Efficient quadratic approximation for statistical design

A highly efficient approach to quadratic approximation of circuit responses is presented. Because it uses a fixed pattern of base points, this approach requires extremely small amounts of CPU time and storage space. Using this approach, a major obstacle for the traditional quadratic approximation for dealing with large problems, namely, the prohibitive requirement for storage and computational effort, is effectively eliminated. The accuracy and efficiency of this quadratic approximation approach are demonstrated by two statistical circuit design examples. >