In this report, a newly modified Newton algorithm (MNA) and a data structure for sparse matrix manipulation are presented for analyzing large-scale electronic circuits on the Cyber-205 supercomputer. The MNA is improved from the Multilevel Newton Algorithm (ML NA) developed by Rabbat Sanjiovanni-Vincentelli, and Hsieh (1979). The time complexity and convergence rate of MNA are analyzed. The computation steps are shown in detail by some example circuits. Scalar and vectorized simulation programs have been tested run on a VAX 11/780 Scalar machine and on the Cyber 205 vector processor at Purdue University. From the results obtained, we observe that the MNA results a speedup of about 100 on the Cyber-205 as compared with using a scalar computer to analyze an electronic circuit containing 500 identical subcircuits.
Date of this Version