Passing to report something very exciting: using the openEMS simulation scripts written by Volker Muelhaus given on the IHP SG13G2 github repo, I ran some simulations on a 50-ohm microstrip line using the sky130 BEOL. And I did some benchmarking comparing the results against Ansys HFSS using the same geometry and BEOL stack. The strip width (M5) is 6 um and the ground width (M1) is 30 um. As basis of comparison I used some performance parameters: the characteristic impedance (Zc), quality factor (Q), attenuation constant (alpha) and propagation constant (beta). First, let's begin by comparing the Zc. Since openEMS doesn't have an automatic meshing option, it's on the designer to find an optimal mesh, but as hard as this task can sound, it is totally doable. So, the comparison will take into account some different meshing density. First figure shows the comparison of 6 different meshing densities: using 3, 4, 8, 10 and 12 lateral cells (using a vertical cell size equal to the skin depth at 60 GHz, 0.5 um) on the signal strip and another, using 12 lateral cells and a refined vertical cell size equal to the skin depth at 110 GHz, 0.25 um. We can see that using more than 10 lateral cells doesn't improve the result, but refining the vertical cells increases Zc. HFSS simulation results in higher Zc (50.2 ohms for the complete stack, 50.6 ohms for a simplified, planarized version). OpenEMS gives 48.6 ohms (vertical cell = 0.5 um) and 49.2 ohms (vertical cell = 0.25 um). Respectively a 3.18% and -1.99% error. Really, really close! The second figure shows the quality factor. HFSS results a 7.6 Q-factor at 60 GHz, while the two most accurate settings on openEMS gives 7 and 7.2. A caveat is that HFSS on this simulation setting does not have the "solve inside" setting activated on the thick metal, thus the conductive losses might be a mit under-estimated. But the values (albeit low) are very close. The third figure shows the attenuation constant. Same tendency of the Q-factor: HFSS results in lower losses than openEMS, but the mesh on openEMS is implementing a "solve inside" on M5 signal, thus I'd say openEMS is more accurate on this front. 1.36 dB/mm for HFSS, 1.5 dB/mm and 1.47 dB/mm on openEMS, respectively for coarser and thinner vertical meshing. Again, close (+10.3 % and +8 % errors). All values at 60 GHz. Last but not the least, the fourth figure shows the propagation constant. 1.397 deg/um and 1.4 deg/um for openEMS and 1.39 deg/um for HFSS. I'd say they are the same because every simulator incurs in numerical error, and small variations are expected. All values at 60 GHz. I will put together a git repo this week to make the openEMS script available to whomever wants, but I have to say that I'm really excited about these results! Now to attack CPW lines ;)