Asynchronous scheduling/binding using a genetic approach

In this paper a new approach to operation scheduling and binding in asynchronous High Level Synthesis (HLS) is presented. We developed a genetic algorithm and integrated it inside Pipefitter, an existing tool for the automated synthesis of asynchronous circuit. A Control Data Flow Graph (CDFG), derived by Pipefitter from an HDL specification, is the input format for our algorithm. The designer can steer the search of an optimized solution either in the direction of minimum area or in the direction of maximum throughput. In the final solution each operation in the CDFG will be assigned to an operative unit (static binding), while the execution sequence will be determined run-time by the control unit (dynamic scheduling) in order to improve performances. This solution is then returned to Pipefitter that will complete the synthesis process down to the layout level.