC going the way of UMC. The struggle with 28/20nm was a sign and the huge delay is going to hurt their business in the long run. Thankfully they developed 16nmFFat the same time as 20nm, but that can easily be negated if both AMD/Nvidia just pass 20nm entirely as TSMC has nothing on the horizon below 16nm close to being ready.
What horrible writing. Good grief, how could this pass an editor. It's completely unclear, and borderline misleading.
For the first sentence, compared to what????? To the 28nm most companies are on? Nope. What then? Compared to what the node would be like if it were not a FinFet? Terrible, terrible writing.
It's not more expensive to build chips on newer nodes. It's generally cheaper. Again, learn how to write clearly. It's more expensive for the same die area, but less expensive per transistor. Since a direct shrink dramatically lowers the size, smaller lithographies cost less once it gets past the low yield phase. 20nm is one where this holds less true.
That's probably what you mean, but it's written so poorly, no one can figure that out unless they already know it.
First, 14nm is not a meaningful term, and doesn't reflect the actual lithography. It's just a name the company gives the process. They try name it as if it were the actual lithography like it was years ago, to indicate the size reduction, but it's not an absolute term. Intel's numbers are always fudged, a little, so that, for example, their 22nm was really more like 24nm in terms of shrinkage. But, they can call it what they want. The point is, this isn't necessarily better than the 16nm process. It could be, but each company names it what they want.
Keeping in mind Global Foundries licensed the technology, and is calling it, 16nm, you get a pretty good idea of how there is some room to play with the names.
While it's interesting AMD will be using this, it's also important to not read to much into it. Since their next generation of 'big' processors isn't out until 2016, and then almost certainly on GF (which works fine for Samsung, since they can switch relatively easily between the two since the processes are so similar). Carrizo is on 28nm, so there will not be any large processors made at Samsung in 2015. This leaves the smaller markets. Since we know Puma's successor is going to 20nm, and the R9 390 is also 20nm, it's not clear what they would switch so quickly to Samsung's process in 2015. More likely, they will be using the process, but in 2016. Of course, anything is possible, but there aren't too many scenarios where their bigger product lines would switch so quickly, considering AMD has had to cut R&D, and does not in any way want to be on the cutting edge of working with new manufacturing processes, except in limited ways. Rory Read called it "burning cash". So, I'm skeptical we'll see any volume from AMD in 2015 on this process.
Of course, I could be wrong.
Well maybe he could've mentioned that the performance numbers are quoted from Sammy's president....
Doesn't look like a sound strategy to me.
maybe for their APU or ARM based server but for GPU i think AMD most likely stay with TSMC. there are rumor that Tonga will be manufactured by GF but in the end it was made by TSMC as well. and so far Samsung only dealing with making SoC type of chip.
For the first sentence, compared to what????? To the 28nm most companies are on? Nope. What then? Compared to what the node would be like if it were not a FinFet? Terrible, terrible writing.
Thank you for your comment. Companies usually compare a new process node with the "previous one", that one being 20nm in this case. It wouldn't make sense to compare to any other existing nodes on the market, since there are usually more than one in use, and would create too much confusion.
It's not more expensive to build chips on newer nodes. It's generally cheaper. Again, learn how to write clearly. It's more expensive for the same die area, but less expensive per transistor. Since a direct shrink dramatically lowers the size, smaller lithographies cost less once it gets past the low yield phase. 20nm is one where this holds less true.
I said it's "more expensive to build chips", not transistors. On newer nodes the transistors would of course become cheaper because they are smaller, but if companies want to build chips of the same die area then it would cost them more on newer process nodes. Therefore companies seem to be choosing to cut the die size/number of transistors in order to make the overall chip less expensive (or rather make it cost around the same amount as it did before).