Effect of Large Voltage (1 MV) on Accuracy When Using relref=sigglobal in Spectre APS/X

Hi,

I’m running transient simulations in Spectre APS/X and I’m concerned about how accuracy tolerances are applied when relref=sigglobal is used.

In my setup, I generate a very large control voltage (~1 MV) to drive a Verilog-A block (for e.g, a 20bit ADC). This high value is only for producing internal control signals and is not part of the “real” signal path.

I know that with relref=sigglobal, Spectre uses the largest node voltage as the global reference for relative error (reltol). In theory, this could loosen the error checks for low-voltage nodes.

To test this, I built a simple circuit: one branch with a 1 MV source connected through a resistor to ground, and another branch with a 1 V sine wave driving an RC filter. I compared relref=sigglobal with alllocal/pointlocal and found no significant difference in waveforms or convergence.

However, I’m not sure if the same holds true for large, complex circuits like a full transistor-level RF PLL, where nonlinearities and tighter accuracy demands might make the impact more visible.

So my questions are:

1. Does a very large node voltage actually affect accuracy in practice when relref=sigglobal is used?

2. Why doesn’t the simple experiment show any difference?

3. In a more complex circuit (e.g., PLL), could this become a real accuracy issue?

4. Is it recommended to avoid such artificial high voltages in Verilog-A, or is it safe as long as vabstol/iabstol and LTE checks remain in place?

Any insights or clarification would be greatly appreciated.

Thanks!