Which simulator is more accurate? HSPICE or Spectre(RF)

[wylee]

spectre conservative moderate

Me and my colleagues always arguing on which simulator produce the most accurate results? HSPICE, Spectre or SpectreRF?

Please vote your opinion and if possible give your reason(s)...

 

[rajath]

hspice or spectre

I thought accuracy was more dependent on the model used, and not that much on the type of simulator. But as far as speed vs. accuracy is concerned, I think spectre performs a tad better.

 

[eda4you]

spectre circuit simulator training malaysia

I also agree with rajath. Further there are a lot of paramters one can turn during simulation setup (incredible). But of course, the accuracy of the models dominate the simulation results.

 

[wylee]

hspice accurate=1

Let us just assumed that we are using the same model from the same foundry for both HSPICE and Spectre, say TSMC 0.18um models...

My personal experience is using the same circuit, simulating under the same condition, HSPICE shows system instability (output oscillates) whereas Spectre don't.....

Thats why I 'think' HSPICE simulations is more accurate

 

[Fom]

hspice option accurate

Hy, wylee!
Did you try conservative option for Spectre. May be you get difference because of different simulator options.

 

[wylee]

spectre hspice

As far as I know, 'conservative' simulation in Spectre will show better resolution on the waveform where more sample points are calculated to plot a smoother waveform.

Fom, I had forgotten and cannot confirm whether I used 'consevative' or 'moderate' settings when I compared results with HSPICE.

Anyway, if both 'conservative' and 'moderate' simulations in Spectre show different results (such as oscillations), surely there is some problem in their simulator

In HSPICE, I always use ".option accurate=1" in all my netlist

 

[terryssw]

Conversative or moderate setting does really affect the simulation accuracy, also there is some occasion that moderate setting provide completly different waveform than the conversative one due to different time-step control, different numerical integration methods and relative tolerence setting. For example, when simulating digital signals, oscillation (which is come from the failure of the simulation) can occur during transition (This is known as trapezodial oscillation as it is caused by the failure of trapezodial numerical integration methods) if the time step is too large. Usually conversative setting can solve the problem as it used gear2 integration methods. I think Hspice also have such problems, but in Hspice the option setting are usually well defined by the user easily, so it work well. Hope this can help and correct me if I am wrong.

 

[eda4you]

I'd like to discuss in this context he following problem.

Some years ago I tried to model switches and its nonlinear behaviour (charge injection, ...) and got some strange simulation results for it. For example if the input signal of a normal nmos switch changes its polarity (the source gets the drain and similar) some charges get loss!!! And that for hspice and spectre, although I used Bsim3v3 modells, which are based on modelling charges in contrast to older modells. A device engineer told me that 's normal, because the models aren't symetric!?! Besides the mos was a simple one, not one designed for switching higher voltages etc.