Simulation discrepancies between Sonnet (standalone) and Sonnet Cadence Interface

Hi,

As per title, I am simulating four custom inductors that'll operate at their target inductance at 60 GHz. All simulations are satisfactory when using Sonnet by itself, however when I re-simulate them using Sonnet Cadence Interface (SCI), all of their inductances are way higher then previously simulated (tens of pH), even their minimum inductances.

May I know is this kind of discrepancy normal during re-simulation at SCI? Should I tune the geometry of the design again so that it'll corroborate with the data that I've previously simulated in Sonnet?


Thanks in advance.

No, this is not normal at all. It seems your model is different (same feedline length? feedline de-embedding set properly?) or you evaluate results using different equations.

The feedlines are the same as I imported their gdsII from Sonnet.

not sure about the de-embedding settings and different equations, may I know where to check it?

(edit: the equations I used are inductance1 and q-factor respectively in both platforms.)

also, my .matl file was translated using one of the .son project files, will it affect the results in SCI as well?

In the xgeom editor, you see an arrow if there is a shift in reference plane for the port.
If there is no ref shift, of course any difference in feed line length from box boundary to the inductor will also change inductance.

If the port reference plane is not the difference, compare the *.son file created both ways.

there are arrows in all inductors in xgeom, in fact, all of them are synthesised by your spiral inductor assistant and come with the arrows by default.

Besides, in SCI, I've taken special attention to the box margins so that it will appear the same with the box size (um) in xgeom. Note that my left margin is always set as 0 um to prevent extra feed line from being created automatically.

(both images from different inductors)

Ok, that is to remove the electrical effect of the feedline. The calculated inductance is then for the oval part only (excluding feedlines).

~~

In your model in the Cadence interface, the feedline are part of your results (no reference shift) which explains the extra inductance. That's the additional feedline inductance.

Both results are correct, you just measure at difference places. You need to decide how much feedline length you need in your layout.

So hypothetically, if I remove the feedlines in SCI (only the oval shapes are left), will the simulation results be the same with my previous simulations in Sonnet?


Speaking of which, these inductors ultimately need to be connected with the main LNA circuit, which involves transmission lines. Will the transmission lines affect the performance of the inductors and the LNA altogether? (I'm concerned because the higher the inductances, the lower the operating frequency of my LNA)

Should I include the routing of the transmission lines when designing the inductors in xgeom?

Yes, but you would need to add margings to keep away the metal box walls. The Cadence interface will then add in feedlines with ref shift.

Yes and yes.

For such small L values where feedline have a significant amount of total inductance, add a realistic amount of feedline length. The length in Sonnet was chose to minimize the effect of box wall metal on inductor performance. If your lines in the circuit are shorter, you can remove only a part of the line, by setting the port reference plane to the desired position.