Wrong Constraint Values in Sequential Cell Characterization

Hi Anuradha,

PLEASE USE LATEST LIBERATE RELEASE, WHEN YOU HAVE AN ISSUE.

My questions/suggestions are,

1. use select_index {1} before loading the template

2. save the pass deck

3. re-run the decks for the constraint that you have problem with.

3a. Assuming you do not see a FAIL for the first 3-5 iteration. Normally, the last 5-10 iterations are for final constraint searching. 

3b. in your logfile, just before the simulation starts, you should see what criteria the tool chooses to use for the arc. For pass/fail look for the point where the output no longer has a full transition or the glitch reaches the threshold you have set(or default). for degradation, find out where the delay is pushed out by the threshold (defined by constraint_delay_degrade).

Guangjun

Hi Guangjun,

Thanks for your input. So this is what I did to debug the issue : 

1. Added following debug commands to the script, so that the pass/fail arc instances with the measurement criteria can be clearly observed: 

select_index -style 1x1 ;# run only 1st point in table

set_var bisection_info 4 ;# print additional bisection search info to output log
set_var power_info 2 ;# print additional power calculation info of table's 1st point to file decks/po

set_var ski_enable 0
set_var extsim_save_passed all ;# save all run decks and output files
set_var extsim_save_failed all ;# save all run decks and output files
set_var extsim_save_verify 2 ;# save verify deck
set_var extsim_deck_dir [file normalize "decks"] ;# specify directory for SPICE decks and output files
set_var extsim_tar_cmd "" ;# disable tgz of simulation decks and run logs

2. The log file says, Liberate uses the default "degradation" criteria for the constraint measurements. By default this must be 10% (?)

It seemed that the automatic bisection search bound was too large to determine the optimal setup/hold times, so I modified following : 

set_var constraint_search_bound 3e-10
set_var constraint_search_bound_estimation_mode 3

This somewhat improves the accuracy, but I still have few problems :

1) Even with the above settings, the bisection search step size is still seen to be larger. How can I reduce the step size ? Although there is a variable called constraint_combinational_step_size, this does not seem to have any effect on the actual step size between two iterations !  And how can I add few more iterations assuming the degradation criteria is not violated ?

INFO (LIB-405): Search setup_5/setup_5_001(TSPC18): type=setup_rising, when=, entry=0, iteration=1, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-3.0125e-10, delay(CP-R->Q-R)=1.21979e-09, criteria: degradation, passed
INFO (LIB-405): Search setup_5(TSPC18): type=setup_rising, when=, entry=0, iteration=2, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=3.0125e-10, delay(CP-R->Q-R)=-1, criteria: degradation, assumed failed
INFO (LIB-405): Search setup_5/setup_5_001_iter002(TSPC18): type=setup_rising, when=, entry=0, iteration=3, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=0, delay(CP-R->Q-R)=-1, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter003(TSPC18): type=setup_rising, when=, entry=0, iteration=4, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-1.50625e-10, delay(CP-R->Q-R)=-1, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter003(TSPC18): type=setup_rising, when=, entry=0, iteration=4, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-1.50625e-10, delay(MNM11_d)=0.239033, criteria: settled, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter004(TSPC18): type=setup_rising, when=, entry=0, iteration=5, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.25938e-10, delay(CP-R->Q-R)=1.30979e-09, criteria: degradation, passed
INFO (LIB-405): Search setup_5/setup_5_001_iter005(TSPC18): type=setup_rising, when=, entry=0, iteration=6, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-1.88281e-10, delay(CP-R->Q-R)=1.43892e-09, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter006(TSPC18): type=setup_rising, when=, entry=0, iteration=7, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.1661e-10, delay(CP-R->Q-R)=1.33195e-09, criteria: degradation, passed
INFO (LIB-405): Search setup_5/setup_5_001_iter007(TSPC18): type=setup_rising, when=, entry=0, iteration=8, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.02446e-10, delay(CP-R->Q-R)=1.37599e-09, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter008(TSPC18): type=setup_rising, when=, entry=0, iteration=9, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.09528e-10, delay(CP-R->Q-R)=1.35207e-09, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter009(TSPC18): type=setup_rising, when=, entry=0, iteration=10, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.13069e-10, delay(CP-R->Q-R)=1.34162e-09, criteria: degradation, passed

2) When we say 10% of delay degradation for "setup", does it mean that the maximum allowable delay on the clk->Q arc is 10% increase of the minimum clk-Q delay (reference) ?

3) How does Liberate measure this reference value ?

Anuradha 

Hi Guangjun,

Thanks for your input. So this is what I did to debug the issue : 

1. Added following debug commands to the script, so that the pass/fail arc instances with the measurement criteria can be clearly observed: 

select_index -style 1x1 ;# run only 1st point in table

set_var bisection_info 4 ;# print additional bisection search info to output log
set_var power_info 2 ;# print additional power calculation info of table's 1st point to file decks/po

set_var ski_enable 0
set_var extsim_save_passed all ;# save all run decks and output files
set_var extsim_save_failed all ;# save all run decks and output files
set_var extsim_save_verify 2 ;# save verify deck
set_var extsim_deck_dir [file normalize "decks"] ;# specify directory for SPICE decks and output files
set_var extsim_tar_cmd "" ;# disable tgz of simulation decks and run logs

2. The log file says, Liberate uses the default "degradation" criteria for the constraint measurements. By default this must be 10% (?)

