DDR3 Series Termination Length Matching

Question

I am implementing a DDR3 interface on a PCB and I have a question regarding the termination of address/control/command/clk traces.

I have series terminations for all the required traces mentioned above, my question is whether or not the traces going to these terminations require length matching or should they just be kept as short as possible and left as uneven termination stubs? To me it makes sense to keep them as short as possible even if lengths aren't matched well, but I wanted some other opinions on the matter :)

Thanks!

Answer 1

In an ideal world where the trace impedance, load impedance, and termination impedance are all ideal and perfectly matched, it doesn't make any difference how long the termination stubs are, or whether they're length-matched, because a terminated TL of any length is equivalent to a pure resistance, absorbing all of the signal presented to it and generating no reflections.

In the real world where there are parasitics and all matching is imperfect, it makes sense to keep the termination stubs short to minimize EMI and to keep the delay of any reflections as short as possible. There doesn't seem to be any compelling reason to length-match them. It's the paths between sources and loads where you need to avoid skew.

Answer 2

I was still unable to find a solid written answer in the JEDEC specification or in design guides but after looking through some reference designs I found that termination resistors are always placed as close as possible to the final signal pin (last ram chip).

Answer 3

In short, you don't need length matching for termination resistor traces but you should keep this length minimum, maximum of 300mils is recommended. You can find same recommendation in Micron reference design (I don't recall what exact UG it was, sorry)

Answer 4

Length matching is needed, you don't want the data or control to switch after the clock. Here is an example of how this is to be done.

Source: https://fedevel.com/blog/ddr3-length-matching-rules

Termination stubs don't matter as long as you keep them to something like 1/20th of the wavelength, if you get into longer then that, you need to run an analysis.