search

eez-open / modular-psu

EEZ Bench Box 3 (BB3) Modular T&M chassis
https://www.envox.eu/eez-bb3/
Other
313 stars 97 forks source link

SDRam trace length match #2

Closed LJJS closed 5 years ago

LJJS commented 6 years ago

The length of the bus for the SDRam chip should have all the same length. The bus operates at 143MHz, so the total trace length should be all equal.

Also, the length of each trace at each layer should have the same length!

prasimix commented 6 years ago

Thanks for that. This is a first prototype and we'll see how it's going to work. I do believe that for keeping all traces of equal length is not ask for introducing 6-layer PCB.

LJJS commented 6 years ago

I have done some testing with the length matching. Just an example how it can be done:

length matching

Sure, this is not the final solution, but I think it shows with some effort it is possible to tune the traces to very similar lengths. The board right now have differences in length to a maximum of 10mm, with a typical trace length of about 35mm. With some work it should be possible to match the length of the trace with a maximum difference of 1mm, even with a 4-layer PCB.

But also said, I think it will work just fine with these lengths at these frequencies, but it would be a nice detail in engineering.

prasimix commented 6 years ago

Thanks for effort. Actually I tried to use meandering but the results was pathetic at least in v.7.4. I'm going to commit a fix for #3 and such changes should be applied in that revision.

Rerouter commented 6 years ago

Most of the top layer can be matched without meandering, I have done it in kicad, but would probably be a pain to recreate in eagle. meander meander 2

The bottom layer needs a bit more agressive length matching, currently i have matched within 10mm, I approached it by shortening the longest traces first, leaving almost nothing to be added to the top layers,

But the control lines, some of them have 30mm differences, that took some manual creativity,

Rerouter commented 6 years ago

LJJS, the timing specifications only require a +-10mm matching, Attached is my current effort to that 10mm margin, but if you want to continue down to under 1mm, be my guest.

If you want to go the whole hog, the RX/TX buses on the ethernet controller are also out of skew by up to 30mm,

EEZ DIB MCU r1B1.zip

Current lengths, Address Group: A0-A5 - 45.0 A6-A11 - 45.5

Command Group: SDNRAS - 45.0 SDNCAS - 53.7 SDNWE - 45.0

Control Group: SDBA1 - 50.0 SDBA0 - 49.5 SDNBL0 - 50.0 SDBLN1 - 54.4 SDCKE0 - 45.0 SDNE0 - 47.0

Data Group: D0-D1 - 45.0 D2 - 48.4 D3 - 48.4 D4-D12 - 45.0 D13-D15 - 50.5

Clock: SDCLK - 55.0

prasimix commented 6 years ago

Thanks, I believe there is no need to go down to +/-1mm. First, we have to see what will happen when I tried to run existing PCB (that should be in a week or two). Should I just add that changes into my next commit or you are going to initiate a new pull request?

Rerouter commented 6 years ago

I'll Initiate, As I have also wrapped up Kicad-ising the schematic, and generating a netlist.

Rerouter commented 6 years ago

https://github.com/eez-open/modular-psu/pull/5

Pull request made.

prasimix commented 6 years ago

I managed to open KiCad file today. Unfortunately it seems that it doesn't include bug reported and fixed in Eagle under #3. Any chance to include that change?

Rerouter commented 6 years ago

https://github.com/eez-open/modular-psu/pull/6

All kicad files are now brought up to eagle master.

Rerouter commented 6 years ago

https://github.com/eez-open/modular-psu/pull/8

Missed an unconnected count,

prasimix commented 5 years ago

Prototype works predictably, can be reopened if some issue arise.