A few weeks back, we covered foundry plans over the next few years for TSMC, Samsung, GlobalFoundries, and Intel. Samsung has since issued a major update on its own roadmap plans and projects. The company is planning to launch a variety of new process capabilities, including new technology nodes, the introduction of Extreme Ultraviolet Lithography (EUV), and its own fully depleted SOI (FDSOI) technology. To date, GlobalFoundries has been the foundry playing up its FDSOI credentials, so it's interesting to see Samsung breaking into this space.
Let's start with a general look at Samsung's roadmap before we dive into the specifics of its technology ramps or how we expect these initiatives to play in various markets. Unfortunately, for reasons known only to itself, Samsung distributed some perfectly nice slides, then refused to allow us to publish most of them, even though we're allowed to talk about the data on them and, in a few cases, recreate them in our own images. Since I have the approximate image design aptitude of Koko the gorilla, I apologize for what you're about to endure.
In its presentation, Samsung noted that Risk Production is a target for an entire year, and that the left and right edges of the roadmap blocks (on the slides I'm not allowed to show you) don't actually mean anything. That's significant, because in some cases, the left edge of a block is read as the beginning of a process while the right-hand edge means that product or process is ready for completion. Different foundries mean different things, sometimes, when they use the same phrases. The time span between risk production, volume production, and actual product launch will vary depending on both the product in question and the vagaries of the ramp-up process.
When Intel refers to entering volume production on a given node, it usually launches parts within 1-2 months. When TSMC, Samsung, or GlobalFoundries announce volume production, the chips that roll off the line still have to go into various devices, which then go through a manufacturer-specific evaluation and test process. That's why Samsung could claim to enter volume production of 10nm SoCs back on October 17, 2016, but the first device to use those SoCs was the Samsung Galaxy S8, which launched on April 21. The six-month gap between those two events is typical, and longer gaps are not unknown.
From this point forward, I'll be describing what the company told us in general terms, rather than backing it up with visuals. The 14nm LPU process node referred to above is Samsung's third generation of 14nm technology, and the company told us it expects the LPU node to offer up to 15 percent improved performance compared with earlier 14nm process lines. This kind of incremental scaling is a relatively recent innovation in the foundry business. In the past, companies like Samsung, TSMC, and GlobalFoundries delivered "all" the benefits of a given node up front, with smaller improvements over the next few years. Now, the gap between early and later production on the same node is somewhat larger.
As for EUV, we've talked before about the difference between EUV and existing Argon-Fluoride lithography (ArF). Samsung's 8nm node will extend non-EUV production one last time, with strictly incremental performance improvements above 10nm. Beyond that point, Samsung believes the mask counts and costs will simply be too high to justify non-EUV technology, while the overall performance of any future non-EUV node would be worse than the company's planned 8nm. Obviously if EUV fails to launch on time, foundries will try to come up with ways to extend existing technology further. But we've already been stretching non-EUV lithography for at least a decade, while adopting multi-patterning and immersion lithography to compensate for EUV's continued development difficulties.
President of Samsung Electronics’ semiconductor division, Dr. Kinam Kim, presented at the Samsung Foundry Forum 2017 in Santa Clara, Calif., on May 24, 2017. Click to enlarge.
The point of offering new FD-SOI progression paths is to give customers who may not have the funds to invest in cutting-edge process nodes a way to continue to extend and improve their existing product lines. If you are a current 28nm bulk silicon customer who doesn't want to do a full chip redesign, 28nm FDS (Full Depleted SOI) offers performance and yield advantages. FinFET will be the performance option of choice for customers who can justify a full architecture change, while 18FDS will give 28nm bulk and 28FDS customers significant performance improvements.
GlobalFoundries has also talked up its investments in FD-SOI, and it's interesting to see Samsung picking up the mantle on this as well. The company expects to bring eMRAM and RF silicon to market on FD-SOI and cites improved costs and performance even for RF parts (RF manufacturing is quite different from traditional silicon manufacturing and the gains from new process nodes are traditionally much smaller).
Click to enlarge
Finally, there's Samsung's advances in packaging manufacturing. The company wants to advance the use of new techniques to minimize thickness when using fan-out, while eliminating the interposer layer altogether for HBM and HBM-style memory technologies. I-Cube (Interposer Cube) is Samsung's own name for its 2.5D interposer technology. The goal is to remove the need for a silicon interposer altogether and boost bandwidth while cutting costs. We know AMD is also sourcing HBM2 from Samsung, so they're clearly ramping these technologies towards mass production.
Overall, Samsung's roadmap is extremely ambitious and reflects a company that wants to challenge TSMC in multiple product markets. Samsung is seeking to position itself to guard against potential unknowns in the market (like the ramp time for EUV), while simultaneously pushing into spaces like FD-SOI, where previously GlobalFoundries seemed to be the only company carrying that particular technology forward.
When Samsung took the lead on 14nm, companies like TSMC were quick to argue that this was a fluke, and that they would resume their historic leadership position on new nodes shortly thereafter. It's now clear that Samsung has no intention of allowing its 14nm success to be a one-time-only proposition. How its foundry competitors will answer this challenge remains to be seen. But the semiconductor industry has been the better for their recent challenges to TSMC, and that's a trend we expect to continue.
