FEDEVEL
Platform forum

Andrew , 10-23-2024, 03:10 PM

Do these results make any sense? Or is there perhaps a bug in the Altium Designer Keysight SI plug-in?

I'm testing the new Altium SI plug-in by Keysight. In particular, I'm curious about using it to test the impact of termination resistors on the DDR3 lines. The plugin has a bug that's keeping it from running our on AD project, so while Altium's tech support is looking into that, I'm testing it on an identical stack-up in a simple PCB project. It's literally just running from one header on the top side of the board to another on the bottom side of the board.

Here is the simple PCB layout without the termination resistor. No top copper pour, just a curved 50 Ω trace routed from one pad to another with teardrops and ground stitching vias.

Andrew , 10-23-2024, 03:12 PM

BUT now check this out. All you have to do is add the termination resistor in, and it screws everything up. Note that the header pads themselves are only about 1500 mil apart.

See how the presence of the termination resistor results in some massive degradation to insertion and return loss? It doesn't make sense to me. While the values might still be arguably acceptable for DDR3, I don't understand how a simple pull-up affects the signal this way. And the results are the same no matter where you position that resistor. Do these results make any sense? Or is there perhaps a bug in the plug-in?

QDrives , 10-23-2024, 07:13 PM

You do have a stub in the shown picture.
What if you route the signal 'through' the pull-up (down)?

Andrew , 10-24-2024, 10:54 AM

Here's what happens when we try eliminating the stub. I've made it adjoin the trace or the header pad (which you'd think would be the best location), but never ran it through the 0402 pad. Let me try... I should also note that I mispoke. Resistor location does in fact change the S11 and S22 response curves somewhat.

BTW, the plane is 4.6 mil under it and tied to GND, although I've also tried tying it to VDD without luck.
Placement 1:

Andrew , 10-24-2024, 10:54 AM

Andrew , 10-24-2024, 10:54 AM

Placement 2:

Andrew , 10-24-2024, 10:54 AM

Placement 3:

Andrew , 10-24-2024, 10:55 AM

Now let's see an example where there's a trace stub but no termination...

Andrew , 10-24-2024, 10:55 AM

Weird, right? Now here's what happens when we make the trace super long... The frequency response is about what I'd expect for those nets.

Andrew , 10-24-2024, 10:55 AM

Moving it farther back down the trace does change the response...

Andrew , 10-24-2024, 10:56 AM

But check this out - when I remove the termination resistor and leave and equal-sized copper pour of the same dimensions as the 0402 pad connected to the signal net, it gives me a very clean response.

QDrives , 10-24-2024, 06:59 PM

1) What is the resistance according to SI?
2) What is the source and receiver?
3) That is the impedance of the source and receiver?

Andrew , 10-25-2024, 12:30 PM

I apologize for the delay in responding. I was out yesterday for a medical procedure and just now getting back in the swing of things.

1) in the screenshots above, it confirms a trace impedance of 51 ohms. This is an agreement with the Simbeor third party libraries which are used in calculating the trace impedance in the layer stack manager.

2) in this very primitive example, there is in this very primitive example, there is necessary not necessarily a source and a receiver. It's simply two headers. And the key sign SI plug-in does not require such to be specified, as it simply calculates the S11 and s12 parameters.

3) The impedances listed in the screenshot are for the trace itself or for the pads.

QDrives , 10-26-2024, 12:18 AM

But if you just have the S11 and S12, the termination will make it worse.
The termination is to compensate for the imperfect system (source -> transmission -> receiver)

Andrew , 10-28-2024, 04:02 PM

The termination resistors are primarily for SI purposes, correct? So why would they cause degradation in the S11 and S12 parameters?

Or is this an instance where the parameters are *not* good indicators of SI simply because they don't take power into account?

QDrives , 10-28-2024, 08:30 PM

If your source, transmission line and receiver are all (about) 50 ohm, adding a termination resistor would reduce the received signal to 50% (or 6dB).
However, if your receiver is high impedance, than it would work correct.

If you as the tool to give you the S11 and S12 values, I assume that the tool goes for 50 ohm for both source and receiver and all 'errors' are from the trace.

Andrew , 10-29-2024, 05:42 PM

No, it doesn't consider the source and receiver. Literally just traces. Or am I misunderstanding what you're saying?

I'm gonna pester Altium again tomorrow.

QDrives , 10-29-2024, 09:22 PM

There is a video with Eric Bogatin. I do not know if it was with Robert or one of his Teledyne LeCroy ones.
"*The signal enters the lines and sees 50 ohm. It goes down and down the line and sees...*" -- **2x 50 ohm in parallel**!

Andrew , 10-30-2024, 11:36 AM

That makes sense... Although in this instance, the resistance value in the simulation doesn't change the frequency responses. I've tried it with termination values as high as 1 MΩ :/

QDrives , 10-30-2024, 04:33 PM

So having a termination changes it, but the value of termination does not have an influence???

Andrew , 10-30-2024, 05:52 PM

QDrives , 10-30-2024, 08:17 PM

Let me guess the response from Altium: "but why would you have any other termination resistance than 50 ohm?"

Andrew , 10-31-2024, 10:58 AM

Oh, they haven't even bothered to respond at all. Even after last week, when the rep put me in direct email contact with Tech support. And I emailed them all again yesterday letting them know that it had been over a week on my three tickets and I hadn't heard back 😐

The rep is trying really hard to sell us on these plugins, but I made it a point to tell him we're not interested purchasing a product that Altium provides no documentation or tech support for.