Every Snapdragon 835 going into this year’s flagship Android phones is a Samsung 10nm chip, including the one in Samsung’s own Galaxy S8 and, more tellingly, the ones in everyone else’s. Winning the Snapdragon flagship outright, after the 14nm generation Samsung shared with the industry, is the clearest sign the leadership it grabbed two years ago was not a fluke. The fast-follower is still, for now, in front.
10nm is a refinement more than a revolution, a tightening of the FinFET process into production rather than a new transistor. The payoff is more cores and more graphics in the same thermal budget, which is what makes the Galaxy S8 and the other 835 phones feel like a real step up. There is no drama in the physics this time, which is itself the point: Samsung has turned the gamble that paid off at 14nm into a repeatable process, and repeatability is the thing foundry customers actually pay for.
Samsung has a reputation problem dating to 2015, and the Snapdragon win is partly about burying it. When Apple dual-sourced the A9 across Samsung 14nm and TSMC’s 16nm, the two chips turned out not to be identical, and the Samsung-built version developed a reputation, fair or not, for worse battery life. The internet called it chipgate, the benchmarks got argued over for months, and the lasting damage was not the battery numbers but the idea that a Samsung chip might be the lesser one. That perception is now the thing Samsung’s foundry has to outrun, and a clean Snapdragon 835 generation is how it starts.
This is also the year Samsung makes it official. It splits the foundry into its own division, separate from the System LSI group that designs Samsung’s own Exynos chips, and the reason is the discomfort that has haunted the business since the A4. A customer like Qualcomm does not love handing its crown-jewel design to the same company whose chip group competes with it, so Samsung builds a wall, at least on the org chart, between the people who build other companies’ chips and the people who design Samsung’s. It is a signal: we are a real foundry now, not a memory company doing fabrication on the side.
The pitch to customers is sharpening. TSMC still books better than half the foundry market to Samsung’s high single digits, and it holds Apple, building the A-series on its own 10nm while Samsung took Qualcomm. Samsung cannot out-scale TSMC, which is bigger and more focused and has the deeper roster, so it competes on being aggressive: first to new nodes and willing to deal in a way the market leader is not. That hunger is real value to a Qualcomm or an Nvidia tired of waiting in line at a sold-out TSMC. The risk is that hunger sometimes means shipping a node before it is fully baked, and a foundry that wins on aggression has to be careful aggression does not turn into yield problems the customer can feel.
The next couple of years are Samsung’s to lose. It has the Snapdragon flagship, a newly independent foundry telling the right story, and a process lead it has held since 14nm. Perso, I think the lead is more fragile than the win column suggests, because TSMC does not stay behind for long and has the resources to take any node seriously the moment it decides to. But right now, in early 2017, Samsung is building the most advanced mobile chips in the world and getting paid to do it for everyone, the rivals included. The company that was renting out spare fab time seven years ago is running the front of the industry. The trick from here is staying there.