Today, the only way to keep signals in the optical domain is to run them through a long length of optical fiber. But an on-chip optical network solution makes it possible to vary the delay, depending on network congestion, which today can only be achieved by converting optical signals to electronic, storing them in memory, then retranslating them to optical signals as needed.
The main problem encountered by other groups, from Bell Labs to Hewlett-Packard to Cornell University to the University of Rochester (N.Y.) to the University of California at Santa Barbara, is that nanosecond-scale delays are only possible by narrowing the signal's bandwidth to the point of making it unusable. On the other hand, if the bandwidth is kept wide, the delay becomes too small to be usablea classic catch-22.
"There is always a trade-off between delay and bandwidthwe can increase the delay by decreasing the bandwidth and vice versaso the idea behind the design of this particular delay line was to have a relatively big delay in a small footprint, but with a big enough bandwidth to have meaningful bit rates of 5 to 10 Gbits/second," said Xia.
IBM engineers achieved a tenfold increase in bits over previous demonstration chips by other research groups. "In our case we have 10 bits in 0.03 mm2, which is a new world's record. The previous record holder used atomic gases to achieve longer delays, but with much lower bandwidth, resulting in only a few bitsso our record is about 10x greater than this previous one," said Vlasov.
The ultimate goal of the project, IBM said, is to go from 10 bits today to hundreds of thousands of bits, which would eliminate the need to ever convert from optical to electronic signals even if the data transmission went through multiple switches and traveled through repeaters for thousands of miles. But IBM estimates that the project will take from five to 10 years to achieve that goal, because it will need not only delay lines but a whole array of other photonic devices that perform functions now done electronically.
"We plan on building a silicon photonics toolboxeverything engineers will need for future on-chip optical interconnects," said Vlasov. "For example, besides waveguides we are also fabricating modulators, photo diodes, switcheseverything in CMOS on very compact silicon-on-insulator substrates."
IBM's next goal will be to increase the delay time to the microsecond-scale, while simultaneously making it tunable. By varying the delay tuning itthe on-chip delay line could synchronize network data traffic at a switch so that data packets make optimal use of the available bandwidth on the network backbone.
"Our main goal is to make the smallest possible footprint and thereby to discover the limits of scaling," said Vlasov.
"We want to create the most compact structures possible for CMOS devices," said Xia. "It is quite a challenge, because we have to demonstrate it is possible to build these very compact waveguides using standard silicon-processing techniques."
The Defense Sciences Office's "Slowing, Storing and Processing Light" program supports IBM's project.