Guest
Hi all,
We are working with cadence signalstorm library characterizer and we
encountered 2 issues. Does anyone has a solution/suggestions for these?
1) According to the SignalStorm Library characterizer User guide page
53, bottom of the page:
SignalStorm library characterizer manages the database by checking the
result timestamp. If the simulation passed, SignalStorm does not rerun
the simulation, unless you use the-force option. But if the database
changed as the circuit changed, the specification file changed, or the
simulation setup file changed, it marks the simulation result as old
and reruns the simulation.
I have tried this. But it does not work as it should.
This is what I do and what happens.
db_open db_lib2 //create database
set_var SG_SPICE_SIMPLIFY true
set_var SG_SIM_TYPE "hspice"
set_var SG_SIM_NAME "hspice"
set_var SG_SPICE_SUPPLY1_NAMES " "
set_var SG_SPICE_SUPPLY0_NAMES " "
// install subckt
db_install -design AND2 -subckt Spi/net.lump_AND2 -lib \
../../models/HSPICE/l18u18v.012_CMOS
db_install -design AOI21 -subckt Spi/net.lump_AOI21 -lib \
../../models/HSPICE/l18u18v.012_CMOS
// create vectors
db_gsim
// setup
db_setup -setup setup.nsgu -process L18U18V_TT
// perform simulation
db_spice -process L18U18V_TT -simulator HSPICE
-> this all works and the simulations are running with the ipsc/ipsd
daemons on 1 machine
Then I change the netlist of AND2 (changing transitor's width)
I run the following commands
db_open db_lib2 //open database
set_var SG_SPICE_SIMPLIFY true
set_var SG_SIM_TYPE "hspice"
set_var SG_SIM_NAME "hspice"
set_var SG_SPICE_SUPPLY1_NAMES " "
set_var SG_SPICE_SUPPLY0_NAMES " "
db_install -design AND2 -subckt Spi/net.lump_AND2 -lib \
../../models/HSPICE/l18u18v.012_CMOS
db_install -design AOI21 -subckt Spi/net.lump_AOI21 -lib \
../../models/HSPICE/l18u18v.012_CMOS
db_gsim
db_setup -setup setup.nsgu -process L18U18V_TT
db_spice -process L18U18V_TT -simulator HSPICE
According to the user guide 'db_spice' should have marked the
simulation results of AND2 as old and re-simulate them but it does NOT.
It doesn't simulate again.
How do I solve this? Because as you can imagine, this is a very big
issue.
2) Second problem is concerning the ipsc/ipsd daemons.
I cannot describe the problem otherwise than saying that the
simulations are unstable. Meaning sometimes it works well and sometimes
not.
I had configured 3 machines running HP-UX to be used as ipsd.
When I start the simulation and monitor the status with ipsmon it looks
good but at a certain point the simulation of some vectors on 2 of 3
machines hanged themselves. I do not think the simulation is hanging
because 'hspice' is not visible in the 'top' command, only '32' which
is related to signalstorm I think.
When I monitor the progress with ipsmon always the same vectors are in
the queue, So the vectors are not proceeding. Now I have to kill the
processes.
I also tried this on linux machines and the same problem comes up.
What is causing this instability and how can I solve it.
Using pseudo LSF (only 1 machine) isn't a solution for us because we
believe in the power of multi-processing and we need it to speed up the
characterisation.
Thanks,
Kristof
We are working with cadence signalstorm library characterizer and we
encountered 2 issues. Does anyone has a solution/suggestions for these?
1) According to the SignalStorm Library characterizer User guide page
53, bottom of the page:
SignalStorm library characterizer manages the database by checking the
result timestamp. If the simulation passed, SignalStorm does not rerun
the simulation, unless you use the-force option. But if the database
changed as the circuit changed, the specification file changed, or the
simulation setup file changed, it marks the simulation result as old
and reruns the simulation.
I have tried this. But it does not work as it should.
This is what I do and what happens.
db_open db_lib2 //create database
set_var SG_SPICE_SIMPLIFY true
set_var SG_SIM_TYPE "hspice"
set_var SG_SIM_NAME "hspice"
set_var SG_SPICE_SUPPLY1_NAMES " "
set_var SG_SPICE_SUPPLY0_NAMES " "
// install subckt
db_install -design AND2 -subckt Spi/net.lump_AND2 -lib \
../../models/HSPICE/l18u18v.012_CMOS
db_install -design AOI21 -subckt Spi/net.lump_AOI21 -lib \
../../models/HSPICE/l18u18v.012_CMOS
// create vectors
db_gsim
// setup
db_setup -setup setup.nsgu -process L18U18V_TT
// perform simulation
db_spice -process L18U18V_TT -simulator HSPICE
-> this all works and the simulations are running with the ipsc/ipsd
daemons on 1 machine
Then I change the netlist of AND2 (changing transitor's width)
I run the following commands
db_open db_lib2 //open database
set_var SG_SPICE_SIMPLIFY true
set_var SG_SIM_TYPE "hspice"
set_var SG_SIM_NAME "hspice"
set_var SG_SPICE_SUPPLY1_NAMES " "
set_var SG_SPICE_SUPPLY0_NAMES " "
db_install -design AND2 -subckt Spi/net.lump_AND2 -lib \
../../models/HSPICE/l18u18v.012_CMOS
db_install -design AOI21 -subckt Spi/net.lump_AOI21 -lib \
../../models/HSPICE/l18u18v.012_CMOS
db_gsim
db_setup -setup setup.nsgu -process L18U18V_TT
db_spice -process L18U18V_TT -simulator HSPICE
According to the user guide 'db_spice' should have marked the
simulation results of AND2 as old and re-simulate them but it does NOT.
It doesn't simulate again.
How do I solve this? Because as you can imagine, this is a very big
issue.
2) Second problem is concerning the ipsc/ipsd daemons.
I cannot describe the problem otherwise than saying that the
simulations are unstable. Meaning sometimes it works well and sometimes
not.
I had configured 3 machines running HP-UX to be used as ipsd.
When I start the simulation and monitor the status with ipsmon it looks
good but at a certain point the simulation of some vectors on 2 of 3
machines hanged themselves. I do not think the simulation is hanging
because 'hspice' is not visible in the 'top' command, only '32' which
is related to signalstorm I think.
When I monitor the progress with ipsmon always the same vectors are in
the queue, So the vectors are not proceeding. Now I have to kill the
processes.
I also tried this on linux machines and the same problem comes up.
What is causing this instability and how can I solve it.
Using pseudo LSF (only 1 machine) isn't a solution for us because we
believe in the power of multi-processing and we need it to speed up the
characterisation.
Thanks,
Kristof