Strategies for Wave-guides on CMOS

1
Strategies for Wave-guides on CMOS

• OIC Strategies for Global Interconnect Upon CMOS
• Dr Terry V Clapp, Dr Jon V DeGroot jnr Dr Ann
  Norris
• February 2004

2
Intent

1. I will show what is driving the requirement to
   consider optical communications at the chip
   level.
2. I will show what challenges the CMOS platform
   offers.
3. I will offer some materials, process and
   waveguide propositions for these applications.

3
The Interconnect Environment
Interconnect level number of layers
Passivation
Etch stop and dielectric cap
Global Up to 5
VIA
Copper damascene Nucleation and barrier
Intermediate Up to 4
Tungsten plugs
Local 2
Pre-metal dielectric
4
Design Window

• The design window appears to favour dielectrics
  as the preferred optical material.
• Spin-on glassy dielectrics would give a low-k and
  low temperature option.
• Polymers either spun-on /or printed might be a
  preferable process option in view of topology and
  the need for electrical connections.
• Polysiloxanes and perfluoropolymers both have a
  demonstrated process compatibility and acceptance
  for deployment in CMOS foundries.

5
PWG/OIC Efforts

• Dow Corning working in collaboration with Gemfire
  and others to develop low cost OIC fabrication
  process as well as prototypes via
  photo-patterning.

Early attempt at single-mode 6 x 5 micron nominal
dimensions
Multi-mode 100 x 100 micron
6
Absorption lt0.02dB/cm at 850 nm
7
Spot Size / Confinement
8
Relative versus Absolute size

• Say 75um pitch for the global interconnect,
• at high bandwidth cross-talk is likely to be an
  issue,
• very high speed modulation will become critical
  in terms of routing and power.
• Optical _at_lt20um pitch for wave-guides, if,
• you work in the visible
• you can provide sources and receivers in fact CW
  lasers off-chip with on-chip modulators seems a
  better architecture.

• For square SM buried rib wave-guides,
• the core-cladding index difference determines
  mode confinement up to the diffraction limit,
• the higher the difference,
• the more difficult it is to make the guide low
  loss,
• the tighter the bends you can have,
• at shorter wavelengths,
• the core diameter is smaller,
• the cladding thickness can be reduced.
• Are mirrors vs bends the answer?

9
Conclusions

• It seems clear that the optical penetration in
  the communications infrastructure will penetrate
  right down to the chip level.
• The demands of that environment seem to favour
  the adoption of technology in the visible
  wavebands.
• Materials are available but much research needs
  to be done to establish the boundary design rules
  demonstrate the technology.

10
Acknowledgement
The contribution of my colleagues at Dow Corning
is gratefully acknowledged