TUSB7340: No PCIe link - noise on power rail?

Ferdinand, the noise is started at red line below where PCIe clk started?

Hello Brian,

yes, exactly. I just made a measurement showing the PCIe clk as well:

 TUSB not found

TUSB found

As you can see the noise starts with the PCIe clock.

The PCIe clock comes from a clock buffer on a COM Express module and uses a different supply voltage.

Regards
Ferdinand

Ferdinand:

     can you probe 1.1v and 3.3v, 48Mhz clock and GRST?

Regards,

Brian

Hello Brian,

sure thing:

 TUSB not found

 TUSB found

And a closer look at the 48MHz:

Do you already have an idea what the issue might be?

Regards
Ferdinand

Ferdinand:

    Pass/fail is happened on the same board  or different board? is the failure consistent?

regards,

brian

Hello Brian,

the measurements I took were all done with the exact same setup. The failure is not consistent, that is, power cycle the system several times and there are occasions where the TUSB is found.
The issue is not limited to this particular board either but has been observed on numerous boards. So far we've been able to find combinations of carrier and COM express board that do not show the issue at all. This has been an minor issue for a while, but the last batch of boards show this behavior very prominently and it has not always been possible to find a set of carrier and COM express that works.

I'll send you the schematic sheet via private message for completeness.

Regards
Ferdinand

accepted

Hello Brian,

I did some further testing where I added additional capacitors to the power rails and replaced the ferrite for the 3.3V analog voltage with a piece of wire. The noise on the 3.3V analog rail was reduced notably, but the system still behaved the same way:
 TUSB not found 8, added caps to C4809 and C4810, added 2x 0402 4,7uF to each Vcc, replaced L55 with wire

I also replaced the oscillator and the divider circuit with a 1.8V oscillator that was supplies externally. This had no effect, either. I also noticed that the TUSB can be detected as a PCIe device if the 48MHz clock is missing.

It seems as if the TUSB has trouble booting; when "soft" resetting the system from U-Boot, and not power cycling it, the TUSB is only detected if it was detected after power up. Soft resetting the system does not change whether it's found or not, and soft resetting only toggles the PRST signal.
We wondered whether it could be a timing issue, that is, the TUSB reset and the expected PCIe behavior from the COM express module don't match up, but in that case a soft reset should work reliably.

Do you have any further ideas?

Regards
Ferdinand

Ferdinand:

'    Can you zoom in waveform for 1.1v, 3.3v 48Mhz clk and GRST?

    Also send schematic for 1.1v 3.3v and GRST?

regards,

brian

Brian,

I've already provided a close up view for the 1.1V, 3.3V analog and the 48MHz clock. Can you please be a bit more specific on what you want to see? Is it the shape of the noise, the timing relation between the signals or something different?

I also already sent you the schematic fot the 1.1V and 3.3V supplies. The GRST signal is generated by a board management controller (STM23) and is used in many places in the system. What exactly are you looking for?

Regards
Ferdinand

Ferdinand:

     I want to see the timing relationship between 1.1v and 3.3v, 48Mhz and Grst. Also need a close look of  GRST, there is a glitch on GRST.

Can you remove GRST signal and just add 0.1Uf Cap to ground on GRST pin?

Regards,

brian

Hello Brian,

here the measurements you asked for:

 TUSB found - additional caps, no ferrite for 3.3V analog and 1.8V oscillator with external supply

I measured the 3.3V with an active probe and not the voltage rail probe, that's why it shows more noise.

 TUSB not found - additional caps, no ferrite for 3.3V analog and 1.8V oscillator with external supply

 GRST

The "glitch" always appears and there is no difference visible whether the TUSB was found or not.

GRST close up

I added a 100nF cap to the GRST signal:

 Rise-time increased by about 1000x, but it seems to improve things.

I changed the cap to 1nF wich caused the rise-time to be about 50ns.

System behavior basically unchanged:

TUSB found

TUSB not found.

I'll try disconnecting the GRST and just adding a cap next.

Regards
Ferdinand

ok, Pls try with just external Cap.

Also, can you adjust 1.1v delay or ramp time with LDO? 1.1v seems up late than 3.3v

Regards,

Brian

Hello,

I tried disconnecting the GRST and adding a cap to ground. Now I have a major glitch/runt in the GRST signal:

(close up)

However, this seems to have no effect on whether the TUSB is detected as a PCIe device. Sometimes it is found and sometimes it's not.

I'll try to use an LDO for the 1.1V next.

Regards
Ferdinand

Brian,

I modified our power supply so that the 3.3V and 1.1V start at approximately the same time. This is controlled by two separate soft start pins, each with a cap to set the time, so depending on component variations this will vary.

This brought a significant improvement! On my test board the TUSB was correctly detected in more than 90% of the time. Previously it was less than 50%.

I then added a small capacitor to the GRST signal to ground to get rid of the reflections I saw, but this seemed to have a detrimental effect - the TUSB was not found more often again. First I used a 1nF capacitor, which I then changed to 220pF to shorten the GRST rise time, but this had no notable effect on the general behavior.

The Input transition time (trise and tfall) are specified as max. 25ns in the datasheet. Do you know how critical this is?
With the smaller capacitor the reflections are becoming visible again.

Do you have any further ideas?

Regards
Ferdinand

adding more delay on 3.3v so 1.1v power up early than 3.3v 

Hi Brian,

the 1.1V now start before the 3.3V and that seems to do the trick. We'll start testing this on more boards, but it looks very promising.

I've left the GRST as is. The reflections we see in the measurement don't seem to matter, but only testing can confirm this.

Thanks!

Regards
Ferdinand

Great,  you still see the noise sometimes but functional ok?

Regards,

Brian

Hello Brian,

I've tried changing the power sequence on a few boards but it didn't solve our issue. However, I think in these cases it was linked to something different.
The PCIe reference clock is sent to the TUSB and and mini-PCIe module (that was not a good design decision...) and whether the TUSB is detected seems to depend on the termination in the mini-PCIe module.

Could you please send me the IBIS model for the TUSB? I'd like to see if I can setup a sensible simulation to see how small changes on the PCIe module might affect the signal at the TUSB.

Thanks!

Regards
Ferdinand

Ferdinand:

    You can download IBIS model from ti.com SLLM133.ZIP.

Regards,

Brian

any more update?

Hi Brian,

thanks for the link to the model - I saw some of the old posts regarding the TUSB and didn't think it was on the product page.

But no news yet - the license for our ADS simulator is in use and we've sent in the differential probes for calibration. I'll also be out of the office the next 2 weeks, so it will be a while before we know more,

Regards
Ferdinand