A power modeling and characterization method for macrocells using structure information

To characterize a macrocell, a general method is to store the power consumption of all possible transition events at primary inputs in the lookup tables. Though this approach is very accurate, the lookup tables could be huge for the macrocells with many inputs. In this paper, we present a new power modeling method which takes advantage of the structure information of macrocells and selects minimum number of primary inputs or internal nodes in a macrocell as state variables to build a state transition graph (STG). Those state variables can completely model the transitions of all internal nodes and the primary outputs. By carefully deleting some state variables, we further introduce an incomplete power modeling technique which can simplify the STG without losing much accuracy. In addition, we exploit the property of the compatible patterns of a macrocell to further reduce the number of edges in the corresponding STG. Experimental results show that our modeling techniques can provide SPICE-like accuracy and can reduce the size of the lookup table significantly comparing to the general approach.