A digital-optical system-on-package module architecture for high performance multiprocessors

Data transport bottlenecks namely memory access delay (MAD) and node crossing delay (NCD) were analyzed using a digital-optical system-on-package module architecture. The MAD bottleneck was minimized by using a 3-D distributed shared memory field in which high speed OIs delivered data to and from each processor. The NCD bottleneck was resolved by the use of non-blocking, high speed, optical switching, each switch having a local controller that was optically synchronized and programmed by an optical header signal. It was found that the coupling efficiency between laser and waveguide was around 45% and the coupling efficiency between waveguide with a 45° end mirror and the PiN PD with a36 μm diameter active area was around 35%.