FAST STATIC IR DROP PREDICTION USING MACHINE LEARNING

By Anmol Salvi

INTRODUCTION: Logic Synthesis , High Level Synthesis , IR drop estimation and Lithography hotspot detection all fall under one broad category in a way which is Electronic Design Automation and are all related to each other to a certain extent.IR drop constraint is one of the fundamental requirements that is enforced in almost every chip design. However, it’s evaluation ends up taking a large amount of time and also the mitigation techniques for fixing any violations may require a number of iterations. Thus , fast and more importantly accurate IR drop prediction becomes essential for reducing the overall design turnaround time. Very recently, machine learning (ML) techniques have been studied very actively so as to allow for fast IR drop estimation owing to their potential promise and recorded success in a lot of fields. IR drop which is also called as voltage drop, is defined as the deviation of a power supply level from its specification which occurs when current begins to flow through the power grids. This must be restricted in order for a particular circuit to meet both it’s timing target and function properly. With advancement in design and manufacturing technologies , the increased current load that it has brought also further exaggerates IR drop violations, which then becomes a critical concern for both VLSI design as well as test. Static IR drop can be used to highlight power grid weakness in the design as it takes into account the average current which is drawn from a power grid while assuming average switching conditions.

IMPLEMENTATIONS:  XGBoost , DNN , CNN , ANN , Linear regression are all possible ML techniques that can be employed to help not only provide accurate IR drop predictions but they can also be used to mitigate them.  XGBoost and CNN are both extremely popular ,  and are in fact the best choices when the objective is IR drop mitigation. Also in order to detect and accurately predict Static IR drop , XGBoost definitely emerges as the definitive choice and is the way to proceed . CNN however is the better choice when it comes to accurately predicting Dynamic IR drop. Stemming further into the model used to predict static IR drop , the model takes two parameters as features: one being the PDN’s topology, the other being various electrical features. For the electrical features, it defines features representing both the pullup and pulldown network for every node that has been analysed which is based on the amount of current drawn and also the effective resistance that lies between each power node, voltage sources, and also current loads. The XGBoost model, proposes a flow to mitigate IR drop incrementally with a large number of iterations. In each iteration, based on the prediction that is being made , the designer will perform incremental mitigation and generate an updated layout. In order to avoid retraining the ML model with each iteration, it defines a model update condition, which determines whether the existing ML model still applies to the updated layout or not. The model is only retrained when the testing data ends up satisfying the present update condition. Such an update condition actually considers the model’s robustness quantitatively, which is very rarely discussed in most ML applications in EDA.

ADVANTAGES: A)Saves time.

                           B)Ensures greater accuracy.

                           C)Thereby ensures better performance

CONCLUSION: Thus various ML techniques can be used in order to better optimize the entire process of predicting IR drops accurately and at a much faster rate thus making the integration of ML techniques with EDA an extremely useful and important step towards the future.