Improved methodology for better accuracy on transistors matching characterization

An improved methodology for the characterization of matching parameters in MOSFETs and bipolar transistors is presented. Because of their statistical nature, only estimation of matching parameters can be provided from given measurements. Considering general σ_Δp = A_p / √(WL) transistor matching model, the aim of this paper is to discuss, for the first time, the most relevant matching parameter A_p estimation accuracy. Dispersion on A_p estimation when using conventional least squares regression on σ_Δp vs. 1/√(WL) plots is analytically characterized. Then it is shown that replacing the conventional least squares regression by a weighted least squares regression leads to increased accuracy on A_p estimation and consequently allows a better detection of physical effects responsible for mismatch.