Might dedicated foundries for TSV eliminate the need for Smaller devices?
Might the availability of oursourced TSV foundries (already serving the market in 2009) lower the barrier to circuit functionality improvements, such that continued feature shrinks become unnecessary for performance gains?
Outside foundry TSVs are large and require significnat keepout areas. This pretty much limits their use to I/O level 3D. I/O level 3D offers some improvement over wirebonding, POP, and other assorted I/O level package 3D, but this will not replace the need to shrink 2D circuitry. An I/O level 3D interconnect is ~40 gate delays using the power of 200 gates. Only the smaller FEOL or midline TSVs offer a real alternative to process shrinks. I think fine scale TSVs will give us a two or three generation boost, but shrinking will continue. Shrinking geometry will advance more generations but at a slower and slower rate with smaller and smaller incremental steps. The combination of 3D and shrinking will give us perhaps another 10-15 years of Moores Law like improvements.
While it's clear that a foundry can do I/O level (a.k.a. "global interconnect"), couldn't a line specialize in FEOL TSV (a.k.a. "local interconnect") and offer such as a foundry service? Ion implant, specialized depositions, and copper metalization (SMIC handling dual-damascene copper for TI) are all offered as foundry services. Why couldn't a fab offer a dense TSV process module (using poly or tungsten) as a foundry service?
I think that the logistics makes that model difficult, but not impossible. The out of fab for TSV and back in for the balance of the metal stack is a hard sell. Our experience with fabs willingness to do back and forth to an outside processor has been pretty low. The cost of the TSV equipment is pretty minimal also. For prototyping, the fabs we work with generally use the STI tool to cut the TSVs. Its a little slow, but for early work it is cost effective. The rest of process, at least for our tungsten TSVs is already in the fab, so no new capital is required.
Thanks for the perspective on prototyping TSV. It's good to remember that most of the unit process steps needed to form TSV already exist in many fabs.
Regarding outsourcing part of the flow, there is precedence for companies outsourcing unit process steps--such as ion implant, epi, and (recently) CMP--so it seems that the logistics are managable.
I think that the logistics makes that model difficult, but not impossible. The out of fab for TSV and back in for the balance of the metal stack is a hard sell. Our experience with fabs willingness to do back and forth to an outside processor has been pretty low. The cost of the TSV equipment is pretty minimal also. For prototyping, the fabs we work with generally use the STI tool to cut the TSVs. Its a little slow, but for early work it is cost effective. The rest of process, at least for our tungsten TSVs is already in the fab, so no new capital is required.
Thanks for the perspective on prototyping TSV. It's good to remember that most of the unit process steps needed to form TSV already exist in many fabs.
Regarding outsourcing part of the flow, there is precedence for companies outsourcing unit process steps--such as ion implant, epi, and (recently) CMP--so it seems that the logistics are managable.