Four Strategies for Pursuing Optical Interconnects
This research note is based on CIR’s new report, “Revenue Opportunities for Optical Interconnects: Market and Technology Forecast – 2013 to 2020 Vol II: On-Chip and Chip-to-Chip”
Over the next five years chip companies will experience an “interconnect bottleneck” that will cause them to create alternatives to current metal/electrical interconnects. The most likely technology direction will be optical, creating new opportunities for several different kinds of companies.
For chipmakers, optical interconnect represents enabling technology for next-generation chips. Hence, Intel and IBM are already actively involved in optical interconnect development. For optical component companies, optical interconnect at the chip level promises an addressable market of billions of units.
CIR research indicates four strategies for firms seeking to take advantage of this new strategic direction in the semiconductor industry.
Strategy #1: Light Engines
Essentially miniaturized optical assemblies, light engines are already available from a handful of vendors including Avago and Reflex Photonics. They use conventional technology, although at least one product contains silicon waveguides.
Light engines are more board-to-board and module-to-module than chip-level. Further miniaturization is possible; perhaps using waveguides and more optical integration. But light engines have limits as far as semiconductor industry applications are concerned because of their size.
Strategy #2: Compound Semiconductors
This overlaps with Strategy #1. InP—and to a lesser extent—GaAs have good optical properties, but are expensive. These materials are standard for fabricating active optical components and waveguides; the optical components industry is based on them.
This industry will benefit from the need to optically interconnect at the chip level. Opportunities exist for smaller, less expensive VCSELs and other innovative hybrid integration products. CIR sees this strategy as inherently more scalable than the light-engines approach.
Strategy #3: Silicon Photonics
Silicon detectors, modulators and waveguides have been used in optical communications for many years. However, several firms—notably IBM—have started to develop silicon photonics products specifically for optical interconnects. Intel has developed a “silicon laser”—actually a Si/InP hybrid. Micron Technology is working on the integration of silicon photonics into bulk CMOS fabrication.
Silicon photonics promises to bring the existing deep understanding of silicon device fabrication to photonics. For chip-level interconnect the vision is of silicon chips that integrate electronic processor with optical interconnect, all in silicon. Indeed, interconnect could be the “killer app” for silicon photonics. But for now, the technical challenges are great and significant revenue generation from silicon-based photonic integration may be well in the future.
Strategy #4: Polymers
Polymer waveguides for optical interconnect were suggested back in the 1990s, but didn’t get much traction. They continue to attract interest from a few large specialty chemical firms including Dow Chemical (Rohm & Haas), DuPont, and Dow Corning, and some start-ups. Polymer detectors and modulators are in limited production by, for example, GigOptix. Lasers using organic materials are possible, but are just a research topic for now.
The selling proposition for polymer optics is low cost. Chip-level optical interconnect requires this. Whether performance and durability of polymers is good enough for this application remains to be seen.
There is no certainty that an “interconnect bottleneck” will emerge in the semiconductor industry. Such a bottleneck may be what independent firms supplying optical interconnect solutions want, but the semiconductor industry would rather avoid it.
If the “bottleneck” appears, will it be a crisis or a slowly emerging issue? If the latter, then some companies that see opportunity in chip-level interconnection may overshoot the market with their products’ functionality. In such cases, it is possible that firms could re-target their products towards board-to-board interconnection, which is an already extant market.
There is also technological risk. Many of the products profiled above are difficult to fabricate. It isn’t just a matter of “getting the technology right.” Photonic devices used in chip-level optical interconnection must also be small and inexpensive. These are immensely difficult requirements to fulfill.