bbspice fitting process taking too much time

Dear Andrew and Tawna,

We have just migrated to Spectre 19.1.0.541.isr14 64bit, IC6.1.8-64b.500.14, EMX 6.0 with Interface 6.0.0.  For Spectre 20.1, we are still requesting a license to use. 

I have re-simulated my layout structure from DC-900GHz with this time a 100MHz step. The data looks smooth to me in ViVA (I could show the screenshot if you are interested in it). To my surprise, the data is non-passive as there are peaks at around 755GHz where the magnitudes are greater than 1. You could see that clearly also in the impedance graph. That is why in transient simulation, the bbspice method produces large fitting error and Spectre instead uses the linear interpolation method (I have here set nportbbsversion=2). This leads to a  "blow-up" transient behavior in my circuit, in which the node voltage reaches 100kV level.   

I will contact customer support as soon as possible. In the meantime, I want to ask if you have some general recommendations about why EMX could generate nonpassive data, and how one should do to avoid it.

Many thanks and regards,

Anh

Dear Andrew and Tawna,

We have just migrated to Spectre 19.1.0.541.isr14 64bit, IC6.1.8-64b.500.14, EMX 6.0 with Interface 6.0.0.  For Spectre 20.1, we are still requesting a license to use. 

I have re-simulated my layout structure from DC-900GHz with this time a 100MHz step. The data looks smooth to me in ViVA (I could show the screenshot if you are interested in it). To my surprise, the data is non-passive as there are peaks at around 755GHz where the magnitudes are greater than 1. You could see that clearly also in the impedance graph. That is why in transient simulation, the bbspice method produces large fitting error and Spectre instead uses the linear interpolation method (I have here set nportbbsversion=2). This leads to a  "blow-up" transient behavior in my circuit, in which the node voltage reaches 100kV level.   

I will contact customer support as soon as possible. In the meantime, I want to ask if you have some general recommendations about why EMX could generate nonpassive data, and how one should do to avoid it.

Many thanks and regards,

Anh

Hi Anh,

Please file a Case at https://support.cadence.com .  When you file the Case, let me know the Case number.   

I'm thinking out loud here ....

For generating an s-parameter file that will work with Spectre, see this appNote:  7 Habits of Highly Successful S-Parameters (Spectre 20.1, 19.1 and IC6.1.8 ISR9) .   

• You have a dc data point in EMX - this is good. 
• I plot the s-parameters on the Smith Chart using ViVA:  How do you plot S-Parameter data directly from ViVA? .  I see that you've done this - great!   
• Make sure your data is spaced closely enough to accurately capture the DUT behavior.   If the data looks disjointed or choppy/piecewise-linear, then there probably aren't enough data points - Your data looks pretty smooth on the Smith Chart, at least for s-parameter S59,27.
• You can use the s-parameter validation tool in ADE Explorer to check for passivity, as outlined in:   S-Parameter Comparison Tool Aids in Troubleshooting.   I expect s-parameter data from EMX to be passive for a truly passive DUT.  

I can see your concerns regarding passivity by looking at your Smith Chart plot.   Is the DUT truly passive? (DUT=device under test)

You *cannot* use bbspice interpolation when simulating an active device.  You must use linear interpolation.  

There are a number of parts to check:

1. Is the DUT as designed truly passive?   And what are you modeling with the s-parameter data (what is the DUT)?

2. Is the EMX setup correct?   

3. Is the resulting s-parameter data of good quality?  

4. Is the spectre setup correct?

5. When you run your spectre simulation, are you getting any warnings or errors (I'm sure the answer is "yes")?   The spectre.out logfile will give excellent troubleshooting information that we'll need to know.

best regards,

Tawna