One of the ongoing story threads we’ve covered in the foundry business is how Samsung beat TSMC to 14nm and the benefits it gained from doing so, in the form of high-profile contracts from companies like Qualcomm. E.S. Jung, executive vice president and head of Samsung’s foundry division, has now told the press that the firm wants a 25 percent share of contract manufacturing.
“We want to become a strong No. 2 player in the market,” Jung told Reuters, before adding that the company will try to boost its market by looking to multiple smaller companies as well as carving out space with cutting-edge firms. This is an ever-more-important component of any foundry’s growth. Fifteen years ago, the benefits of a node shrink were sufficient enough to attract even companies that don’t benefit as much from die shrinks. Analog components, Wi-Fi and cellular radios, and RF components never gain as much from a node switch as a conventional transistor. But over the long term, the costs and scale of manufacturing made it worthwhile for most companies to transition.
That’s no longer the case. Instead, we’ve seen the industry pivot to a complex discussion of ‘long’ versus ‘short’ nodes. 20nm, for example, was a short node, used by a limited number of mobile customers. 14nm, in contrast, is considered to be a long-lived node. As for 10nm and below, GlobalFoundries is skipping it entirely to head straight for 7nm (which is more like Intel’s 10nm), while TSMC and Samsung have different views of 10nm as well. Samsung sees 10nm as a long-lived node, while TSMC has suggested that it’s more of a speed bump on their own path to 7nm.
But there’s an interesting discrepancy between Samsung’s overall position in the market and its specific position in contract manufacturing. Overall, Samsung is expected to overtake Intel as the world’s largest chipmaker this year, thanks to ongoing demand for its memory (both NAND and DRAM) hardware. But as far as doing contract manufacturing, Samsung actually lags GlobalFoundries and TSMC.
TSMC commands ~50.6 percent of the contract manufacturing market, compared with 9.6 percent for GlobalFoundries and 8.1 percent for UMC. UMC isn’t one of the top tier foundries any longer; the vast majority of its business is tied to nodes between 3,500 nanometers (not a typo) and 90nm. It has one 12-inch fab that’s been outfitted for 14nm, Fab 12A, but it’s not a huge competitor compared with TSMC.
Samsung’s plan for this transition is to lean on revenue from its memory products while plowing an estimated $5.3 billion into developing new foundry technology that both the contract and non-contract parts of the company will share. Samsung is estimated to have earned about $5.3 billion in contract manufacturing sales in 2016 and wants to grow that by ~10% in 2017. Samsung has picked up Nvidia and Qualcomm as customers, but is believed to have lost the entirety of Apple’s new business over the past few years, as the folks at Cupertino transitioned from Samsung SoCs exclusively to TSMC.
According to Jung, Samsung’s fortunes may be critically linked to whether it can get EUV working. “You need a technology that can wow your clients. Without such advanced technology, it’ll be difficult to win back customers from your rivals,” Jung said — while declining to make any mention of who those customers might be.