A novel approach for high-level power modeling of sequential circuits using recurrent neural networks

In this work, we propose a novel power model for CMOS sequential circuits by using recurrent neural networks (RNN) to learn the relationship between input/output signal statistics and the corresponding power dissipation. The complexity of our neural power model has almost no relationship with circuit size and the numbers of inputs, outputs and flip-flops such that this power model can be kept very small even for complex circuits. Using such a simple structure, the neural power models can still have high accuracy because they can automatically consider the nonlinear characteristic of power distributions and the temporal correlation of the input sequences. The experimental results have shown that the estimations are still accurate with smaller variation even for short sequences. It implies that our power model can be used in various applications.