Samsung's foundry is having a darn good 2015. The Samsung Galaxy S6's Exynos 7420 is the first chip built on the 14nm process, and early reviews suggest that the company's decision to wait for its own homegrown solution instead of Qualcomm's Snapdragon 810 paid off. With multiple companies, including Qualcomm and (it's rumored) Apple itself tapping Samsung for their 14nm needs, the manufacturing arm of the Korean giant is set to have a banner year. It's natural, therefore, to see a headline talking up an Nvidia-Samsung GPU collaboration and think "Hey, that could totally happen."
According to the Korea Times(Opens in a new window), Samsung has won a contract to manufacture GPUs for Nvidia. That publication quotes a source as saying: "The latest agreement between Samsung and Nvidia is another positive factor lifting Samsung's logic chip business unit. The timing looks good as increased foundry customers justify the Korean chip giant's moves to find new revenue sources." The source then goes on to claim that the deal is worth "a few million dollars."
It's possible that this is a translation error, or that the deal in question is merely for early pilot production. But we believe a Samsung-Nvidia deal is dubious for several reasons. In addition to the low dollar value specified in the Korea Times story -- "a few million" dollars would barely pay for initial design work, much less any major product ramp -- there are questions of timing and fundamental process suitability.Which 14nm node?
Samsung has two 14nm nodes: 14nm LPE and 14nm LPP. 14nm LPE stands for "Low Power Early," and promises improved power consumption and performance, but the cream of the crop will arrive later, with 14nm LPP (Laser-Produced Plasma) and a further 15% improvement. Critically, however, both of these are low-power nodes with a specific focus on SoCs and IoT (Internet of Things) types of devices. To date, no manufacturer has announced that they intend to build a high-power product on a Samsung process node, and Samsung has no experience in building that kind of hardware.
Now, one might ask -- what about GlobalFoundries? GF, after all, has at least some limited experience in building GPUs thanks to its work on Kaveri, Carrizo, and socketed Kabini. It also knows something about building chips that consume up to 140W thanks to its CPU work with AMD. While all these things are true, they don't change the fact that GloFo's 14nm process technology is a straight-up port of Samsung's work -- GF's own 14nm-XM tech is off the table and under the rug.
None of this means that Samsung or GlobalFoundries couldn't build a 14nm high-power process line, but that's the kind of specialized work that usually results in a great deal of money changing hands. Foundries don't agree to build out a specialized iteration of a technology if they don't have cash up front, and companies like Nvidia don't pay to build out that foundry line unless they're certain it'll be worth the cost.
It's possible that Nvidia is prototyping a low-power variant of a GPU architecture to see if it's worth moving its mobile GPUs to Samsung's 14nm technology, or that it wants to get a feel for what a 14nm GeForce on 14nm silicon would look like. It's also possible that the Korea Times goofed, and that this new collaboration between the companies is Tegra-specific. A next-generation Tegra on 14nm silicon -- a follow-up to the X1 -- would align perfectly with Samsung's existing specialization and wouldn't require such heavy lifting on the GPU front.
Early exploration isn't out of the question. TSMC's 16nm process isn't expected to make much of a dent in the semiconductor industry; most early movers have leapt for Samsung. The company's 16nm+ process, which is viewed as analogous to Intel and Samsung's 14nm, isn't actually expected to hit the market until Q3 2015.
