It is there for two reasons. First, in Spice it is a required parameter. And when reading a Spice netlist, we needed something to translate the Spice step parameter to.

Second, it has real meaning, but it is different than the meaning for the Spice parameter. In Spice it set print interval, and the print interval set the maximum time step.

In Spectre, the value of step sets the minimum step size the simulator will use for aesthetic reasons. Meaning that Spectre is watching the activity on the waveforms and will reduce the step size when the waveforms are changing rapidly. It does this for two reasons. In places where the simulator computes a finite-difference derivative, say with capacitor voltage to compute capacitor current, or with inductor current to compute inductor voltage, it must do this to maintain the accuracy of the solution. Any error made in this situation tends to accumulate. However there are other times when it will reduce the timestep because of rapid changes in voltages not associated with capacitor or  currents not associated with inductors. Using large timesteps in this situation does not create an error that accumulates, however it can produce waveforms that are misleading or hard to interpret. So in this case Spectre reduces the time step. The step parameter places a limit on how much the simulator will reduce the timestep in this case. I believe its default value is something like 0.1% of the simulation interval.

-Ken

Paul,
   The step parameter does not specify the minimum time step, it specifies minimum step it uses to maintain the aesthetics of the waveform. If you see Spectre use a smaller step, which it does quite often, it is in order to maintain accuracy or because a device told it that it needed a smaller step.

The transition time on the sources was chosen to be compatible with Spice.

-Ken

Yes, using a small value for step will reduce the chance of a misleading waveform. However, if you have a lot of discontinuous signals, it could also slow the simulation. Generally, I find Spectre's timestep control to be pretty good and as such there is little need to play with step. Though I will reduce it if I am particularly interested in the fidelity of the current waveforms.

-Ken

Paul,
The cmin parameter add small capacitors from every node in the circuit to ground to aid convergence in transient analysis. Sometimes models have discontinuities, and those discontinuities can cause convergence problems. Adding a bit of capacitance can often help transient analysis get past the discontinuity.  However, you should be aware that to Spectre these are not real capacitors. They are modeled as true capacitors, but Spectre feels no obligation to take small timesteps to accurately compute the current through these capacitors.

The cmin parameter is not provided to help with discontinuous signals, which may occur naturally in the circuit. Rather it is available to help with unintensionally discontinuous models.  In general the models are pretty good and cmin is rarely needed. I have not used it in years.

There is nothing that would cause step to be ignored completely. If that were allowed to happen, then Spectre could not get beyond a discontinuity in a signal because it could not find a time step small enough to allow it to smoothly render the jump.

-Ken