Structural technique for fault-masking in asynchronous interfaces

Abstract

Asynchronous VLSI circuits have been proven to be more tolerant to persistent defects, such as stuck-at faults, than their clocked counterparts. However, such circuits are directly reactive to input stimuli and so they can be more vulnerable to transient faults at their inputs. The ability to tolerate such faults is most crucial for interface circuits, or transducers, which are the informational kernel of any system. For example, an unspecified signal change at some of the transducer's links can produce a harmful effect on other links and influence the behaviour of the whole system. Types of faults and specific requirements of asynchronous transducers, under the various assumptions about the correctness of their and the environment's implementation, are investigated. A structural method, based on synthesising a correct circuit for the transducer from its original specification made under the assumption that the environment is correct, and then augmenting it by a structurally separate ‘wrapping’ of a special protection logic, is proposed. This logic consists of the perfect mirror model and adjudicator components which are built separately for each of the transducer links. The discussion makes use of the formalism of trace structures and their conformance introduced by Dill. The approach appears to be most efficient when the links between the transducer and the environment utilise standard handshake protocols.