A Hybrid Approach to Automated Power and IR drop reduction and Classification using Support Vector Machines

With the evolving technology and designs being scaled drastically, power dissipation and IR drops are some of the most challenging aspects of a chip production. This paper deals with Python based automatic key parameter extraction to train a system efficient enough to classify cells of any design into critical and normal classes. Support Vector Machines (SVM) with radial basis function has been exploited to train the system and classify with an error rate of only 0.7%. Standard mitigation techniques like use of MTCMOS/ VTCMOS, decap cells and cell padding have been used to achieve static power reduction by 85.4% and IR drop reduction by 36.34%.