It seemed that the automatic bisection search bound was too large to determine the optimal setup/hold times, so I modified following : 

set_var constraint_search_bound 3e-10
set_var constraint_search_bound_estimation_mode 3

This somewhat improves the accuracy, but I still have few problems :

1) Even with the above settings, the bisection search step size is still seen to be larger. How can I reduce the step size ? Although there is a variable called constraint_combinational_step_size, this does not seem to have any effect on the actual step size between two iterations !  And how can I add few more iterations assuming the degradation criteria is not violated ?

INFO (LIB-405): Search setup_5/setup_5_001(TSPC18): type=setup_rising, when=, entry=0, iteration=1, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-3.0125e-10, delay(CP-R->Q-R)=1.21979e-09, criteria: degradation, passed
INFO (LIB-405): Search setup_5(TSPC18): type=setup_rising, when=, entry=0, iteration=2, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=3.0125e-10, delay(CP-R->Q-R)=-1, criteria: degradation, assumed failed
INFO (LIB-405): Search setup_5/setup_5_001_iter002(TSPC18): type=setup_rising, when=, entry=0, iteration=3, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=0, delay(CP-R->Q-R)=-1, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter003(TSPC18): type=setup_rising, when=, entry=0, iteration=4, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-1.50625e-10, delay(CP-R->Q-R)=-1, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter003(TSPC18): type=setup_rising, when=, entry=0, iteration=4, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-1.50625e-10, delay(MNM11_d)=0.239033, criteria: settled, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter004(TSPC18): type=setup_rising, when=, entry=0, iteration=5, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.25938e-10, delay(CP-R->Q-R)=1.30979e-09, criteria: degradation, passed
INFO (LIB-405): Search setup_5/setup_5_001_iter005(TSPC18): type=setup_rising, when=, entry=0, iteration=6, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-1.88281e-10, delay(CP-R->Q-R)=1.43892e-09, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter006(TSPC18): type=setup_rising, when=, entry=0, iteration=7, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.1661e-10, delay(CP-R->Q-R)=1.33195e-09, criteria: degradation, passed
INFO (LIB-405): Search setup_5/setup_5_001_iter007(TSPC18): type=setup_rising, when=, entry=0, iteration=8, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.02446e-10, delay(CP-R->Q-R)=1.37599e-09, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter008(TSPC18): type=setup_rising, when=, entry=0, iteration=9, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.09528e-10, delay(CP-R->Q-R)=1.35207e-09, criteria: degradation, failed
INFO (LIB-405): Search setup_5/setup_5_001_iter009(TSPC18): type=setup_rising, when=, entry=0, iteration=10, cslew(D-R)=2.5e-12, rslew(CP-R)=2.5e-12, align(D-R->CP-R)=-2.13069e-10, delay(CP-R->Q-R)=1.34162e-09, criteria: degradation, passed

2) When we say 10% of delay degradation for "setup", does it mean that the maximum allowable delay on the clk->Q arc is 10% increase of the minimum clk-Q delay (reference) ?

3) How does Liberate measure this reference value ?

Anuradha 

Have you checked the waveforms ? This is the must-do, unless you fins an obvious issue and solution. The logfile information can be used as a guide, in terms of which waveforms to start with or to look into.

constraint_delay_degrade=0.1 by default. The delay does not change if the two signals are set up well before the constraint. you should find this very obviously through the waveform. constraint_delay_degrade_abstol(default 1ps, minimum supported 0.1ps) is another variable that sets the delay/degradation criteria. 

if constraint_combinational=0, then the min step may be controlled by constraint_search_time_abstol.

I would not recommend change the default constraint_search_bound_estimation_mode and even other settings for constraint, unless you checked have checked the waveform and know what parameter to change.

guangjun

Hi Guang,

Guangjun Cao said:

The delay does not change if the two signals are set up well before the constraint

This is obvious. So according to your answer what Liberate seems to be doing is, first, two signals are setup well in advance and measure the minimum clk-Q delay and based on that calculate the point where this delay degrades by 10%.

Guangjun Cao said:

Have you checked the waveforms ?

Yes, I have. I don't see any problem in the waveforms for given iterations. The thing is when I set the set_var constraint_search_bound 3e-10, the search algorithm works better and arrives to more accurate measurement than automatically determined bounds.

So my last doubt : 

When both constraint_delay_degrade=0.1 and constraint_delay_degrade_abstol = 1ps (by default) are set, what is the criteria for delay degradation measurement ? Because I do not see the 10% degradation point is met  in the iterations (or in waveforms). i.e. The last "pass iteration" shown for the setup time in the log does not correspond to the 10% degradation point (initial constant clk-Q delay=65 ns, 10% degradation point should be 71.5 ns) . Kindly clarify.....

Thanks

Anuradha

Yes, the tool will find the normal delay, at the boundary that gives the largest setup time. then, the tool will simulate at the opposite boundary, which is supposed to give a more than 10% delay degradation. after this, the tool will reduce the setup time between the signal to find the nearest (no bigger that 2ps tolerance) point where 10% delay degradation is found. with 65ns normal delay, delay degradation by 10%, which means a delay of 71.5, at which the setup is measured.

Guangjun

Thanks. The definition of the criteria is clear to me now although it's bit foggy in the manual. You can consider this as "closed" now.

Anuradha