HSPICE on PC and SPARC - different results

[paulc]

ultrasim waveform awaves

Anyone tried to run simulation on PC and SPARC? I tried and find that the results are different for switching capacitor circuit. Anyone has same experience?

Thanks

 

[golohoyeah]

I have simulated in PC before, and find the
result not so good...
I think SPARC has more accurate result

 

[dacadc]

I think main reason lies in bad mathematical models. Please check used models.

 

[paulc]

I am using BSIM3.3 Model. I think the models are OK since I get same result for some other circuit in AC and transient analysis. But, for switching capacitor circuit, I get different results.

 

[cswang]

please check your option seting
and transient step and transient periors!

 

[paulc]

Both files are the same, model, simulation card, circuit netlist. Everything is the same, in the same directory

 

[andy2000a]

On 2002-03-28 12:56, golohoyeah wrote:
I have simulated in PC before, and find the
result not so good...
I think SPARC has more accurate result

hspice on X86 CPU use 2GHz floating > 700M UltraSparc , and the same CPU clock winXP OS is faster than win98 /win2k
by the way , have anyone use evercad aditspice ?? (it support verilog + spice)

 

[tlihu]

On 2002-04-01 05:05, andy2000a wrote:

On 2002-03-28 12:56, golohoyeah wrote:
I have simulated in PC before, and find the
result not so good...
I think SPARC has more accurate result

hspice on X86 CPU use 2GHz floating > 700M UltraSparc , and the same CPU clock winXP OS is faster than win98 /win2k
by the way , have anyone use evercad aditspice ?? (it support verilog + spice)

I've tried ADiT for a couple of years and found that it did run simulation faster than HSpice in the native SPICE mode, although the slowest among the modes ADiT supports (SPICE, Turbo, & Xgate for verilog/SPICE co-simulation). However, I don't think that the product is mature enough because, except for SPICE mode, there are a few bugs, or incompatibilities, in the other two modes.

 

[paulc]

Today, i even found that the results on P4 and Athlon are different! Both simulations are on Win 2K and model, netlist, simulation card are the same.

I use HSPICE because of its proved accuracy and good convergence. In fact, I tried HSIM before, it is several times faster than HSPICE, but for some circuit like switching capacitor circuit, the results are obviously wrong.

 

[Nobody]

Avanti local agent provide different prices in PC and UNIX version.
PC version's list price is about US$8000 and UNIX is double of PC's.
And the sales tell me that some limit in PC's version.
BTW, u may upload the lib and encrypted hspice in the forum to let
others help u to test the case. Maybe the problem is only in ur enviroment.

 

[paulc]

Nobody, sorry, i cannot upload the model and spice file.

 

[MOS21]

hsim

Today, i even found that the results on P4 and Athlon are different! Both simulations are on Win 2K and model, netlist, simulation card are the same.

I use HSPICE because of its proved accuracy and good convergence. In fact, I tried HSIM before, it is several times faster than HSPICE, but for some circuit like switching capacitor circuit, the results are obviously wrong.[/quote

hsim will give good results, if the right tolerances are set.
try seting the analog (most sensitive portion in a single matrix)
dont rely on the outo partitioner, that is bad.]

 

[andy2000a]

Anyone hear about ultrasim ??

I still use hspice for analog design , and if some "digital like" circuit , I use
aditSpice or starSim or hsim speed it ..
synopsys nanosim = powermill +VCS , but I find other tool , like smash
ultrasim .. anyway , hspice is Golden simulator for analog design...
but hspice have many little bug..

 

[CatKing]

I don't think the sim result will be different on different platform. The speed may be different, but the result must be same, otherwise, forget this tool and try some other one.

 

[edaguy69]

The different between PC version and Unix versions (Sun, IBM, HP or SGI) is PC version only support less models than the Unix versions. And believe it or not in PC it's usually faster ! We pay more money to get the result slower ! And yes, the result are different in different version because the default setting are different.

 

[andy2000a]

about PC version spice simulator

recentlly , I try use smash for win2k (support Hspice L49 model) and try to run it, compare it with hspice 2002.4 (Unix) , till now look like well,
but I use sweep result (smash use paramsweep) , smash waveform viewer is "poor" than awaves for display "sweep result"
Have anyone "special case" for "verify" smash spice engine ?? aditSpice run is fast , But some time simulation result is wrong (like Hsim /powermill in some case have "wrong" result )

 

[edaguy69]

The accuracy of the result depends on many factors: initial condition you set and the value of some options of the tool. The accuracy of the waveform depends on how many timestep you use. If you don't use enough point, the waveform viewer will intrapolate to give you the result in between. Hspiceis a different breed with Hsim, PowerMill
Well, Hspice is accuracy but very slow, it take weeks or months (!) to simulate a PLL circuit. Hsim or PowerMill/Time takes only a few hours. There are always trade off between speed and accuracy. As a matter of fact, with some appropriate options, Hsim can give you a very near result of Hspice. Have you ever tried to simulate a circuit of 100,000 transistors with Hspice? Hsim can easily !

 

[andy2000a]

about hspice & hsim

I know hsim is "fast" than hspice , I ever use hsim/adit/starSimXt/ hspice, and we usually use strsim/adit for whole chip simulation ,
but some tricky in use hsim/adit analog block , I ever use hsim simulation
a PLL case (no initial condition) hspice will fail for oscillator , but hsim/adit
it work (really chip is sometime fail) , and I think hsim/starsim or aditSpice is usefully for our design, but how to setting hsim/adit/Starsim ?
hsim/starsim maybe suit for SRAM/ROM design , but some circuit , I think hspice is better simulator ...in my case I think
SRAM/ROM , Analog-digital block interface --> use hsim/powermill like
PLL --> use hspice
I hope some EDA mix simulator can do it , analog call spice, interface call hsim , digital call Verilog ..
smash support spice/verilog, but lack of interface (or digital circuit in analog block)

 

[edaguy69]

It's just a wish, different jobs require diferent tools. Our problems are their business opportunity. There is no all-cure tool. we need both Hspice and Hsimto do the job right. Have you heard about Hspice RF you have to pay double to get that feature that Spectre and other tools had for a long time.

 

[sunjimmy]

I have used Hspice for many years.
Although Hspice is slow. But if you know
how to set the options correctly, it will produce reasonable & good result. From my experience, the simulation result is very close to the real measurement.
In my opinion, for switched-cap ckt , you must pay attention to the options about how charge conservation is preserved during simulation. Otherwise, the accumulated error charge may lead you to the wrong answer.

Hope it helps : )