Soldering issue with U48

[mdavidsaver]

With a batch of 20x Marble 1.4.3, 11 boards failed the MGT loop back test with transceivers Y0 and/or Y1. Also one failed on Y6.

[mdavidsaver]

The commonality of Y0/Y1 is a path through U46 and U48 mux. chips. Tests performed:

1. Attempt to toggle the mux. and test loopback through the FMC connector.

This also failed, indicating that the issue was either in the common part of both loops, or somehow in both the FMC and QSFP circuits.

2. The MGT test was modified to use different transmitter and receiver pairs (@keithpenney I believe you can say how). This isolated the transmit side.

[mdavidsaver]

Inspection of the leadless U46 was difficult as the pads did not extend past the package body.

A resistance to ground test of pin 2 (power down) reported ~6.8k ohms on a functioning board, and ~90k ohms on a non-functional board. Indicating a poor solder.

[mdavidsaver]

@ldoolitt Noted that the footprint used for Marble <= 1.4.3 does not match the OEM recommendation.

[mdavidsaver]

With a batch of 20x Marble 1.4.3, 11 boards failed the MGT loop back test with transceivers Y0 and/or Y1.

Removal and replacement remediated the Y0/Y1 MGT test failures.

Also one failed on Y6.

This appears to be a separate issue.

[mdavidsaver]

5435c09 enlarges all of the pads.

Below is screen capture where I have tried to overlay the old PI3DBS16212ZBBEX footprint above the new as a quick confirmation that the pads do not move.

[ldoolitt]

Closing with the release of v1.4.4.
I hope this isn't unduly optimistic.