Fault-Tolerant Delay-Insensitive Inter-Chip Communication

Y. Shi, S.B. Furber, J.D. Garside and L.A. Plana

Abstract

Asynchronous interconnect is a promising technology for communication systems. Delay Insensitive (DI) interconnect eliminates relative timing assumptions, offering a robust and flexible approach to on- and inter-chip communication. In the SpiNNaker system - a massively parallel computation platform -a DI system-wide communication infrastructure is employed which uses a 4-phase 3-of-6 code for on-chip communication and a 2-phase 2-of-7 code for inter-chip communication. Fault-tolerance has been evaluated by randomly injecting transient glitches into the off-chip wires. Fault simulation reveals that deadlock may occur in either the transmitter or the receiver as handshake protocols are disrupted. Various methods have been tested for reducing or eliminating deadlock, including a novel phase-insensitive 2-phase to 4-phase converter, a priority arbiter for reliable code conversion and a scheme that allows independent resetting of the transmitter and receiver to clear deadlocks. Simulation results confirm that these methods enhance the fault tolerance of the DI communication link, in particular making it significantly more resistant to deadlock.

PDF (379K) IEEE Copyright