RF calibration of on-chip DfT chain by DC stimuli and statistical multivariate regression technique

The problem of parameter variability in RF and analog circuits is escalating with CMOS scaling. Consequently every RF chip produced in nano-meter CMOS technologies needs to be tested. On-chip Design for Testability (DfT) features, which are meant to reduce test time and cost also suffer from parameter variability. Therefore, RF calibration of all on-chip test structures is mandatory. In this paper, Artificial Neural Networks (ANN) are employed as a multivariate regression technique to architect a RF calibration scheme for DfT chain using DC- instead of RF (GHz) stimuli. The use of DC stimuli relaxes the package design and on-chip routing that results in test cost reduction. A DfT circuit (RF detector, Test-ADC, Test-DAC and multiplexers) designed in 65 nm CMOS is used to demonstrate the proposed calibration scheme. The simulation results show that the cumulative variation in a DfT circuit due to process and mismatch can be estimated and successfully calibrated, i.e. 25% error due to process variation in DfT circuit can be reduced to 2.5% provided the input test stimuli is large in magnitude. This reduction in error makes parametric tests feasible to classify the bad and good dies especially before expensive RF packaging.