Efforts increasing at older nodes and with advanced packaging as end markets fragment.
By Ed Sperling & Mark LaPedus
Major foundries are stepping up their offerings across a wide swath of technology nodes, specialty processes and advanced packaging—a recognition that end markets are fragmenting and that the path forward includes a mix of new and established processes.
As the smart phone market flattens, there is no single “next big thing” to drive volume at the most advanced nodes. Consequently, foundries have begun broadening their focus by tapping into more markets with minimal investment. While the race at the leading edge continues unabated, there is growing competition at established nodes in preparation for the IoT, as well as a big push to drive down the cost of multi-chip packaging in 2.5D and fan-outs.
TSMC, the largest commercial foundry, is farthest along in comprehensive multi-market push, but GlobalFoundries and Samsung are quickly ramping up their efforts in this part of the market. So are UMC, SMIC and big outsourced semiconductor assembly and test (OSAT) providers such as ASE and Amkor.
“There is a recognition that continued growth of smart phones will not be enough of a driver for leading-edge fabs,” said Joanne Itow, managing director of manufacturing at Semico Research. “At its recent technology symposium, TSMC emphasized areas that were not at the leading edge. The demand for mature technology nodes is still strong, and foundries can make a lot of money on fully depreciated fabs and mature technologies.”
Samsung this week unveiled plans to commercialize its 8-inch fab offerings, including internally developed IP for embedded flash, image sensors, and high-voltage devices, along with design enablement services for the consumer, mobile, networking, server and automotive markets. That expertise also will include advanced packaging.
“Samsung has been packaging for a number of years, but most of that has been in-house,” said Kelvin Low, senior director of foundry marketing at Samsung. “We will make different shades of that packaging available to our packaging customers. This is a big change. We also have showcased a silicon interposer in the past, but the cost is still too high. We are creating a road map to get the cost down for different versions of interposers so they can be used in cost-sensitive markets.”
Both TSMC and GlobalFoundries have been offering design services for a number of years. TSMC has a stake in Global Unichip. GlobalFoundries has developed an entire ecosystem, including a partnership with Invecas. And SMIC leverages Brite Semiconductor. SMIC also teamed up with Jiangsu Changjiang Electronics Co. (JCET) in 2014 to buy STATS ChipPAC, a Singapore-based packaging house.
Battle at the leading edge
While all of these foundries continue to expand their offerings at every node, the battle at the leading edge of Moore’s Law continues unabated—not the least of which is a battle of words. It’s difficult to compare one foundry’s 7nm process with another because the back-end-of-line process may not be the same as the transistor line widths or pitch.
So far, there isn’t even certainty about what type of transistor will be used at 7nm.
“There still is discussion around that,” said Samsung’s Low, adding that the goal is to stretch finFETs to 7nm. “But 7nm with EUV is a necessity. With EUV there are less masks, which equals less defects.”
Sanjay Jha, CEO of GlobalFoundries, said in a speech at CDNLive earlier this month that the amount of data moving to data centers is doubling every few years. He said that will drive advanced technologies such as silicon photonics, which he estimated will be a $33 billion to $35 billion market, as well as 7nm chips because the amount of processing will continue to rise. But he noted that focusing just on the leading edge would be a mistake.
“Growth will be dramatically higher in non-leading edge technologies,” Jha said. “By the time we get to 5nm, we will have to change to a vertical FET and we will have to have EUV working. 5nm will be a very expensive node. The rate of innovation will be coming down because the ROI is not there. As the industry consolidates and the return on innovation goes down, we then have to do more innovation. I believe we will go to more heterogeneous architectures.”
Dave Hemker, CTO at Lam Research, agrees. He said that process technology and manufacturing equipment will continue to be developed at the leading edge of the Moore’s Law road map for years to come. But that also does not mean every chip will be developed using the most advanced technology.
“It doesn’t make sense to make a 14nm analog chip,” Hemker said. “It’s much easier to package together with something else.”
This helps explain why there is so much effort being put into established process nodes and specialty fabs. Samsung said this week it is entering the specialty foundry business, a move that puts the company in an increasingly crowded field. GlobalFoundries, SMIC, TowerJazz, TSMC, UMC and others compete in the large but fragmented specialty foundry field.
The specialty foundry market includes a number of technologies, such as analog, mixed-signal and RF. There are several sub-segments within these categories. And the processes are typically more mature and fabricated in 200mm fabs.
Surprisingly, though, 200mm fab utilization rates are tight right now. “At key foundries, 300mm fab utilization rates continue to lag behind 200mm fab utilization,” said Samuel Wang, an analyst with Gartner. “200mm fabs at some foundries are fully utilized.”
Demand is strong for various 200mm technologies, especially 180nm processes. “In 180nm, the volumes will continue to go higher,” Wang said.
One of the hotter 180nm markets is bipolar-CMOS-DMOS (BCD). BCD combines the advantage of several processes, including bipolar for analog, CMOS for digital, and DMOS for power and high voltage.
In fact, seeking to meet demand, TSMC is readying its third-generation BCD processes based on 180nm technologies, according to Mark Liu, president and co-CEO of TSMC.
There are other strong markets in the specialty foundry arena. Seeking to get a piece of the action, Samsung plans to provide the following technologies in the arena—embedded flash; power management ICs; CMOS image sensors; and others. It will provide these processes within its more mature 200mm fabs.
For some time, Samsung has developed these technologies for its own systems, namely smartphones. “We want to make them available beyond just for internal consumption,” Samsung’s Low said.
Samsung also is providing various chip-packaging technologies for foundry customers. The company is no stranger to chip-packaging. For years, it has been developing various 2D-based package types for its own chips. It also has expertise in through-silicon vias (TSVs). In fact, it sells a DRAM module based on TSVs. More recently, Samsung began mass producing the industry’s first DRAM package based on the second-generation High Bandwidth Memory (HBM2) interface.
Now, Samsung wants to enable this technology for foundry customers, a move that follows similar efforts by TSMC and Intel. For some time, TSMC has developed 2.5D, chip-scale packages (CSPs) and other technologies. The company is also pouring billions of dollars into the fan-out packaging arena.
Intel is also offering its advanced packaging technologies for customers. Not all foundries are moving into the packaging business, however.
All told, Intel, Samsung and TSMC are not providing commodity packages for customers, but rather more advanced packaging types. Still, the move puts vendors on a collision course against traditional OSAT providers.
“You look at the vertical conglomerates. When they do packaging, do they have an intrinsic advantage? In some respects, the answer is yes. They will have a better sense of design for manufacturing from the IC chip design to the foundry process and all the way to packaging. They will have an integrated P&L. That means they understand how to allocate R&D dollars and cost,” said Tien Wu, chief operating officer at Taiwan’s Advanced Semiconductor Engineering (ASE), the world’s largest OSAT.
“The foundries also have an intrinsic disadvantage. For example, they are not a dedicated OSAT. They are not a dedicated packaging house. If you go with a company to do the foundry, and maybe the chip design and packaging, customers may lose a portion of their supply chain flexibility. That involves pricing, capacity or the need to have multiple suppliers,” Wu said.