Levenberg-Marquardt Backpropagation model augmented with Prim's algorithm approach (LMBP) to minimize power in FSM synthesis

Recent advancements in intelligent learning enhance the logic synthesis domain, especially for low power minimization. This paper presents a low power synthesis by implementing AI with conventional greedy Prim's algorithm. Initially, by applying Prim's algorithm, an MST (Minimum Spanning Tree) has been formed from the undirected state transition graph, which has been obtained from a conventional FSM benchmark with the help of a probability method. After finding the MST, the minimum power is found from a weighted Hamming distance by state encoding from the SIS (synthesis for sequential circuit) tool, which is an NP-Hard problem and thus time-consuming. This motivates us to apply the AI integrated technique with the greedy algorithm. The database has been initially generated, and with that Levenberg-Marquardt back propagation model is trained to replace the SIS tool to predict the power for multi-level realizations. The results are impressive, and the model is even more efficient for prediction. Experimental results can show that the augmented-based framework speed up the process by two times compared to the basic simulation-based process.