Based on coplanar waveguide, I am designing a meta-material waveguide by inserting lumped-element inductors along the center line.

Here are some parameters:

  • Target frequency: 10-20 GHz
  • Size of lump capacitor: 10 um, sit along the CPW (insert into a cut along the centerline)
  • CPW dimension: CL width = 20 um, CL gap = 11.8 um (nominally 50 Ohm TL)

The lumped-element inductors are separated by 20 um. To achieve 50 Ohm matching, between the inductors I inserted a pair of open stub, extending from the center line into the GND without touching. Together with the inductor, I define a unit cell using T-model as "LCL", where L is half the inductance of each inductor from the front and rear of a cell.

A similar figure can be found in Fig. 1 in this work: https://arxiv.org/pdf/2007.00638.pdf

In COMSOL, I simulated a short array of 20 such cells terminated by 50 Ohm lumped ports, where I swept the stub length as a parameters: 50Ohm_LumpedPort_1 --- (LCL) --- (LCL) --- ... --- (LCL) --- (LCL) --- 50Ohm_LumpedPort_2

With the above 20-cell simulation, I found that I needed a stub length of 100 um (each side) to obtain sufficient shunt capacitance to get good 50 Ohm matching, characterized by a minimum peak reflection over a wide band of freq (8-12 GHz)

Then I expanded the simulation from 20 cells to 200 cells. The models are similar except that now we have a longer chip as the array is straight. I also kept the lumped-element inductor the same. To my surprise, now the stub length that gives 50 Ohm matched condition is significantly shorter at 80 um.

To analyse the issue, I fitted the 20-cell model in Python to an expected transfer matrix to extract the L and C values of the cell at a given stub length. Using the obtained L and C, I formed a 200-cell array and analyse the Bloch impedance, I do not see a big difference from that of a 20-cell array, and they are both very close to Z_0 = sqrt(L/C) expected from a single cell.

Much appreciated if you can advise on it.


1 Answer 1


At the end, I found that it was due to a wrong meshing. By ensuring the same mesh density in the two models, I can see the same reflection and extract the same equivalent lumped-element values of L and C.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.