Differential LVDS twisted pair vs parallel track PCB

Hello all, I recently had to modify an FPGA based board to add a couple of LVDS clock lines between neighbouring FPGAs (about 100mm long and working at 150MHz). The existing clocks, in the PCB layer stack, were single ended and suffering from severe cross-talk from adjacent data lines. The new connections used previously unused and unconnected pins.

My first prototype was made using fine enamel wire, twisted to keep the wires in close contact and to provide some rigidity to the wire mod. This appeared to work fine and I remember 'scoping up the receive end and seeing nice monotonic signals. All tickety boo until we made a stitch on 0.2mm thick PCB which basically solders on to vias to do the same job in a much neater and easier way from a production point of view. Being on a single sided PCB, the tracks were routed

0.1mm spacing and track width with about 5mm between adjacent pairs for safety. The PCB sits directly on top of the back face of the board over a solid ground plane (for the most part excepting where it sits over the FPGA vias to gain access to the required vias).

The signals now look different so I think I have changed the impedance radically. The received signal is now more square like with a small amount of ringing (though not excessive).

So my question is, can anyone explain what might have happened/what differences should be expected between the two types of mod?

BTW the receive end has an on-chip differential term of around 100 ohms.

Many thanks,

Reply to
davew
Loading thread data ...

tried putting the numbers in this:

formatting link
ed-microstrip-impedance

-Lasse

Reply to
langwadt

I think you are saying that you created an add-on board to carry the signals between the chips. The add-on board is single-sided and has differential pairs of traces. Is that the situation?

The differential impedance of the add-on board is probably wrong. Since it's single-sided, the usual impedance calculator programs won't work. It is sort of borrowing the main board ground plane, but that's not well controlled.

I'm guessing that your differential impedance is too high. TXLINE calculates your geometry at 130 ohms assuming a ground plane opposite, so will be even higher without a ground. The wires and vias and traces to the chips complicate life more.

If you were to add a piece of copper or aluminum foil or tape to the back side of your little board, then it becomes differential pair over a ground plane, maybe 130 ohms as noted. You could try that and see if things look better or worse.

One trick is to add dielectric on top of the traces, layers of tape or some plastic or whatever, to bring the impedance down.

But squarish with a little ringing isn't bad. Does it work OK?

John

Reply to
John Larkin

formatting link

Hey, that only differs from a TXline calculation by about 10%. Not bad in this buisness.

But both programs assume a backside ground plane, and Dave doesn't have one.

John

Reply to
John Larkin

Thanks to both for your help.

Yes it's a single sided board with two differential pairs.

Thanks for the calculator link, the impedance seems to come out at about 160 ohms given the substrate height we have used (it's actually about 0.5mm) so in any case too high.

As you say the quality of the received signal is actually OK in my view (perhaps a little too much ringing but not bad at all). The problem I have is to adjust timing in the firmware to compensate for the change in waveshape I think. The edges are better defined with the stitch on board because they are more vertical. A lot more sluggish with the twisted pair version and of course with a wire mod the results will be much more variable and therefore prone to potential problems.

Thanks again

Reply to
davew

Differential impedance won't care a lot about the ground plane, it is mostly due to the spacing and diameter of the two wires and the insulation on them. Most likely, the capacitance of the vias and the traces to the chip leads is adding a capacitance at the receive end. I doubt a twisted pair of small magnet wire will really have a differential impedance of 160 Ohms. Most pairs of closely-spaced wires end up near 110 Ohms.

Jon

Reply to
Jon Elson

Oh, yes, if the OP did this, then you are certainly right. I thought he was just using existing vias to make hand-wiring the enamel magnet wire easier for the rework folks. If you make a differential pair on a single-sided board, which you can do, you need to put the two traces INSANELY close together to keep the impedance down. Remember the twisted pair of magnet wire? The "traces" are only a couple thousandths of an inch apart. You need to get the copper traces pretty near to the same spacing to get down to the 110 Ohm region.

Putting the pair on a double-sided board doesn't lower the DIFFERENTIAL impedance all that effectively.

Jon

Reply to
Jon Elson

I thought that at first, but on re-reading it seems to be a little single-sided kluge board with paired microstrip traces.

If you make a differential pair on a

Depends on the dimensions. If a board is single-sided and the board is thin relative to trace width, the impedances will be higher than the same board with a ground plane opposite the traces.

If the board thickness is much greater than the trace width and spacing, the presence or absence of a farside ground plane won't matter much. Think of a differential pair on a plane, below which half of the universe is filled with FR4.

I think we agree that his differential impedances are probably well above 100 ohms. Wider traces would help. Or a farside ground plane. Or both.

Two uncoupled (ie, far apart) 50 ohm traces *are* a 100 ohm differential trace!

John

Reply to
John Larkin

Right, but that REQUIRES a ground plane.

Jon

Reply to
Jon Elson

Right, the dielectric constant of the universe is too low.

John

Reply to
John Larkin

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.