Does ngspice and the GF180mcu PDK support transient noise simulation?

Yes it does

That is excellent news. How do I turn it on?

Please see the discussion with Lukas Bongartz on Nov. 22 in this channel. There are some different opinions as to whether transient noise simulations can be done reliably, or at all.

👍

You can't run meaningful transient noise analyses in ngspice, it does not have the required time step management. You can read the documentation.

Boris, could you give a short description on the time step management required here?

You probably know a lot more than me about this, but in most cases that I'd be interested in, it is a requirement to process white noise up to tens if not hundreds of GHz, i.e. mandating time steps in the single digit ps. The way the time step is described in the manual (15.3.11), as a fixed quantify that equally divides the total simulation time, does not seem to be practical for this (and I haven't been able to get good results). A related hint the manual gives is: Transient noise analysis (at low frequency). Eldo, BDA, Spectre, others, seem to have found a way to manage the time step requirements differently. You of course still need to specify how far out in the spectrum you want to look, but the simulation time does not blow up under a small time step requirement. A good test case is an RC circuit with a 3-dB bandwidth of 10 GHz (a model for a high speed sampling switch). One needs to look at noise up to about 100 GHz to get the well-known sqrt(kT/C) result for the observed RMS noise.

Sorry for bumping an old thread, but this approach seems to be generating a transient signal by sampling from a Gaussian distribution at a rate (1/NT) such that the Nyquist frequency is the upper end of your noise power integrating range. Would it be possible to (1) set up the circuit solver to try to find AC magnitude up to the cutoff for noise power integration for transient steps larger than NT, (2) apply appropriate time shifts to the AC magnitude for each time slice sampled, and (3) use Parseval's on the sum of all shifted AC magnitude functions to find the overall frequency domain transfer function between a noise source and a target node? The closest analogy I can think of is trying to find the average frequency domain expression of a signal by averaging all of the slices of a spectrogram. Part of me thinks this shouldn't be possible since it seems like a way to cut around the Nyquist criteria.