Samsung Unveils Scaling, Packaging Roadmaps

Foundry unit rolls out ambitious plan down to 4nm, along with 18nm FD-SOI and advanced packaging developments.


Samsung Foundry unveiled an aggressive roadmap that scales down to 4nm, and which includes a fan-out wafer-level packaging technology that bridges chips in the redistribution layer, 18nm FD-SOI, and a new organizational structure that allows the unit much greater autonomy as a commercial enterprise.

The moves put Samsung Foundry in direct competition with Intel, GlobalFoundries and TSMC, as well as vendors, across a wide swath of markets ranging from mobile devices to IoT, and RF. Samsung also announced its plan to begin risk production for 8nm LPP this year, and to begin using EUV in its 7nm LPP process next year. EUV is expected to reduce the number of photomasks required at that node by about 20.

Exactly what these numbers mean, and how they compare to other foundries, isn’t entirely clear, and there is discussion throughout semiconductor manufacturing sector that one foundry’s numbers don’t match another’s. But what is clear is that Samsung is attempting to pick up business at every node and half-node, including 8nm, 7nm, 6nm, 5nm, and 4nm, and it plans to introduce an 18nm version of its FD-SOI technology in 2019.

“This is now an independent foundry,” said Kelvin Low, senior director of foundry marketing at Samsung Foundry. “We will still leverage Samsung R&D and memory and logic. And we will utilize the company’s advanced packaging and manufacturing capacity. But we are now an independent business organization.”

This result, in effect, is being able to utilize the deep-pocket R&D of an IDM with a thriving end-market business in everything from televisions to smart phones, while also being able to utilize capacity of a newly constructed 300mm fab equipped with the latest technology, including EUV systems. Low said that with EUV the company is turning out 1,200 wafers per day, and he expects that number to improve.

Samsung has been in production on 10nm since late last year. The company issued a press release in March to that effect, saying it has shipped more than 70,000 wafers of its first-generation Low Power Early (LPE), although it didn’t discuss any additional details.

Joanne Itow, director of manufacturing at Semico Research, noted that all of the major foundries are scrambling to find out which process technologies will work best for which applications. Because many of the hot growth markets—automotive, industrial and regular IoT, augmented/virtual reality and medical—require new technology, it’s uncertain which process will win.

“Everyone is trying to figure out which process is best for which product, so foundries are opening up all of these processes,” Itow said. “This is coming from a variety of inquiries. Not every one can be a winner. Eventually, this will filter down to a few technologies. But customers do want different options this year, and we’re seeing the same thing happening with TSMC. They need to roll out something new. This is becoming a pervasive electronics applications market, and there are so many different types of products that we are seeing different processes to accommodate them.”

For Samsung in particular, this is a signal that the company is now courting a broader base of customers, she said. “In the past Samsung was very selective because they wanted to make sure they were successful with their product launches. Now, the next step is to get a broader base and to expand its foundry revenues.”

Samsung’s announcement also included developments at 8nm and 6nm. Samsung did not elaborate. But Sam Wang, a research vice president at Gartner, said the 8nm is a competitive move. “From a time-to-market point of view, it is inevitable that Samsung must offer a relaxed 7nm technology, in response to TSMC’s aggressive 7nm DUV schedule before EUV becomes ready,” he said. “Customers could not fully rely on Samsung’s 7nm EUV-only schedule, which has uncertainty because of the the exact progress of ASML (on EUV). In a way, Samsung’s 8LPP node is a relaxed 7nm node, which should be equivalent to TSMC’s N7, and Samsung’s 7LPP should be equivalent to TSMC’s N7+.”

Samsung also announced plans to introduce its first gate-all-around FET at 4nm in 2020 using EUV. This is the first time a foundry has publicly discussed a timetable for delivering GAA FETs, but roadmaps exist within a number of companies for several more nodes. That includes EUV lithography, GAA FETs using vertical and horizontal nanowires, and nanosheet FETs.

Kinam Kim, president of Samsung’s Semiconductor Business, said during a presentation at a recent event sponsored by Imec, a Belgium R&D organization, that the company sees a path to logic transistor scaling down to 1.5nm.

