Samsung, a foundry supplier of bulk CMOS circuits to ST, has agreed to license a 28nm silicon-on-insulator process and act as a primary and second-source of the process. The license allows Samsung to implement the 28nm fully-depleted silicon-on-insulator (FDSOI) manufacturing process that has been developed by STMicroelectronics NV.
The signing gives the process additional credibility as ST itself has only relatively limited production capacity to run prototype-to-medium volumes of FDSOI at its own pilot wafer fab and R&D center in Crolles, France. As well as offering 28nm FDSOI as a foundry option and as a high-volume supplier to ST, the licensing deal allows Samsung to make its own circuits using the process, the companies said.
Samsung's support is all the more important because a similar FDSOI licensing agreement between ST and Globalfoundries, announced in January 2013, has yet to produce any obvious results. In addition, the process sits outside the mainstream bulk-CMOS to FinFET transition that has been pioneered by Intel and embraced by leading foundry Taiwan Semiconductor Manufacturing Co. Ltd.
Only three weeks ago, Samsung and Globalfoundries announced they would work together to offer a common 14nm FinFET manufacturing process for foundry customers.
Jean-Marc Chery, chief operating officer of ST, told Electronics 360 that the agreement with Samsung is an extension of existing manufacturing agreements between the companies. "This will make FDSOI a competitive innovative standard." Chery said that ST has co-operated with Samsung on 32nm/28nm bulk CMOS since 2009 and Samsung is the largest supplier of CMOS foundry services to ST.
The 28nm node will be a long-lasting one because of additional costs associated with going to 20nm bulk CMOS or to FinFET production which, in the absence of extreme ultraviolet lithography, require more expensive double patterning of wafers, Chery said. The 28nm FDSOI process offers the option to migrate from bulk while improving performance or power consumption.
The claim is that 28nm FDSOI is faster, cooler than 28nm bulk CMOS and simpler than a move to FinFET production. While few have disputed these claims, Intel and TSMC have expressed concern about the higher cost of the specialized SOI wafers and the limited number of suppliers.
Kelvin Low, senior director of marketing for Samsung's foundry operation, said: "This signifies our commitment to serve players in the foundry space. FDSOI is perfectly positioned to extend the 28nm node without going to double patterning. HKMG architectures are very similar, allowing us to leverage the learning we have already made."
PDK, IP support
Samsung and ST will support common 28nm FD-SOI library and IP and Samsung has also licensed the physical design kit, foundation libraries, advanced IP and design flow. The PDKs are available immediately. "We expect to qualify the process early in 2015 and then go to volume production."
ST and Samsung have also broadened the scope of applications for 28nm FDSOI compared with some previous statements. Networking infrastructure, consumer multimedia and the Internet of Things have been mentioned previously. Now smartphones have been added and automotive.
Chery said there are now 17 designs in progress for FDSOI up from 15 design wins in January. "14 are ASIC and 3 are ASSPs," said Chery.
Although benchmarks are provided of ARM Cortex-A9 processor cores and Imagination PowerVR graphics processing unit implemented in 28nm FDSOI the absence of flagship IP providers and EDA companies as supporters of FDSOI is notable. On both the IP and EDA fronts Chery said: "We expect an announcement very soon."
Len Jelinek, senior director and chief analyst at IHS Electronics & Media, the publisher of Electronics 360, did not dispute the claimed cost advantages of 28nm FDSOI over 20nm bulk CMOS and other technologies.
"FDSOI offers a significant cost/performance advantage, especially if you do not need to go to double patterning. However, the technology is still gate-first, so it is not industry standard and will require significant engineering," Jelinek said. "But then again if you are going into a cost-sensitive market using 40nm you could migrate to FDSOI and get performance improvements with the cost of 20nm bulk CMOS or 14nm FinFET."
Jelinek, concluded: "This might be a very opportunistic solution just at the right time or at least be a bridge until lithography has a cost-effective solution."
Related links and articles: