EPSRC Grant JOINT FINAL REPORT

There are at least two contexts in which distributed multi-processor real-time (R-T) systems present increasingly difficult technical challenges to continuing the ‘synchronous-digital-design’ methodology that has been standard practice for many years: (i) complex systems where the parts are widely spaced (and often in relative motion) cannot easily operate with a common clock and (ii) advances in integrated circuit technology are leading to a capability to integrate huge numbers of highperformance processors on a single chip, and commercial pressures are ensuring that these complex components will be utilised in many future systems. The increased speeds and reduced feature sizes make it increasingly problematic to distribute a common clock around such integrated circuits. As a result, there is increased interest in trying to design systems in an asynchronous manner – the potential advantages and the difficulties have been known for many years, and so far, many digital systems designers are not yet convinced of the need to abandon the synchronous design philosophy, but it appears likely that in the two contexts referred to above, there is a good potential for improving performance by a change to an asynchronous design philosophy. An intermediate approach (Globally Asynchronous, Locally Synchronous, known as GALS) is widely expected to become popular. It is within all these contexts that asynchronous communications arise, with a requirement for a reliable means for data transfer between their data-processing subsystems. Traditional techniques using handshakes and message-passing protocols have limitations with regard to important R-T requirements including safety and temporal predictability. Asynchronous data communication mechanisms (ACMs), however, offer a number of advantages as data links between subsystems, including providing full temporal independence for one or both of the communicating sides. The COMFORT project has contributed to a number of advances in the theory and practice of ACMs. Particular emphasis has been on: (i) improving the modelling and analysis techniques for ACMs; (ii) developing and improving systematic approaches to ACM design; (iii) investigating automatic synthesis of ACMs; and (iv) implementing and testing example ACM hardware initial investigations have been made into system construction using ACMs as the “glue” between subsystems. This research has been closely linked with a number of ongoing projects being conducted in both NU and KCL. These include one concerned with the synthesis of systems with heterogeneous timing, with emphasis on multiprocessor R-T systems, one concerned with the visualization aspect of asynchronous system design, and one in the field of nonlinear dynamics. The COMFORT project results have been reported in over 40 academic publications, including PhD theses, book chapters, technical reports, and journal and conference papers, and in seminars presented at various universities. The technical advances have been made available to Matra BAe Systems via our close collaboration with them, and to other academic and industrial partners through our deep participation in the asynchronous systems community in the UK (Asynchronous UK Forum), EU (ACiD Working Group), and worldwide (annual ASYNC symposia and the asynchronous mailing list) and the general circuits and systems community (annual ISCAS symposia and various other IEEE meetings). Furthermore, through its initiation and planning of an international workshop (AINT’2000) in Delft, Netherlands, the COMFORT project contributed to the wider exposure and understanding of ACMs among such industrial companies as IBM, Intel, and Philips.