Convergence Developments

Simulation Precision and Iteration Options

Most SPICE programs use double precision arithmetic throughout for all calculations. Double precision calculations are performed to an accuracy of approximately 16 decimal digits. This is often inadequate due to numerical noise; this is similar in concept to quantization noise found in data converters. Numerical noise increases inversely to the time step and in some cases can exceed the tolerance requirements for iteration convergence. Numerical noise is a major source of non-convergence in transient analysis especially in power circuits.

Since its inception, SIMetrix has used extended precision in its matrix solver to reduce numerical noise and enhance convergence. Extended precision has a decimal accuracy of approximately 19 digits. From version 8.1, SIMetrix now provides a new mode which also uses extended precision in the device equation calculation. This mode also uses the advanced iteration algorithm introduced in version 8.0 to provide a simulation mode that can solve difficult circuits even with tight tolerance requirements.

As well as extended precision, SIMetrix 8.1 also offers two more modes that use quad precision in both the matrix solver and device equation evaluation. Quad precision has 34 decimal digits of accuracy. With such high precision, numerical noise is eliminated as a source of convergence failure. This makes diagnosis of convergence problems caused by other factors easier to locate as numerical noise tends to mask other causes.

Back to What's New Overview