Just adding my opinions on this subject based on 20 years of SI work.
100 Ohms was usually a target impedance for most differential signaling (PCIX, SAS, SATA, and a bunch of others.) It made sense to have 2 50-Ohm traces (50 was a good round number for most VNAs, too.
I believe Intel came up with the 85 Ohms target (about 20 years ago) for 2 main reasons.
1. 100 Ohms was had to achieve on PCBs at that time since trace widths were greater than 6 mils usually. To get to 100 Ohms, we used to have to void GND planes under traces to use a secondary GND further away.
2. If you built a receiver with 2 50-Ohm resistors in series (to get your 100 Ohms), you would usually have at least 5 to 10 picoFarads of die capacitance in parallel to each 50 Ohm resistance, which would look closer to 37 to 38 Ohms at the rise times of the new Gbps signals. So, they figured that 85 Ohms would be a better match to the traces.
So, 85 Ohms became the PCIe standard – but it is not that relevant today with smaller die capacitances and 3 mil traces.
As the article states, the loss at 85 Ohms is usually greater than the loss at 100 Ohms for the PCB too, not just the cabling. But that has changed with low-loss dielectrics they didn’t have in 2000 for most commercial PCBs. Yeah, there were exotic materials (Teflon), but only special needs for microwave and military applications could afford to use them.
Add to that HVM (High Volume Manufacturing) of traces on PCBs have historically produced tighter tolerances for 85 rather than 100 Ohms, and sometimes it is better to be consistent (match those impedances) in your design than to achieve an ideal value.
93 Ohms comes about because connectors are difficult to design to get down to 85 Ohms, a mid-target of 93 also allows the connector to be dual-purpose for both 85 and 100 Ohm systems, reducing the cost of making 2 different versions.
It is tempting to select cable impedance to match what is used on the PCB. However there are a couple of anecdotal considerations. First, higher frequency reflections are a greater function of the mating connector impedance than the PCB itself. If an incoming signal experiences an increase from 85 on PCB up to 93 ohms for a given connector, it is best to remain at 93 ohm and not create new reflections by following it with a decrease to 85 ohm cable
Secondly, the relationship between impedance and loss is inverse of the PCB relationship. That is, higher impedance cable has less insertion loss whereas higher impedance PCB would otherwise be more loss. The loss difference between 85 and 100 ohms is as much as 14%, and is explained in the actual geometry of the twinax. This is the experience for Samtec twinax eyespeed cable and may not be the experience for all cable suppliers.
Ultimately, cable impedance choice depends on the priorities for your system. If insertion loss is the greatest constraint, then 93 or 100 ohm cable is the best choice.
