Designer’s Guide Community

ethan

Community Member

Offline

Posts: 36

warning messages from spectre when doing PSS
Aug 15^th, 2005, 11:29am

Hi there,

Recently, I frequently met the several repetitive warnings from spectre when I tried to do PSS and PAC simulation. One case is shown in the following. I am wondering whether it will cause inaccurate results and how I can get rid of them. Thank you so much.

I have used "errpreset" with moderate and liberal two cases. the warnig messages are almost same. And maxacfreq=50M; there are two phase non-overlapping clocks in the netlist.

Spectre Log file for the case with errpreset=moderate:

Command line:
spectre -env artist5.0.0 +log ../psf/spectre.out +inter=mpsc \
+mpssession=spectre0_4894_5 -format psfbin -raw ../psf \
input.scs
spectre pid = 5076

Loading /usr/cadence1999/ic5032/tools.sun4v/spectre/lib/cmi/3.0/libinfineon_sh.so
.................. (omitted)
spectre (ver. 5.00.32.070703 -- 07 Jul 2003).

Notice from spectre.
Includes RSA BSAFE(R) Cryptographic or Security Protocol Software from RSA
Security, Inc.

Simulating `input.scs' on anubis at 2:24:33 PM, Mon Aug 15, 2005.

Circuit inventory:
nodes 46
equations 97
bsim3v3 12
capacitor 14
relay 18
resistor 19
vsource 8

Entering remote command mode using MPSC service (spectre, ipi, v0.0,
spectre0_4894_5, ).

****************************************************
Periodic Steady-State Analysis `pss': fund = 200 kHz
****************************************************

================================
`pss': time = (0 s -> 2.0075 ms)
================================
Important parameter values:
start = 0 s, outputstart = 0 s, stop = 2.0075 ms
period = 5 us, steadyratio = 1e-03, step = 2.0075 us
maxstep = 20.075 us, ic = all, skipdc = no
reltol = 1e-03, abstol(V) = 1 uV, abstol(I) = 1 pA
temp = 27, tnom = 27 C, tempeffects = all
errpreset = moderate, method = gear2only, lteratio = 3.5, relref = alllocal, cmin = 0 F, gmin = 1 pS

pss: time = 50.2 us (2.5 %), step = 11.2 ns (558 u%)
.................................................
pss: time = 1.957 ms (97.5 %), step = 237.3 ns (11.8 m%)
Conv norm = 45.8e+03, max dV(net064) = -2.45447 V, took 27.71 s.

Warning from spectre during periodic steady state analysis `pss'.
pss analysis might be slow because at least 2500 time-steps are needed to
meet `maxacfreq' constraints.
======================================
`pss': time = (2.0025 ms -> 2.0075 ms)
======================================
pss: time = 2.003 ms (2.68 %), step = 11.09 ns (222 m%)
..................................................................
pss: time = 2.005 ms (50 %), step = 210 ns (4.2 %)
Zero diagonal found in Jacobian at `Vswitchp' and `Vswitchp'.
Zero diagonal found in Jacobian at `Vswitchp' and `Vswitchp'.

(continued in the next post)

Back to top

IP Logged

ethan

Community Member

Offline

Posts: 36

Re: warning messages from spectre when doing PSS
Reply #1 - Aug 15^th, 2005, 11:29am

(cont'd)

Warning from spectre at time = 2.005 ms during periodic steady state analysis
`pss'.
Convergence difficulties resulted in error requirements being unsatisfied.

pss: time = 2.005 ms (52.6 %), step = 12.9 ns (258 m%)
..................................................................
pss: time = 2.007 ms (100 %), step = 171.8 ns (3.44 %)
Conv norm = 417e-03, max dV(I85.M9:int_d) = -18.3628 uV, took 110 ms.
======================================
`pss': time = (2.0025 ms -> 2.0075 ms)
======================================
pss: time = 2.003 ms (2.5 %), step = 1 ns (20 m%)
..............................................................
pss: time = 2.004 ms (37.5 %), step = 1 ns (20 m%)

Warning from spectre at time = 2.0044 ms during periodic steady state analysis
`pss'.
Minimum time step used. Solution might be in error.
Warning from spectre at time = 2.0044 ms during periodic steady state analysis
`pss'.
Minimum time step used. Solution might be in error.
Further occurrences of this warning will be suppressed.

Zero diagonal found in Jacobian at `Vswitchp' and `Vswitchp'.

