XILINX RocketIO / MGT signal quality problems

Hi anyone out there,

we have designed a board with XILINX Virtex II Pro using the RocketIO / MGT serial high speed transceivers. Recently we experienced a problem with bit error rate (BER) and measured the signal quality of the MGT serial links with a 20Gsample scope and 5GHz differential probe. We found a very poor signal waveform and got an almost closed eye diagram. We analyzed this phenomenon and now we assume, that the signal degradation is caused by high reflections on the line. The overshoot and undershoot amounts to 50% of the singal swing. It seemed, that the MGT receiver's input termination does not work properly. Then we tried a TDR ("time domain reflectometer") measurement to check the impedance characteristics of our board even into the MGT's termination. The board traces are fine. Some impedance mismatches are to be seen at vias, AC-coupling capacitors and the Virtex II Pro package. But we think, these are not too bad, the mismatch is in the range of 20%.

Does anyone have experience with Virtex II Pro RocketIO? Did anyone measure signal quality or eye diagrams on such a link? May the impedance mismatches cause the high ringing we found? Can anyone imagine the reason for the reflections though the signal path's impedance seems to be not so bad?

At the moment I don't have a clue. Thank you for any hint!

Michael

Reply to
Michael Mustermann
Loading thread data ...

Michael,

At the speeds of the MGT signals, just about anything can be a 'bump in the road', and cause reflections.

No, the termination in the receiver is not perfect, (nothing is perfect), but it is just fine regardless. Thousands of customers have pcb's working at 3.125 Gbs error free, so it is more likely that you have a pcb issue in your board.

I suggest you immediately contact your local FAE, and arrange to go visit one of our RocketLabs locations, where we have all of the equipment to troubleshoot just such an issue, and the FAEs associated with the RocketLab are all trained and familiar with the equipment, and how to address the issues.

One of the most common mistakes made in measuring the input impedance, or return loss of the 100 ohm differential receiver, is that they measure it single ended (50 ohm) and fail to take into account that a differential return loss measurement is not a trivial or simple thing to characterize accurately. For example, two single ended 50 ohm traces are NOT 100 ohms differential (they are less if they are routed together as they should be to be differential). Bad mismatch right there!

This goes for TDR as well. Unless it is a true differential TDR measurement, you are not measuring what you need to measure (eg the Tek CSA8000 is the only true differential TDR scope that I know of, although I think Agilent now has one as well -- check! does it send two impulses (or steps) at the same time of opposite voltages? If not it isn't differential).

As well, the time resolution fo the TDR may be much faster than the rise time of the MGT signal, and may be showing issues that do not affect the MGT operation (ie a mis-match at 20 GHz is not an issue, as the signal has no energy at 20 GHz).

Save yourself time and money, and go use the services that we provide to get your problems solved. After all, we wnat to sell you chips, not have you endlessly troubleshooting.....

Aust> Hi anyone out there,

Reply to
Austin Lesea

I've used an Agilent 54754A dual 18.4GHz TDR plugin in an 86100A scope for testing 10Gbps connections. I *think* it does a true differential measurement. [ I don't have the documentation handy. ]

Yes, but better time resolution means better spatial resolution, allowing you to work out what went wrong with your board design.

(I found this out the hard way.)

Regards, Allan.

Reply to
Allan Herriman

Hi Allan, hi Austin,

thank you for you reply, but it won't help. Of course, at MGT's speeds everything becomes just esoteric. But we know this, we've designed serial gigabit board for several years. We are familiar to single ended and differential impedance and know how to lay out a board for these signals.

We've carried out _true_ TDR measurement with Tek CSA8000 series scope which is able to send a positive and a negative pulse exactly at the same time. Both the board and the MGT receiver don't seem to be the issue, as I wrote earlier (see below).

When I cut off the signal path by lifting the AC-coupling capacitors

100mil before the MGT receiver and short the differential pair by a simple 100 Ohms termination resistor, then I get wonderful waveform and eye diagram!!! So it is proven, that I do not have a problem along the traces, but together with MGT receiver.

We use such links in the opposite direction as well, with the MGT as the transmitter and a competitor's chip as the receiver. All other details are absolutely identical - same trace geometry, same connectors, same distances - just different direction. Those links work _excellent_ and the eyes look as pretty as in a schoolbook!!! So what do you thing? Do I have a problem with my pcb???

We have contacted the local FAE and he had no idea except the noise at the MGT supply voltage. We were advised to measure at the regulator and found app. 70mVpp noise. But even a lower noise supply with 30mV hardly provided signal improvement.

After 15" FR4 and two special high speed connector (AMP HM-Zd) we get an 65mV eye opening at the MGT receiver package pin, but an

650mV overall signal swing! App. 90% is degraded by reflections! I could double the transmitter voltage swing, but I wonder, if this is the right approch...

I'm out of ideas now. Thank you for your assistance!

Regards.

Michael

Allan Herriman wrote:

Reply to
Michael Mustermann

I checked the 54754A. It claims to be a true differential tester.

Regards, Allan.

Reply to
Allan Herriman

Michael,

I apologize if I seemed to imply that you were new at this.

Series capacitors? Why? When you AC couple, now you have the impedance bump due to the layout from the caps, and their own ESL/ESR.

Has anyone extracted the pcb layout and simulated the pcb traces using a

3D field solver? Have you looked at the impedance mismatch created by the series caps?

