Shifting to photonic clocking

Siepmann: One-size photonic clocking will not fit all. Alternative systems must be compared for specs, efficiency, scalability and cost.

In the microprocessor industry, copper interconnects are finding limitations as data-transmission bandwidth and processor speed continue to rise. With higher data rates and clock speeds, and the current trend toward distributed multicore architectures, very high-speed interconnects within and between computing elements will be necessary. The recognized solution is to change from electronic to photonic interconnects for both data transport and clocking.

Photonic clocking not only solves the limitations of electronic clocking, but also reduces jitter, skew, delay, crosstalk and power consumption while maintaining clock signal integrity for longer distances. Furthermore, it can be implemented in the next high-volume manufacturing cycle. Photonic clocking can meet the demands of both single-core processors at low gigahertz rates and multicore processors that run at 100GHz and higher.

Though the transition to optical clocking is a "when, not if" situation, this price-sensitive market will demand a photonic solution that is also low in cost. So the question is how to make this transition and to what architecture.

One-size photonic clocking will not fit all. Alternative systems must be compared for specs, efficiency (power or heat issues), scalability and, most important, cost.

For a smooth transition, make the most of what you have. Most silicon-based microprocessor chip companies aren't going to change to gaas so that you can build your laser on-chip. Thus, you have to concentrate on efficient coupling to the chip or hybrid mounting onto it. If you are already working in GaAs or inp, then life just became a lot easier.

Since microprocessor technology will continue to increase clocking rates, photonic clocking technology must scale with relative ease to at least one order of magnitude. Already, 2GHz photonic clock systems have been demonstrated that can scale up to 20GHz. Another laser design is scalable from 10GHz to more than 100GHz. And with the increasing number of cores in microprocessors, the distribution system should also be readily scalable.

Embrace compromise. I would love to develop a photonic clocking system with a 100GHz bandwidth, but it doesn't make sense financially. If there are unacceptable cost penalties in changing to a photonic solution, the industry will push the current VCO/copper technology until those cost penalties go away. The good news is that acceptable photonic clocking solutions should be cost-viable in a 2008-2012 high-volume manufacturing cycle.

Lastly, don't make it expensive. To be implemented, any new technology needs to be cheaper or at least in cost parity with the incumbent technology. Microprocessors are a commodity and decisions will be based on the bottom line rather than what will make a product better.

- James Siepmann
LightTime LLC