Warning from spectre at time = 2.0044 ms during periodic steady state analysis
`pss'.
Convergence difficulties resulted in error requirements being unsatisfied.

pss: time = 2.005 ms (42.5 %), step = 1 ns (20 m%)
pss: time = 2.005 ms (47.5 %), step = 1 ns (20 m%)

Warning from spectre at time = 2.005 ms during periodic steady state analysis
`pss'.
Convergence difficulties resulted in error requirements being unsatisfied.

pss: time = 2.005 ms (52.5 %), step = 1 ns (20 m%)
.................................................................
pss: time = 2.007 ms (97.5 %), step = 1 ns (20 m%)
Conv norm = 1.92, max dV(I85.M9:int_d) = -84.448 uV, took 1.59 s.

Warning from spectre during periodic steady state analysis `pss'.
The high-order finite difference method PSS simulation failed to converge.
You can turn off the method by setting 'highorder' to no in the PSS
options. Outputing results without high-order finite difference
refine.

Zero diagonal found in Jacobian at `Vswitchn' and `Vswitchn'.

Warning from spectre at time = 2.0025 ms during periodic steady state analysis
`pss'.
Convergence difficulties resulted in error requirements being unsatisfied.

pss: time = 2.003 ms (2.5 %), step = 1 ns (20 m%)
.........................................................................
pss: time = 2.007 ms (97.5 %), step = 1 ns (20 m%)
pss: The steady-state solution was achieved in 3 iterations.
Number of accepted pss steps = 5146.
Total time required for pss analysis `pss' was 31.91 s.

Warning from spectre.
1 warning suppressed.

modelParameter: writing model parameter values to rawfile.
element: writing instance parameter values to rawfile.
outputParameter: writing output parameter values to rawfile.
designParamVals: writing netlist parameters to rawfile.

End

Back to top

IP Logged

Andrew Beckett

Senior Fellow

Offline

Life, don't talk to
me about Life...

Posts: 1742
Bracknell, UK

Re: warning messages from spectre when doing PSS
Reply #2 - Aug 15^th, 2005, 12:27pm

If your clock frequency is 200kHz, having maxacfreq of 50MHz seems rather high - this is forcing short timesteps in the simulator. Why is it this high?

However, the simulator is trying to take even shorter timesteps than this, which suggests some sort of timestep convergence problem. Do you have large ideal components (like very large inductors or capacitors) in the circuit? Have you connected the circuit up properly? Are some parts of the circuit floating for extended periods of time with no leakage path? This could lead to the singularity problems as you have non-isolated solutions.

I'd advise a later version than IC5032 (which is no longer supported), and for this kind of circuit I'd turn off "highorder" (since it is probably fairly non-linear and wouldn't benefit from highorder). You might want to try (if you've checked the circuit for the issues above) setting cmin=1e-15 (say) to add a small parasitic capacitance to every node, to see if that helps. Missing parasitic capacitances around the transistors combined with discontinuities in the transistor models can lead to transient convergence issues.

That said, it's hard to debug without the circuit...

Regards,

Andrew.

Back to top

IP Logged

byang Community Member Offline Posts: 46	Re: warning messages from spectre when doing PSS Reply #3 - Aug 15^th, 2005, 1:29pm Abrupt state change in vswitch might have caused the zero-diagonal problem seen in the log file. It might help to add a gmin across the vswitch. Baolin
Back to top	IP Logged

ethan

Community Member

Offline

Posts: 36

Re: warning messages from spectre when doing PSS
Reply #4 - Aug 16^th, 2005, 8:08am

Hi Andrew,

Thank you for your prompt reply yesterday.

I have tried your suggestion with cmin=1e-15 and the warning message for convergence difficulties dissappears. But the warning message was still there even by trying maxfreq less than 10MHz alone before.

Actually, there is no specific reason for maxfreq=50M. I know it is high (from the Ken' paper for SC filters simulation method in this website). However, if in case I want to see the opamp (with sc-cmfb) open-loop AC behavior up to 400Mhz in PAC, how much "maxfreq" should I set in PSS in this case? assuming the clock frequency, which is beat frequency, is 200kHz, the number of harmonics=0, and maximum sidebancs=0. I have used maxfreq less than 10M or 20M to 30M before for this case, but each time, the spectre displayed warning message, which was something related to maxfreq too small and would be less accuracy (something like that). So, what's your suggestion on this?

Another question is related to above that BYANG posted "add a gmin across the vswitch". Actually in my upper posts, Vswitchp and Vswitchn are two interconnect wires in schematic, which are all connected to two switches, respectively. My question is that how I can add a specific gmin across the specific wire or trasistor? In some cases, I know I can attach a very big resistor to the certain node. But I don't know how to do that (specfic gmin across specific wire or transistor).

Appreciate your further help.

Good day.

ethan

Back to top

IP Logged