Hi,

I am trying to run PSS analysis in order to simulate the noise contribution of my opamp design to a SC integrator. While running pnoise analysis, I repeatedly get a message saying

"Relative residual of linear system is significantly above tolerance level, results computed may be inaccurate".

My problem is that I dont know how I can ensure high enough accuracy. Can anyone suggest a good solution?

The second issue (although I am not sure what impact it has on the simulation) is the PSS analysis itself. Normally, I would expect that the PSS analysis would choose a "tinit" and "tstop" in such a manner that the all PSS analyses begin from a point where there are no strong transients. However, it chooses the starting point between the falling edge of one phase and the rising edge of the second.

The Spectre manual says that the PSS analyses should start from a point where the circuit has had enough time to settle from the last clock edge. I would expect it to start at a point just before any of the clocks changes. In my SC integrator, I would therefore expect the PSS analyses to begin just before the samping phase strarts. However, it begins just after the previous integration phase has ended and the next sampling phase is about to start. Thus, it is off by about half a clock cycle from where I would expect it to start, and nothing I do induces it to change the "tinit". I can only guess that it chooses the "tinit" correctly after all.

The other point I have in mind is that in an SC circuit with a sample phase and an amplification phase, computing the correct transfer function should require that the analysis begins just prior to the sampling phase and ends just around the time when the amplification phase starts. If the whole process is delayed by half a clock cycle, then the computed transfer function is apparently in error, or am I mistaken?

Thanks
Vivek