HSPICE on PC and SPARC - different results

[rfsystem]

Forget about the simulator if your result seems questionable. If the integration algorithm and error control is equal the difference is only in the model or setup.

The simulator defines only the way it converts a spice netlist into a waveform result shown on display.

I have used Spectre, Hspice, Hsim, Smash, Eldo, Titan and Pspice in the past.

I was looking for differencies many times. In the end it was the user, setup or model which give different results.

 

[jiang]

how about linux version?
is hspice on linux still less models than other unix version?

 

[okguy]

Hspice doc seem to be the same for PC and UNIX :
which models are not supported on PC???

For HSIM, level 28 is a problem, as it is an avanti proprietary model.

OkGuy?

 

[CatKing]

I have never heard some level's model was not supported by the latest release of HSPICE, but there's difference of win and sol versions:
1. There's netlist size limitation of HSPICE on windows version(<= 1M netlist file, can not remember exactly), But no limitation on solaris version.
2. Tools interface: sol version have good I/F with cadence tools, win version dosen't have it(the reason is obviously)/
That's all i have found between them.

 

[andy2000a]

about hspice on PC & UNIX

I think hspice limit on PC is not Tx size , for my design , hspice run over 10000 mos + some Res , Cap will very slow .... if you said PC's hspice limit Tx size to 1M , I think it is too "huge" in UNIX & PC ..
If whloe chip simulation , we use hsim or powerMill ... not use hspice ..

by the way , have anyone use smash runco-sim ?? I use nanosim for whole chip .. but still some error message ..

but Smash waveform tool is too bad , I hope it can use "nova nWave"

 

[RC13]

the pc and unix uses differenct type of processor, especially the memory for calculation. normally, the unix system uses error-correction type. hence, for high accuracy calculation (a lot of decima point!), the software can only run on unix.

for the case of the circuit simulation, i don't the different can be significant. the software setting might be different in both.

 

[Hughes]

The accuracy and speed depend on the software and hardware environment. The solutions to the equations depend on the word length, which is different in different platform, so come the different results. The SPARC version will give better accuracy.

 

[wpwang]

We try hspice on pc and unix. We used the same design.
The result is a little bit difference. We get PC is faster than UNix.
We can't dicesion what is more accuracy.

Pc : PIII 600 RAM 128 Win2000
Unix : Brad1000 RAM 1G SUNOS 8.0(64 bit)

 

[andy2000a]

Re: HSPICE on PC and SPARC

Hi wpwang
Do you feedback this issue to Synopsys ?? maybe Hspice on PC have some problem on "source Code" , In general , EDA vendor already test many case on Hspice for PC .. If your case is specify .. maybe hspice have bug , I try some case both hspice on Unix /PC is ok ..

another problem is CPU issue , like early X86 have FPU bug ..
can you try your case on other X86 CPU ? like P4 or K7

If you setting simulation condition is the same case , only platform difference, I think Hspice develop team maybe check hspice ..

some EDA tool sometime will be find a bug by RD's , maybe you find a new bug ...
on hspice ...

 

[gerryhsu]

Re: HSPICE on PC and SPARC

Hi wpwang
If you setting simulation condition is the same case , only platform difference, I think Hspice develop team maybe check hspice ..

...

I heard that the whole team of Hspice is gone ! tough luck

 

[kunjalan]

Re: HSPICE on PC and SPARC

Please use HSPICE2003 PC veersion. May be it has more accuracies compare with old version.
But the differece between PC and UNIX machine is still exist.
Maybe the reason of different accuracy comes from CPU floating point calculation accuracy.

 

[Btrend]

Re: HSPICE on PC and SPARC

I think there are always difference between application of EDA tools, but
the final results should make sense or have the same response to your simulation of behavior models. That's your goal of your design. The transistor level simulation can affect speed, gain, accuracy but the
functionality should be the same with the behavior model.
So, analog designer always need "options" or "programmable ability" in their circuits to compensate the errors from device models, EDA tools, and FAB uncertainties.

 

[edacw1]

HSPICE on PC and SPARC

Do you use the same version to run simulation on PC and SPARC? I think the SPARC has the higher accurate grade than PC's.