Tolerance analysis in MOSFET analog integrated circuits

The goal of the design and optimization phase in analog integrated circuits conception is to get the structure yielding the best system performance. The parameters of various components (resistance, capacitance, transistors, etc) of the circuit are set to the optimized nominal values, but no guarantee exists that these values will be conserved during the fabrication process. As a matter of fact, during fabrication the real values of the circuit parameters deviate from the nominal values and consequently the performance of each sample is affected in an unpredictable way. Manufacturers sometimes propose the upper and lower bonds of the acceptability interval, but generally they are obtained after performing measurement of an important number of samples. Furthermore, any circuit whose performance is outside the pre-established interval of acceptability is definitively rejected. When a tolerance analysis is included in the design process, the designer has the possibility to estimate, according to the tolerances imposed on each component, the maximum possible fluctuations of the performance before manufacturing the circuit. The first- and second-order sensitivities of the performance with respect to all circuit parameters must be previously computed. In MOS integrated circuits the designer has no direct access to electrical parameters, thus the tolerance analysis with respect to these parameters is insufficient. Our work consists in developing a procedure to compute the sensitivities of the performance of interest with respect to the geometrical parameters of MOSFETs, which are the only parameters under designer's control. A software package has been developed to link these sensitivities to those with respect to the electrical parameters, as in any traditional approach. From these results, and using the tolerances provided by the manufacturer on the geometrical and technological parameters of the MOSFETs, we can accede to the lower- and upper-bonds of performance fluctuations around its nominal value.