Power Models for Semi-Autonomous RTL Macros

Most power macromodels for RTL datapath modules are both data-dependent and activity-sensitive, that is, they model power in terms of some activity measure of the data inputs of the module. These models have proved to be quite accurate for most combinational RTL datapath macros (such as adders and multipliers), as well as for storage units (such as registers). They tend to become inadequate for RTL modules that are control-dominated, that is, having a set of control inputs that exercise different operational behaviors. Furthermore, some of these behaviors may be input-insensitive, that is, they let the module evolve (and thus consume power) in a semi-autonomous way, independently of the input activity. We propose a procedure for the construction of ad-hoc power models for semi-autonomous RTL macros. Our approach is based on the analysis of the functional effect of such control inputs on specific macros. Although the resulting models are tailored to individual macros, the model construction procedure keeps the desirable property of being automatic.