LONDON – Foundry Taiwan Semiconductor Manufacuturing Co. said this week its has prepared reference design flows in support of 20-nm planar CMOS and chip-on-wafer-on-substrate (CoWoS) assembly within its Open Innovation Platform (OIP). Both technologies are open for design starts.
[Get a 10% discount on ARM TechCon 2012 conference passes by using promo code EDIT. Click here to learn about the show and register.]
The 20-nm reference flow culminates in double-patterning-aware place and route to allow manufacturing using deep ultraviolet immersion lithography. Leading EDA vendors' tools have been qualified to work with the 20-nm process on timing, physical verification and design-for-manufacturing. In addition, TSMC and its ecosystem partners are developing 20-nm IP for double-patterning compliance to accelerate 20-nm adoption.
TSMC did not list specific vendors' 20-nm compatible tools, but the reference design flow provides a direct link with simulators for voltage-dependent design rules. An RF reference design kit provides high-frequency design guidelines, including 60-GHz RF model support, electromagnetic characterization and integrated passive device support.
The 20-nm high-k metal gate CMOS planar process is expected to begin volume production in 2013.
The CoWoS reference design flow enables multiple die to be integrated on an interposer. It is a heterogeneous approach in the sense that the interposer, or substrate, usually made of silicon, is unlikely to be in a leading-edge process. Similarly, multiple die on the interposer can be made using different processes.
The flow has now been validated with silicon runs and could be adopted to support high-bandwidth and lower power connections between chips, TSMC said. It includes: management of placement and routing of bumps, pads, interconnections and C4 bumps; an innovative combo-bump structure; accurate extraction and signal integrity analysis of high-speed interconnects between dies; thermal analysis from chip to package to system; and an integrated testing methodology for die-level and stacking-level tests.
Well the 22 nm intel fin has been reverse- engineered and publicized, not sure if that will be copied. Indeed, 14 nm could be more interesting battleground. Still this has been worked on for some years already, not much to fumble with, maybe channel materials.