Clarity 3D - Dielectric materials model selection

Hi Romain, 

What material model is most applicable for a given fabric and dielectric doesn't come with a simple answer, but let me share with you something that hopefully will be helpful.

The material model most applicable to a given dielectric depends on the bandwidth of interest and the material relaxations dominating that band. The Debye model (dating back to 1929) was defined to model ideal dipolar materials with a single relaxation. The Debye single pole model is not very suitable for a lot of the materials we work with in the PCB/packaging industry and we could instead either use the multi-pole Debye and infinite pole Debye model, aka the Djordjevic Sarkar model. Have a look at this paper which has a lot of details on the modeling as well as some very interesting references for further studies. I encourage you to make some experiments and correlation on your own to get a better grasp of acceptable assumptions. We also have a presentation on the 16th June as part of an SIJournal event, that discusses correlation for a pcb to 50GHz and the material modeling used.

For semiconductor materials, I can't pretend to know much so would hope somebody else could chime in. From speaking with colleagues working in this domain, it seems that Silicon is typically modeled with a constant permittivity simply because of lack of information in the PDKs, but the good news is that judging from the number of successful chips made to date, it is probably a factor that doesn't play a significant role in todays designs. If you want to have a further read, this paper, elaborates further on frequency dependency of both Si, GaA and other materials.  

From a practical standpoint, I can say that we typically see people modeling silicon material with a constant permittivity value and loss tangent of 0 and then specifying the bulk conductivity. For both Djordjevic Sarkar and Debye models, setting the loss tangent to 0 will force permittivity to stay constant over frequency. In other words, even if you specify in Clarity that you want to use e.g. Djordjevic Sarkar to capture the frequency dependent behavior of the dielectrics on package/pcb, you can still model the silicon as frequency independent by setting the loss tangent to 0.

Thanks,

Kristoffer

Hi Kristoffer,

Thanks for your detailed response.

For the presentation on 16^th June, I already registered 

For materials related to the PCB and the package, I probably have less concern about the use of either a Debye model or a D-Sarkar model. However, I am not comfortable with a loss tangent equals to 0.

Coming to the semiconductor, in particular a silicon bulk substrate, I agree with you to set the loss tangent to 0, since the losses will come from the conductivity of the silicon. I think you can use only one of these parameters: conductivity or loss tangent. From my past experience including silicon substrate in my designs, I always used constant values over large frequency ranges without any issue.

In Clarity, when combining multiple domains (edit-in-concert in Virtuoso RF) like the IC within its package and onto the PCB, there is only one possible way to set all the dielectric and semiconductor materials model with enforced causality.

From the different readings, the use of a Debye model with a very high resonant frequency would make the dielectric characteristics constant over the frequency range of interest. I would be interested in having some insights from the community as well.

About correlation, for sure I will make some trials to evaluate the models available in Clarity (and I will let you know).

Best regards

Romain

Hi Kristoffer,

Thanks for your detailed response.

For the presentation on 16^th June, I already registered 

For materials related to the PCB and the package, I probably have less concern about the use of either a Debye model or a D-Sarkar model. However, I am not comfortable with a loss tangent equals to 0.

Coming to the semiconductor, in particular a silicon bulk substrate, I agree with you to set the loss tangent to 0, since the losses will come from the conductivity of the silicon. I think you can use only one of these parameters: conductivity or loss tangent. From my past experience including silicon substrate in my designs, I always used constant values over large frequency ranges without any issue.

In Clarity, when combining multiple domains (edit-in-concert in Virtuoso RF) like the IC within its package and onto the PCB, there is only one possible way to set all the dielectric and semiconductor materials model with enforced causality.

From the different readings, the use of a Debye model with a very high resonant frequency would make the dielectric characteristics constant over the frequency range of interest. I would be interested in having some insights from the community as well.

About correlation, for sure I will make some trials to evaluate the models available in Clarity (and I will let you know).

Best regards

Romain