Statistical Analysis Modeling Of MOS Device Based On Process Drift

With the development of integrated circuit and technology, IC design has already in the era of sub-micro meter. The influence of process variation on the circuit performance is becoming more and more significant, and the cost of producing the IC becomes higher. In order to make IC products meet the need of expectable design and improve the design yield, besides a accurate SPICE device modeling, we also need a statistical model to describe the influence of process variation.In this article, we focus on a certain technological device bsim3v3model. With HSPICE simulator, firstly we use standard Worst Case method and Monte Carlo analysis method to extract the corner model for device, then draw the performance distribution histograms. Based on the histograms and corner model, we compare the performance distribution with the measurement data for these two methods, and analyzes their advantages and disadvantages. Then another method Location Depth Corner Model Analysis(LDCM) is introduced. With using the generated corner model based on LDCM method, we analyze the simulation result. It shows similar accuracy with the result using Monte Carlo method, and it is fast compared with Monte Carlo method. At last, we describe Response Surface Method(RSM). Instead of using Monte Carlo method,RSM method can save much time and still reflect the influence of process variation on the simulation result. And for large scale integrated circuit, Response Surface Method can reduce the simulation time significantly.In conclusion, this article analyzes and verified the variation of device performance caused by process variation with standard Worst Case method, Monte Carlo analysis, LDCM method and RSM method. According the comparison between simulation result and measurement data, Monte Carlo analysis has the highest accuracy, and the LDCM method and RSM method can be a good compromise between accuracy and speed.