Nonlinear Timing/Power lookup table for cell lib

[legend03u8z]

How Library Compiler determine the index interval for a 2-D lookup table?
the variables listed in x-axis,and y-axis are not normally distributed.

Thanks

 

[oratie]

LibraryCompiler does not determine the index (interval and count of index). It just convert your .lib file into .db file.

It is library designer (engineer) responsibility to determine number of indexes. It may be done manually (by tunning some spice simulations) or automatically (some library characterization tools may do it). Number and interval of indexes must provide good correlation between StaticTimingAnalysis (based on .lib file) and spice simulation (based on transistor netlist, which mimic the real chip timing).

In the begining of index set, the interval is smaller, then it is increased. It is because, in case of small output pin capacitance, the cell delay (as well as transition) is not linearly increased with capacitance increasing. So, we need to have more indexes in this area.

 

[legend03u8z]

LibraryCompiler does not determine the index (interval and count of index). It just convert your .lib file into .db file.

It is library designer (engineer) responsibility to determine number of indexes. It may be done manually (by tunning some spice simulations) or automatically (some library characterization tools may do it). Number and interval of indexes must provide good correlation between StaticTimingAnalysis (based on .lib file) and spice simulation (based on transistor netlist, which mimic the real chip timing).

In the begining of index set, the interval is smaller, then it is increased. It is because, in case of small output pin capacitance, the cell delay (as well as transition) is not linearly increased with capacitance increasing. So, we need to have more indexes in this area.

First of all thank you very much!

Currently, I'm looking for how the building-lib software work. I have found a equation to generate the number and value of the breakpoints for LUT.
Refered to SiliconSmart userguide, the index can be calculated by = smallest_slew+a(largest_slew-smallest_slew), and a=(I^2.5+I^1.75)/(Imax^2.5+Imax^1.75), where integer I varies from 0 to Imax, and Imax=numsteps_slew-1.

I have done a test based this equation, the resulting indexs are corresponding to a generated .lib file