Authors: C.C. McAndrew and P.G. Drennan
Affilation: Freescale Semiconductor, United States
Pages: 698 - 702
Keywords: statistical modeling, SPICE modeling, device correlations
Partial correlations between parameters of different types of devices, such as effective channel length<br>for PMOS and NMOS devices, are often modeled and simulated statistically via correlation coefficients.<br>However this is cumbersome and inefficient from a modeling perspective, from a characterization<br>perspective, and from a simulation perspective. We have found that, with physical understanding,<br>it is possible to formulate models with a combination of parameters that are common to,<br>and completely correlated between, different types of devices and that are unique to,<br>and completely independent between, different types of devices. This gives a modeling<br>basis of independent statistical parameters, which is ideal for simple statistical simulation,<br>that nevertheless completely encompasses statistical correlations between different device types.<br>The key then is how these underlying parameters, which are not directly observable,<br>can be characterized from measurements. We show that by identifying suitable ratios<br>and/or differences of electrical measurements between device types, that “hidden” physical parameters,<br>not observable from direct measurements of a single type of device or electrical performance,<br>can statistically be effectively and easily characterized. We show that the technique gives error free<br>values for the variances of correlated parameters, and allows oxide thickness variation to be characterized<br>from simple DC measurements. As far as we aware, this is the first non-tunneling approach that<br>allows oxide thickness variations to be determined from DC only measurements.