Liberate-max_transition

Hi fengye

Just I read your post. I assume your library operates in subthreshod regime. I'm also characterising a subthreshold library and I faced the same issue like you.

The node I'm using is a ULP (not GP) technology with relative high threshold (near to 0.6V). To specify max_transition I run a Monte Carlo simulation on the extracted netlist of the largest bufx12 (not inv) in my library with a Load of 50fF, e.g. for TT-corner (25°) and vdd =0.4V. Then I measured the worst rise/fall transition of the output from 30% to 70%. This is in the mentioned corner 11.243 ns (fall). Actually this cell, with x12 strength, should drive 24fF, but I drived it for a fanout of 2.

For the min_transition I selected for the same cell a load of 0.1fF and I measured the best rise/fall transition of the output from 30% to 70% and halved it. This is 0.3635 ns. For min_output_cap I set the same load of 0.1fF which I use to specify min_transition.

I defined the index_1 / index_2 of all templates (delay, power and constraint) the same: # I'm not sure this strategy is optimal?

{0 1/12 2/12 3/12 ... 12/12}  # actually this is in form of {0 0.08334 0.16667 0.25 0.33334 0.41667 0.5 0.58334 0.66667 0.75 0.83334 0.91667 1} referring to strengths x1 to x12 (as index_1)

in case of index_2 the max load (here 50fF should be assigned to the last value of the index, i.e. '1' and the other loads should be assigned to the other values of the index appropriately (That is what I expected, but I'm not sure?)

However I've an issue with the cells with the smaller than x12 strength. Some cells have, say, max 8 or 6 or even 2 strengths. What happen for such cell with such strategys? e.g for a D-FF there are only two strengths of x1 and x2. It is not possible to drive such load (I mean 50fF) with a D-FF of x1 in subthreshold regime. Therefore it does not make sense to assign such big load to the last value of the D-FF index. One Solution could be to specify for each cell class, i.e. the same count of strength, an index schema, e.g with max 9 values for the cell class with 8 strengths or only 3 values for D-FF. I'm not sure this is a good or even common strategy?

How have you done?

The other question is the simulation of the cell, even largest buf in SS / -40° /0.4V corner to specify max_transition. My cell has issue, i.e the time is too larg. Have you faced also this issue? Or, in other words have you specified also a corner of SS / -40° /0.3V for your library)?

Thaking you in anticipation!

BarPouy

Hi fengye

Just I read your post. I assume your library operates in subthreshod regime. I'm also characterising a subthreshold library and I faced the same issue like you.

The node I'm using is a ULP (not GP) technology with relative high threshold (near to 0.6V). To specify max_transition I run a Monte Carlo simulation on the extracted netlist of the largest bufx12 (not inv) in my library with a Load of 50fF, e.g. for TT-corner (25°) and vdd =0.4V. Then I measured the worst rise/fall transition of the output from 30% to 70%. This is in the mentioned corner 11.243 ns (fall). Actually this cell, with x12 strength, should drive 24fF, but I drived it for a fanout of 2.

For the min_transition I selected for the same cell a load of 0.1fF and I measured the best rise/fall transition of the output from 30% to 70% and halved it. This is 0.3635 ns. For min_output_cap I set the same load of 0.1fF which I use to specify min_transition.

I defined the index_1 / index_2 of all templates (delay, power and constraint) the same: # I'm not sure this strategy is optimal?

{0 1/12 2/12 3/12 ... 12/12}  # actually this is in form of {0 0.08334 0.16667 0.25 0.33334 0.41667 0.5 0.58334 0.66667 0.75 0.83334 0.91667 1} referring to strengths x1 to x12 (as index_1)

in case of index_2 the max load (here 50fF should be assigned to the last value of the index, i.e. '1' and the other loads should be assigned to the other values of the index appropriately (That is what I expected, but I'm not sure?)

However I've an issue with the cells with the smaller than x12 strength. Some cells have, say, max 8 or 6 or even 2 strengths. What happen for such cell with such strategys? e.g for a D-FF there are only two strengths of x1 and x2. It is not possible to drive such load (I mean 50fF) with a D-FF of x1 in subthreshold regime. Therefore it does not make sense to assign such big load to the last value of the D-FF index. One Solution could be to specify for each cell class, i.e. the same count of strength, an index schema, e.g with max 9 values for the cell class with 8 strengths or only 3 values for D-FF. I'm not sure this is a good or even common strategy?

How have you done?

The other question is the simulation of the cell, even largest buf in SS / -40° /0.4V corner to specify max_transition. My cell has issue, i.e the time is too larg. Have you faced also this issue? Or, in other words have you specified also a corner of SS / -40° /0.3V for your library)?

Thaking you in anticipation!

BarPouy