The authors describe how systematic mapping methodologies can be used to derive special-purpose processor arrays for the estimation of the bispectrum via the third-order moments. A novel design that is optimal in terms of total execution time for multiple pipelined data blocks is proposed, and it is shown how formal verification of the design can be achieved analytically. The discussion focuses on mapping regular algorithms to parallel architectures, bispectrum estimation from the third-order moments, and systolic array design for the third-order moments. >