Reliability considerations in parallel optical interconnects

Random failure in a population of lasers after burn-in has profound consequences for the deployment of the optical backplane in high performance computing. It is argued that without an effective field repair strategy the mean time between fails for a large population of lasers can become prohibitive. A field repair strategy must be an integral part of the system design from the outset and must be based on low component replacement cost, minimum system down time, minimum redundancy, and low mass fabrication cost for the replaceable sub-assembly. It is further argued that an example of an acceptable repair strategy is based on an active and electrically pluggable optical harness consisting of a parallel array of polymer waveguides on a flexible substrate. Lasers and photodetectors are pre-aligned to the waveguides and make the harness active. Low cost is ensured by a parallel fabrication process in which all waveguides are simultaneously aligned to their respective lasers and photodetectors in one step.