Hi, I am comparing the CCS characterization accuracy of cells in a shift register with its spice simulation. The power characterization settings used in Liberate looks like below:
#---------------------------------------------------- Accuracy -------------------------------------------
set_var extsim_cmd_option "+spice +lorder MMSIM:PRODUCT"
set_var extsim_deck_header "simulator \lang=spectre\nOpt1 options reltol=1e-4\nsimulator \lang=spice"
set_var extsim_leakage_option "method=gear gmin=1e-15 redefinedparams=ignore rabsshort=1m"
set_var extsim_option "method=gear gmin=1e-15 redefinedparams=ignore rabsshort=1m"
set_var extsim_tran_append "lteratio=10"
set_var extsim_reuse_ic 3
#----------------------------------------------- Power ---------------------------------------------------
### Leakage ###
set_var max_leakage_vector [expr 2**10]
set_var leakage_float_internal_supply 0 ;# get worst case leakage for power switch cells when off
set_var reset_negative_leakage_power 1 ;# convert negative leakage current to 0
### Power ###
set_var voltage_map 1 ;# create pg_pin groups, related_power_pin / related_ground_pin
set_var pin_based_power 0 ;# 0=based on VDD only; 1=power based on VDD and VSS (default);
set_var power_combinational_include_output 0 ;# do not include output pins in when conditions for combinational cells
set_var force_default_group 1
set_default_group -criteria {power avg} ;# use average for default power group
#set_var power_subtract_leakage 4 ;# use 4 for cells with exhaustive leakage states.
set_var subtract_hidden_power 2 ;# 2=subtract hidden power for all cells
set_var subtract_hidden_power_use_default 3 ;# 2=subtract hidden power from matched when condition then default group
set_var power_multi_output_binning_mode 1 ;# binning for multi-output cell considered for both timing and power arcs
set_var power_minimize_switching 1
set_var max_hidden_vector [expr 2**10]
#-------------------------------------------------------------------------------------------------------------------------
Based on the observations, I have few questions to get clarified.....
1. In CCS white paper, the char accuracy was said to be ~2% with CCS modeling, however with these specific settings I cannot reach that accuracy level. In my case, the power accuracy lies within 6-10% of spectre spice simulations. Is there any extra option/switch I can use to improve the accuracy of the simulations (I use auto indexing mode with min/max transitions defined in the ttemplate) ?
2. It's recommended to use the analytical PWL pre-driver waveform for the best accuracy. However it is not very clear, how this is calculated. Is it based on some empirical values or based on template/model information ?
Thanks in advance