Guarded atomic actions and refinement in a system-on-chip development flow: bridging the specification gap with Event-B

Modern System-on-chip (SoC) hardware design puts considerable pressure on existing design and verification flows, languages and tools. The Register Transfer Level (RTL)description, which forms the input for synchronous, logic synthesis-driven design is at too low a level of abstraction for efficient architectural exploration and re-use. The existing methods for taking a high-level paper specification and refining this specification to an implementation that meets its performance criteria is largely manual and error-prone and as RTL descriptions get larger, a systematic design method is necessary to address explicitly the timing issues that arise when applying logic synthesis to such large blocks.Guarded Atomic Actions have been shown to offer a convenient notation for describing microarchitectures that is amenable to formal reasoning and high-level synthesis. Event-B is a language and method that supports the development of specifications with automatic proof and refinement, based on guarded atomic actions. Latency-insensitive design ensures that a design composed of functionally correct components will be independent of communication latency. A method has been developed which uses Event-B for latency-insensitive SoC component and sub-system design which can be combined with high-level, component synthesis to enable architectural exploration and re-use at the specification level and to close the specification gap in the SoC hardware flow