This is definitely a magic number.

Almost e

Hmm.

You can find some by searching "VLSI exponential horn"

cc @Philipp Gühring

Optimal with parasitics is ~3.6 not 2.6... shrug

Is this related to fanout? I had a professor who had done some studies on where to put in inverters for optimal drive strengths or something like that

@Olof Kindgren: No. This is just where they (or their tools) choose values for the timing tables. These values are the indexes into the tables, so they're independent variables. They just need to be chosen in a way that the values cover a large range of timing, and the STA tools (1) don't have to extrapolate out into the unknown, and (2) don't get too far off the actual timing value internally. If the actual surface being represented by the timing tables in the liberty file is exponentially related to slew and capacitance, then it makes sense to sample the slew and capacitance at exponentially spaced values to minimize the error when the STA tool interpolates between points. So it doesn't matter what the spacing is. You choose your low and high values to be within an expected range (probably with margin), then you choose the number of points per index to minimize the error caused by worst-case interpolation, and then you go measure the timing values at all those points and create your tables. Or else you say screw the analysis, choose seven points because that's what everybody else does, and pick some endpoints that somebody else had in their liberty file for a similar process, and call it done. I really don't know how rigorous people are about these things.