Hacker News Reader

Comment by pythonguython | Hacker News Reader

Can someone explain why those differential pairs are routed with so many curves instead of straight paths? (E.g. see the photo under the “ADC and DAC rotting” section)

View on HN

"The traces are length-matched with squiggly lines[...]. The trace matching requirement is ±10 ps according to the Zynq PCB design guide, which is approximately ±2mm in trace length. [...] There is also some delay difference inside the FPGA package which should be considered in the length matching."

This is a shortened excerpt from the article. It can be found below the image with six colorful images of PCB layers. I'm curious how the delays inside the FPGA package are known. Is there a table which pin adds how much delay to a signal or something like that?

The FPGA manufacturer has characterized all their package pin delays. It's possible to export a csv file with internal delays of the package for each pin from the FPGA design tool. With Xilinx Vivado it's just File -> Export I/O Ports.

Does KiCad allow to take these delays for each pin into account automatically or do you need to do all that manually?

Edit: it's using the pad property of "pad-to-die-length" if doing it manually, right?

No, it needs to be done manually. It wasn't as tedious as it sounds though. Most of the pins are very close in delay already and there were just few traces that I had to adjust a little.

How is kicad getting along with RF work in 2024, would you say?

I did some prototyping but never got the stage if actually fumbling around with PCBs a few years ago, things seemed to progressing quite well.

One of the main benefits of using an FPGA is that you can compensate for trace length mismatch with timing constraints

Not usually with IO ports directly associated with hard IP blocks like the DDR controller in the Zynq though.

Path matching, as the other comments say, but also if you have a corner, at >Ghz frequencies it will emit RF.

Matching overall length with other traces, is my guess.

Reply on news.ycombinator.com