Real circuits do not operate in a world with fixed values of device parameters, power supplies and environmental data. Even if a ngspice output offers 5 digits or more of precision, this should not mislead you thinking that your circuits will behave exactly the same. All physical parameters influencing a circuit (e.g. MOS Source/drain resistance, threshold voltage, transconductance) are distributed parameters, often following a Gaussian distribution with a mean value (\mu)and a standard deviation (\sigma).

To obtain circuits operating reliably under varying parameters, it might be necessary to simulate them taking certain parameter spreads into account. ngspice offers several methods supporting this task. A powerful random number generator is working in the background. It is not providing true random numbers, but a long sequence of pseudo random numbers. This sequence depends on a seed value. The same seed valuue will deliver the same sequence of random numbers.

ngspice offers several methods to set this seed value. If no input is given, then ngspice sets the seed (stored in variable rndseed) to 1 upon start up. With the option SEED you may either set a value to rndseed upon start up of ngspice (option SEED=nn, nn is an integer greater than 0), or obtain a “random” number as seed, that is the number of seconds since 01.01.1970 (option SEED=random). This command is best set in .spiceinit (16.6). With the command setseed (see chapt.17.5.67) you may choose any other seed value (integer greater than 0).

The following three chapters offer a short introduction to the statistical methods available in ngspice. The diversity of approaches stems from historical reasons, and from some efforts to make ngspice compatible to other simulators.