Again, I strongly suggest getting to a RocektLab which is equipped to handle this. A quick call to a disti FAE (or even a Xilinx FAE) that has no real knowledge of your situation is not really going to get you the help you need.

Is there a case on this? Who is the CAE?

Working with our Hotline is the best, fastest, and most effective way to solve any issue. This forum here is probably the worst from a time/accuracy/solution point of view for a specific customer issue. It is very useful to ping others in the community.

Aust> Hi Allan, hi Austin,

Reply to
Austin Lesea

Solder some extras in parallel with them, but I doubt this is his issue. We've sent 2.5 Gb/sec SONET through capacitors and jumpers (stubs when open) without problems (the FR-4 helps you by reducing the high frequencies).

Here's a low tech long shot: maybe a ball is opening when the chip heats up. It wouldn't be always open, otherwise the TDR would have picked it up. So is it broken in one instance (one particular board, one location) or in several?

Rule out other low tech problems as well, like incorrect supply voltage.

--
/*  jhallen@world.std.com (192.74.137.5) */               /* Joseph H. Allen */
int a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time(0)
 Click to see the full signature
Reply to
Joseph H Allen

We had to manage different bias voltages for transmitter and receiver. Thus we separated both by AC-coupling capacitors.

No, we didn't exctract and simulate pcb traces. But TDR measurement showed an impedance bump from 100 Ohms down to 80 Ohms diff. And behind the AC-coupling we saw app. 115 Ohms which represented the MGT receiver.

We are in steady contact with Xilinx representative and FAEs. They have been to our lab. But there has not been any progress for a few weeks and no solution to our problem ! That's why I turned to this forum and asked for assistance.

What means "case on this" and what is a CAE?

I'm sorry for the bad public relation. Xilinx claims to have thousands of customers who have boards at 3.125Gig working fine. But until this very moment, nobody (even in this forum) could tell me: "I truly measured a beautiful eye at my board's MGT receiver! Here is the screen shot..." Actually, can _SOMEONE_ give me such a confirmation? Please!

The whole thing is weird and for my personal feeling quite nebulous!

Allen, sorry for my honesty! I hope you don't feel offended!

Regards.

Michael

Reply to
Michael Mustermann

Hi again (I confused the names Allan und Austin in my previous posting),

We had to manage different bias voltages for transmitter and receiver. Thus we separated both by AC-coupling capacitors.

No, we didn't exctract and simulate pcb traces. But TDR measurement showed an impedance bump from 100 Ohms down to 80 Ohms diff. And behind the AC-coupling we saw app. 115 Ohms which represented the MGT receiver.

We are in steady contact with Xilinx representative and FAEs. They have been to our lab. But there has not been any progress for a few weeks and no solution to our problem ! That's why I turned to this forum and asked for assistance.

What means "case on this" and what is a CAE?

I'm sorry for the bad public relation. Xilinx claims to have thousands of customers who have boards at 3.125Gig working fine. But until this very moment, nobody (even in this forum) could tell me: "I truly measured a beautiful eye at my board's MGT receiver! Here is the screen shot..." Actually, can _SOMEONE_ give me such a confirmation? Please!

The whole thing is weird and for my personal feeling quite nebulous!

_Austin_, sorry for my honesty! I hope you don't feel offended!

Regards.

Michael

Reply to
Michael Mustermann

Michael,

Please email me directly with your contact information, FAE name, distribution FAE, etc. so I may escalate your case. By entering a case in the hotline, and communicating your expectations, we know how to address the issue, and we can track it. It is is not resolved quickly, we can escalate the issue to a team of experts. Not dealing with a customer case is a serious issue for the support staff, and is dealt with accordingly. Without a case number, and a contact, no one is taking the issue seriously, as it is not in the system as a "problem."

FAEs may just consider this something that you may be 'playing' around with, and does not require their attention.

I will personnaly find out why the field organization has not provided you with the answer, but without a case open, I can tell you right now that is a major factor!

Visiting a RocketLab(tm), you could see many eye patterns on real boards.

formatting link

Is the on line tool that has been checked to show that the simulated eye patterns are accurate, and represent the actual eye patterns on the thousands of combinations.

I would suppose that others could send you their eye patterns, but that is up to them to decide to take the time and effort to do that. As well, posting graphics on this newsgroup is highly discouraged (as the graphics will be removed -- text only) so they would have to respond to you directly, and we do not know if the email address that appears in your posting is valid (many are not due to spam problems).

My direct email is austin @ xilinx . com (without the spaces).

One last comment, if the driver is also not 100 ohms matched, then any discontinuities upon reflection will re-reflect back to the receiver. There are some devices out there with very low, or very high transmit impedances (not ours) that have this problem. In any system or standard, the SI engineering makes assumptions. The assumption here is that both the Transmitter and the Receiver are to be as good a match as possible, so as to minimize the distortion to the eye pattern from 'normal' (that is, unavoidable)impedance discontinuities from vias, connectors, series AC blocking capacitors, package, and die variations.

A simulation of an improper impedance transmitter will demonstrate this issue.

(rant on)

For all those who are reading this: do not let this happen to you! If you have an issue, get a case number! Follow the case progress on-line through our on-line tracking system. If at any poinht you feel progress is not satisfactory, ask to escalate the case (bumps it up the system to the next level of technical expertise). You are the customer: you are always right (by definition). We do not have a choice. If you ask, we must respond. And we do.

(rant off)

Austin

Reply to
Austin Lesea

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.