Electrical – Differential impedance calculation: reference plane doubt

The calculation looks correct, but the tracks are too wide.

For example, a single 0.75 mm track on a 0.36 mm substrate, has an impedance to ground, unbalanced, of about 49 ohms. Thus a "differential pair" any long distance apart will have an impedance of 98 ohms, without any coupling between the lines, i.e. infinite mutual impedance. It only takes a tiny bit of mutual impedance to bring the differential impedance down to 90, which is why your package puts them so far apart.

Hence the warning about spacing/height. You don't really have a differential pair, you have two single-ended lines, which as you say isn't ideal for rejection of coupled interference, or reduction of radiation.

A differential pair should have a significant mutual impedance, at least similar to or smaller than the impedance of each line to ground.

Keeping the final differential impedance constant, you can make each track narrower, raising its impedance to ground, and move them closer together, lowering their mutual impedance. This makes it more like a differential pair, less like two separate lines.

On this thin substrate you need to make the traces a lot narrower, so they can be closer together, so the mutual capacitance dominates the ground capacitance. Try 0.25 mm if your tracks can be this narrow, and see how far apart they need to be to get 90 ohms.

An advantage of a local power plane is that you can leave all the power routing out of your signal layers and in stead focus on the coupling, routing and impedance control of your signals.

Other than that the best advice is always based on your complete and exact design, so I'll tell you some of my preferences and their reasons, and leave them for you to consider.

For reasons of know-variables I prefer to keep no other layers between the GND and important signals, so in complex designs I try to make as many Signal layers directly next to a ground that fits my stack-budget (of course I'm not spending the money for 16 layers on each design I make!). And if I can only get 1 reliable layer like that, I make sure that layer has only signals and hosts at least the signals that are most important or highest frequency.

For the distances of the stack-up you best call the fab you are having the PCB made at, they know what they can do and what they stock. Once you have those numbers you can use them for your impedance control if you need to.

They can also tell you how accurate their PrePreg procedure is. If it's not very accurate or the layer it is spread on has a lot of copper areas and a lot of gaps as well (this makes PrePreg harder to get uniform) sometimes you will want your Signal and GND on either side of a normal plate, to be able to perform good impedance control. If that is a demand you might want to go for your first choice, but swap the "SIG" and "Sig/Pwr/Gnd" layers.

Another thing you put in your title is Analogue, if you have high-fidelity requirements of analogue signals you are not going to regret splitting your Analogue and Digital power domains completely, including the ground planes and only connecting them at the power-input of your board. You'll be thanking yourself for the extra effort once you find you measure very little digital noise in your analogue signals.