Artificial Neural Network-based Prediction and Alleviation of Congestion during Placement

In this paper, we propose an artificial neural network (ANN) based model to predict global routing congestion during placement. We derive relevant design features impacting routing congestion by computing correlation coefficients and variance inflation factor analysis for various design attributes. We demonstrate the effectiveness of the proposed technique on ISPD 2014/2015 benchmark designs and compare the results with the golden global routing congestion obtained using a commercial placement and routing (PnR) tool. The results show that the proposed ANN model can reduce the root mean square error (RMSE) of the predicted congestion by 68% compared to the prediction done by the PnR tool employing early global routing. Additionally, we demonstrate that we can reduce the routing congestion by leveraging more accurate congestion values predicted by the ANN model, leading to a reduction of 8 - 30% congestion on the benchmark designs. Thus, the proposed ANN-based congestion prediction and alleviation framework can help reduce the design turnaround time and improve figures of merit.