Improved statistical method for extraction of MOSFET effective channel length and resistance

The standard electrical method for the extraction of the effective channel length and external resistance of a MOSFET, based upon regression analysis of resistance data, is found to underestimate the variability of these parameters. This was found using statistical experimentation on the method in which various error components were simulated using a Gaussian random-number generator. The problem with the standard method is that it tends to suppress the device-to-device variability inherent in the data and gives confidence intervals that are too narrow and too optimistic. Two alternative methods for data analysis are presented that alleviate this problem: a multiple regression performed on a transformed set of data variables, and an iterative method. Using either of the methods, it is shown that the accuracy of the extracted effective channel length is only as good as the accuracy with which the mask lengths are specified. It is possible, however, to detect the presence of these errors after a data transformation.