The unscented transform (UT) is proposed as an alternative to the Monte Carlo (MC) method for assessing performance variability of electronic circuits based on emergent technologies. We show that the discrete UT approximation of a continuous probability distribution of random variables can be solved by Gaussian quadrature. As an application example, we compare UT and MC yield predictions for CNTFET current-mode-logic ring oscillators. We discuss the exponential nature of our first implementation and point to the use of sparse grids and symmetry analysis as means to optimize the technique.