PATTERN-BASED MAXIMAL POWER ESTIMATION FOR VLSI CHIP DESIGN

In recently year, the analysis of power management becomes more import ant. It is difficult to obtain the maximum power because this is NP-co mplete. For an n-input circuit, there are 2(2n) different input patter ns to be considered. There are two major methods for this problem. Fir st method is to generate input patterns to obtain the maximal power by simulating these generated patterns. This method is called pattern ba sed. The other one uses probability method to estimate the power densi ty of each node of a circuit to calculate the maximal power. In this p aper, we use a pattern based method to estimate the maximal power. Thi s method is better than that of probability for the simulation of powe r activity. In practical applications, these generated patterns can be applied and observe the activity of a circuit. These simulated data c an be used to examined the critical paths for performance optimization . A simulated annealing algorithm is proposed to search input patterns for maximum power. Firstly, it transforms this problem into an optimi zation problem to adapt the simulated annealing method. In this method , there are three strategies for generating the next input patterns, c alled neighborhood. In the first strategy, it generates the next input pattern by changing the status of all input nodes. In the second stra tegy, some input nodes are selected and changed randomly. In the last strategy, it separates the input nodes into many groups and then selec ts one, randomly, to change its status. These different strategies hav e different solution space for the same input space. They will improve the searching speed and the accurate result. When a new result is acc epted we use a Walk-Through method to search the local optimal result. To analyze the cost function in simulated annealing, a power dissipat ion model is proposed including the glitch phenomena, fanout load, and circuit type. For the criterion of equilibrium condition of annealing temperature, we use a statistical model to estimate the optimal itera tion number in a temperature.