search

The-OpenROAD-Project / OpenSTA

OpenSTA engine
GNU General Public License v3.0
404 stars 173 forks source link

About driving resistance and the Arnoldi model order reduction #165

Closed gzz2000 closed 1 year ago

gzz2000 commented 1 year ago

I know this should be better posted as a discussion. I have a technical question regarding the driving resistance implementation in the reduced order model.

In OpenSTA, the original R and C matrices are first gone through the Krylov subspace computation. After that, the driving resistance is computed, and added to the top-leftmost element in the reduced order model matrix (the 5x5 matrix H in the original CT-Arnoldi paper) after multiplied by the total capacitance.

This mysteriously works very well and my experiment shows that the only changed element is the top-leftmost element, and it is indeed linearly dependent on the applied driving resistance. I wonder if there is any mathematical reference of this property. Thanks for your help and your elegant open-source Arnoldi implementation!

jjcherry56 commented 1 year ago

The only documentation of the arnoldi delay calculator implementation is in dcalc/Arnoldi.txt

gzz2000 commented 1 year ago

Hi James, thanks for pointing me to the documentation. Do you know how to contact the original author William Scott? It seems some technical details are not present there and I want to ask if he could kindly give me some hints. Thank you.

jjcherry56 commented 1 year ago

I do not know how to get in touch with him. If you find anything additional please let me know.