Polynomial pattern finding in scattered data

A new numerical procedure to extract the threshold voltage in MOSFET transistors has been developed by means of polynomial pattern recognition. The technique proposed here is based on the use of particular properties of discrete orthogonal Chebyshev polynomials, it allows the extraction of polynomial curves of different degrees within a set of experimental or simulated data. For the MOSFET threshold voltage determination we have detected linear patterns in the logarithmic representation of MOSFET transfer characteristics (drain current versus gate voltage curves). The results have been compared with the threshold voltage obtained with a classical technique, the transconductance change method, where the maximum of the drain current second derivative is assumed as the threshold voltage. Reasonable and comparable results are obtained. The new technique has shown more immunity to numerical and measurement noise, which is an important feature in the current industrial context.