SpectreRF Turbo: Parasitic Reduction

I wanted to share some experiences I had with SpectreRF-Turbo and Parasitic reduction on a recent large benchmark. The things I learned may be helpful to anyone who wants to get the most out of turbo and parasitic reduction.

Available in MMSIM7.0.1 and later for RF, turbo accelerates simulation in three ways: 1) by more efficiently evaluating the BSIM3 and BSIM4 devices, 2) the matrix is handled more effeciently and 3) turbo divides the job accross multi-core machines.

Parasitic reduction reduces the RC matrix on a nodal basis, but does not currently reduce the L and K parasitics. If you are going to extract L and K, its best to create regions where the L and K are critical and just extract them in those regions. This helps the RC reduction to be more efficient.

Suggested Settings:

In the ADE window, go to Setup->Turbo/Parasitic Reduction...

1) Errpreset = Do not Override (default)

2) Multithreading = Auto (default)

3) Parasitic Reduction options = RF

4) One more setting that helps trim out the small resistors. In the ADE window, go to Simulaton->Options->Analog and go to the bottom of the list where it says "Additional Arguments". Type: rshort=.001 in the field. Extraction tools tend to dump out a ton of really small resistors (< 1 mohm) and this trims those out without affecting the solution.

We are working on a way to trim the small K elements as well with a new parameter. I will keep you posted on that, but it should be out very soon.

The sweet spot for Turbo and Parasitic Reduction is large designs with BSIM devices. As a typical example, I have seen reductions from 8h to 1.5h and memory usage go from 58G to 19G on a very large receiver with 23K devices and 365K parasitic devices. I have also used parasitic reduction on a large circuit with no parasitics and found a substantial speed up. This is due to the fact that the parasitic reduction works on a nodal basis and not just on nets with parasitics. Using turbo/parasitic reduction on small circuits with low device count has small gains as you might imagine.

I hope this helps. If you find this kind of post helpful, I will continue to put up "Best Practices" posts to share insights as we uncover them.