Advanced Cell Modeling Techniques Based on Polynomial Expressions

Abstract

This paper presents an advance cell modeling technique based on polynomial expressions for physical and logic design. Till now, there lacks a modeling method to handle multiple operating points such as input slew, output capacitance, voltage, process, temperature, etc. at a time for the logic synthesis and physical design. Our novel curve-fitting algorithm for cell modeling can process a table with thousands of data points in a modeling equation compared to piecewise equations used in the existing methods [1]-[4]. The number of data storage and CPU run time are significantly reduced without compromising the accuracy of the data. This ensures that the novel curve-fitting algorithm can exactly recover the original data points in a short period of time. In addition, the table-partitioning algorithm is developed to reduce the number of storage and flatten the peak situation occurring at some of the unsampled data. Furthermore, in order to filter the “pass" points with 99% of all data points, the algorithm is developed to select the best order for input variables, further causes to shorten CPU run time again. Our benchmark shows that the new approach has significantly better accuracy and less storage for both the sampling data and unsampling data.