https://open-source-silicon.dev logo
#analog-design
Title
# analog-design
s

Stefan Schippers

01/19/2021, 1:58 PM
@User ideal inductors are not in the sky130 lib as these are not process specific, these are in the xschem standard devices/ library, like ideal resistors, capacitors etc. Realistic inductor can be made using metal loops/spirals, but i don't have any device models/subcircuits for these. If anyone has some information on how to models these structures i can add sky130-specific inductor components.
๐ŸŒ 1
r

Rafael Marinho

01/19/2021, 2:33 PM
I't is a very good idea! Though not a easy task. The model is relatively simple, but need information of sky specific process. Not sure if it can be easily obtained. Here a model could be used. Values of the lumped-elements are tech related and dependent of the inductor itself (no of spirals, width, and distance between spirals, inductor shape, distance to the substract). I'm not specialist, but I'd kind help with it if someone is interested. It is a good chance to learn inductance modeling.
๐ŸŒ 1
s

Stefan Schippers

01/19/2021, 3:14 PM
yes, i would say Rsub can be neglected, Rs and Csub can be obtained from process parasitic parameters, I don't know about Cs and Ls... These depend on the specific layout (spiral / loop(s), single/multiple layers,...)
@Rafael Marinho what i can do is create a symbol for this inductor with the above subcircuit, with parameters for Ls, Rs, Cs, Csub; the designed needs then to fill these values, depending on the metal layer/width/length/shape being used. May be @Tim Edwards can comment if it is reasonable to assume Rsub=โˆž ...
๐ŸŒ 1
t

Tim Edwards

01/19/2021, 4:18 PM
@Stefan Schippers: One of the designs on the shuttle run has a set of inductors that will be fully characterized. So by all means, put in some placeholders, and we can fill in the characterization results when they are available.
๐ŸŒ 1
I can't really answer the Rsub question other than that infinite is probably a reasonable first-order approximation.
s

Stefan Schippers

01/19/2021, 8:26 PM
@Tim Edwards Great! will create the component, then all the details can be updated later on. Thanks
I have just learned that in integrated inductor models Rsub resistors are not accounting for (probably inexistent) conduction in the intermetal oxide, but for losses due to induced currents in the substrate. A patterned substrate with lower metal lines orthogonal to the spiral directions could probably yield higher Q inductors. This is a nice presentation. http://www-smirc.stanford.edu/papers/Orals98s-cpyue.pdf
๐ŸŒ 1
c

Chris Jones

01/23/2021, 1:40 PM
As well as the magnetically induced currents in the substrate, any capacitance to the substrate will have resistance in series with it, so thst would be a resistor in series with the Csub component in the model shown above. As Stefan points out, this resistive loss can be reduced with a patterned metal1 shield (at the expense of somewhat increasing the value of the parasitic capacitances, but it is worthwhile). I think that it is probably not worth guessing the form of the equivalent circuit until you have the model, because that will depend on the modelling tools used as well as the number of terminals on the inductor ( if it has centre taps, or other taps and maybe additional windings). I generally use two models, a complicated one with hundreds of components in it that my field solvers give me as a netlist, and also a simple one with about 5 or 6 elements, that I draw a schematic for, and fit to the complicated model at the frequency of interest.
๐Ÿ‘ 2
๐ŸŒ 1
r

Rafael Marinho

02/03/2021, 1:33 PM
Hi @Chris Jones, would you share more information about the inductor design flux? I was wondering to try open-source tools (ngsolve, FeniCS, freefem...) to simulate the inductor model, but since I never try such flux I do know if the effort worth. In the past I have play with MoM simulation for inductors with Keysight ADS. Then I could use a s-param file (generated at ADS) inside Cadence tools, this way I could simulate the extracted version of the circuit with the extracted circuit in cadence + the inductor model (EM extracted). I wonder if I could make something similar, but with open-source tools...
๐ŸŒ 1
c

Chris Jones

02/04/2021, 11:25 AM
What I have used in the past is FastHenry and FastCap, with a script to generate the input geometry for each of these, with a separate conductor for each segment of each turn. I used FastModel from fastfieldsolvers.com to view the 3d geometry to check that my script generated it properly. Then the script makes a subcircuit model from the output of FastHenry and FastCap. I can't use the script that I had at my previous job, however I could write something equivalent again if I have time. Also there might be better open source tools these days.
๐ŸŒ 1
t

Tim Edwards

02/09/2021, 4:59 PM
@Chris Jones (@Rafael Marinho): There is a variant of the "extresist" command in magic that does full R-C extraction that I added a long time ago when I was working for MultiGiG and we were using FastHenry to analyze transmission line geometry. "extresist fasthenry" will extract output for FastHenry. I have not used it in years. Also, I did not make any equivalent routine for FastCap. But that's one open-source solution.
๐ŸŒ 1
c

Chris Jones

02/13/2021, 12:34 AM
Thanks @Tim Edwards, that sounds like it could be helpful for a more general solution than my script that only models specific structures.
๐ŸŒ 1