how many segments of the RLGC model is accurate?
if i know the RLGC parameters in terms of Ω/m, H/m, F/m. How many segments (or basically what's the length of each segment) should I set for accurate transmission line model simulation?
Any comments about that?
Thanks,
You need at least 10 segments (better 20 segements) per wavelength. Notice that wavelength here refers to the effective wavelength in the line, not the wavelength in air. So each segment should be no longer than 360°/10 (better 360°/20) in transmission phase delay.
I doubt that RLGC model gives "accurate transmission line model simulation", because it can't model frequency dependent losses.
Yes and no.
Agreed, the normal format is exact only for narrow bandwidth. However, some simulation software like Cadence Spectre provides an extension where multiple RLCG parameter sets (extracted at different frequencies) are all combined in one file, and the software knows how to handle these frequency dependent tables. Some EM solvers like Sonnet can create these frequency-dependent RLCG data files.
So I can just estimate the effective wavelength from the transmission signal, e.g. 10Gb/s data rate, just estimate as 10GHz, i don't know what's the microwave index on-chip, so just use speed of the light, then the wavelength of the signal is 3 cm. 1/10 wavelength is 3mm, if my on-chip transmission line is 3mm long. I think i have to take another 1/10 based on the transmission line length, I have to use 300um (or 150um) per segment length. is that correct? what if the transmission line is even longer?
do you know any paper or book talked about this?
Thanks
Yes, you are right, RLGC is frequency dependent, but it I don't know how to incorporate freq. variables into the Verilog-A model to capture this effect.
You can choose as many segments as you need for harmonics and analyze all these frequencies.
This applet does 159 harmonics
If you chose 159 segments, you can compare your results with this Fourier java Applet
http://www.falstad.com/fourier/
Analysis methods
http://www.jtl-engineering.nl/signal..._analysis.html
As mentioned by SunnySkyGuy, for non-linar simulation you should consider not only the fundamental frequency but also the harmonics. For your 10GHz signal let's include up to the 5th harmonic, i.e. 50GHz.
With SiO2 as intermetal dielectric, we have permittivity around 4, so that on-chip wavelength is about half of the wavelength in air.
50GHz wavelength in air is 6mm, so on-chip it is about 3mm. Then with the lambda/10 estimate, each line segment should not be longer than 0.3mm (at 1/10 wavelength). To be safe, you could use half of that value for the segment length.
Build the model by cascading short segments of the length calculated above.
No, not that I am aware of. When I worked on this topic as an EDA expert in EM support, I wrote some tool-related application notes on this topic, but these are too specific for your case.