Synthesis of multi-cycle circuits from guarded atomic actions

Abstract

One solution to the timing closure problem is to perform infrequent operations in more than one clock cycle. Despite the apparent simplicity of the solution statement, it is not easily considered because it requires changes in RTL, which in turn exacerbates the verification problem. Another approach to the problem is to avoid it altogether, by using a high-level design methodology and allow the synthesis tool to generate the design that matches design requirements. This approach hinges on the ability of the tool to be able to generate satisfactory RTL from the high-level description, an ability which often cannot be tested until late in the project. Failure to meet the requirements can result in costly delays as an alternative way of expressing the design intent is sought and experimented with. We offer a timing closure solution that does not suffer from these problems. We have selected atomic actions as the high-level design methodology. We exploit the fact that semantics of atomic actions are untimed, that is, the time to execute an action does not change its outcome. The current hardware synthesis technique from atomic actions assumes that each action takes one clock cycle to complete its computation. Consequently, the action with the longest combinational path determines the clock cycle of the entire design, often leading to needlessly slow circuits. By augmenting the description of the actions with desired timing information, we allow the designer to split long paths over multiple clock cycles without giving up the semantics of atomicity. We also introduce loops with dynamic bounds into the atomic action description. These loops are not unrolled for synthesis, but the guards are evaluated for each iteration. Our synthesis results show that the clock speed and performance of circuits can be improved substantially with our technique, without having to substantially change the design.