Authors: N. Lu
Affilation: IBM, United States
Pages: 818 - 821
Keywords: statistical models, compact models, SPICE models, process variations, spatial correlations
We present an innovative method to model the spatial correlations in semiconductor process and device variations or in VLSI circuit variations. Without using the commonly adopted PCA approach, we give a very compact expression to represent a given spatial correlations among a set of similar statistical variables/instances located at different places on a chip/die. Our compact expression is easy for implementation in a SPICE model and is efficient in circuit simulations. In semiconductor processes and devices as well as in logic circuits, the degree of correlation between any two intra-die instances of a process/device/ circuit decreases with increasing separation between them. Various measured hardware data has revealed such a gradual de-correlation of spatial correlation over distance. Describing intra-die variations using spatial (i.e., distance-dependent) correlations unifies various descriptions of intra-die variations, such as mismatch, across chip variations, random uncorrelated variations, and random correlated variations, etc.