Then, using a 2D material called molybdenum disulfide (MoS2), Samsung believes it could scale logic technology even further. Samsung and others are exploring so-called MoS2 FETs. “We believe around 1nm is possible,” Kim said. Still in the R&D stage, MoS2 is a family of transition metal dichalcogenide (TMD) materials. The TMDs have remarkable electronic, optical and mechanical properties.

Fig. 1: Cross-section simulation of (a) finFET, (b) nanowire, and (c) nanosheet. Source: IBM.

In addition, Samsung announced its second-generation FD-SOI at 18nm in 2019 using immersion lithography. The company has been adding RF and other IP capabilities to its 28nm FD-SOI technology. Going forward, it plans to add embedded MRAM to FD-SOI, and eventually into finFET processes, as well. The company also is looking at FD-SOI for the automotive industry because it has a superior soft-error rate over bulk CMOS, said Low.

How this fares in the market is not clear. “On FD-SOI, we now have Samsung 28, 18nm, and GlobalFoundries’ 28, 22, and 12nm,” said Gartner’s Wang. “Since Samsung’s 18nm is independently developed by itself, it remains to be seen how GlobalFoundries will response to this 18nm offering.”

Back in CMOS, it appears that 10nm will be a long-lived node for Samsung. “We expect 10nm will be a very usable, long-life node,” Low said. “But some customers require a new node every year. So 7nm will be finFET on CMOS. 6nm will be smart scaling for area and power. 5nm will be finFET on CMOS. After that, we will show a post-finFET device, using a gate-all-around multi-bridge channel finFET.”

Perhaps the biggest surprise is a shift in the company’s advanced packaging. Samsung had hinted last year that it was looking for an alternative to silicon interposers because they are too expensive. The company is relying on a redistribution-layer (RDL) interposer to bridge logic to high-bandwidth memory, sidestepping commercial interposer technology for its 2.5D technology, and adding the same technology into its fan-outs.

Fig. 2: RDL interposer. Source: Samsung

Intel has introduced its own low-cost Embedded Multi-die Interconnect Bridge (EMIB), which is a silicon bridge that runs through the package substrate.

—Mark LaPedus contributed to this report.

Related Stories
Extending EUV Beyond 3nm
Now that EUV is finally shipping, companies are working on extending it much further using anamorphic lenses and high numerical aperture technology.
Uncertainty Grows For 5nm, 3nm
Nanosheets and nanowire FETs under development, but costs are skyrocketing. New packaging options could provide an alternative.
What Will Transistors Look Like At 5nm
As finFETs run out of steam after 7nm, what comes next? The debate is just beginning.
5nm Fab Challenges
New transistor types, plus issues with masks, patterning, materials, process control and interconnects, add up to a very tough transition.


David Ral says:

Ok this is getting ridiculous. Sansung’s marketing team is on overdrive. Their so-called “10nm” is currently worse in terms of drive current and less dense than Intel’s “14nm” based on their PDK. I can’t wait to see what they think a “4nm” node geometry should be. Someone needs to put an end to those names.

Dolan1998 says:

Drive current is not the only aspect affecting performance, there is many other aspects and overal performance should be better (depending on design).

And most importantly, it is definetly better than Intel’s so-called “10nm”, “10nm+” and probably “10nm++” too.

Also we should finally clarify difference between pitches of benchmarking cells and density achieved in real world chips. In one case, Intel surely will have advantage, other will be simmilar to what we see today (Skylake ~15 mtr/mm2, ZEN 25 mtr/mm2).

Jack says:

So 4nm = 40 nm? Nobody can stop marketing from going forward.

Its like the story with the LCD displays. They didnt really got faster, but year after years marketing coined new ways to “measure” it, so now we have “1ms” displays that still need 30+ms for a full transition.


Mehmed says:

14nm is 42-70nm class
10nm is 34-54nm class

And I guess Intel’s 14nm was better than Samsung’s 10nm.

memister says:

Latest rumor is that Samsung is skipping nodes to go direct to 6nm, after losing 7nm Qualcomm order to TSMC. But I think for it to be safe, it needs to go direct to 5nm (TSMC might pull that in) or maybe even 4nm.

Leave a Reply

(Note: This name will be displayed publicly)