In our product we use a TLK10232 for a connection (10G-KR, Clause 45) via a backplane to a central unit with a switch from Marvell 98CX8129 Hooper.
Our current problem is, that we are not able to reduce errors in the direction from the central unit to the slave unit (down-stream) counted by the HS_ERROR_COUNTER (dev.: 0x1e, reg.:0x0010) to zero, even with modifying the registers in the TLK10232 from the default values.
The amount of errors depends on 2 things:
1. The slot of the backplane. The shorter the distance between the central unit and the slave unit, the bigger the amount of errors. The issue occurs mainly for short distances.
2. The temperature. The issue increases with low temperatures (< 0°C).
The distance on the PCB between backplane and TLK10232 is less than 4 cm.
The following register settings are changed from the default values on the new unit:
1. dev.: 0x1e, reg.: 0x0002 => 0x811c (better than default)
2. dev.: 0x1e, reg.: 0x0004 => 0x1320 or 0x1820 (but very bad result after new link training)
During our tests we accidentally changed the register dev.: 0x1e, reg.: 0x0002 to 0xa11c, then we had even better results of the error counter than with 0x811c. Since bit 12 is for "TI use only", the impact of the errors is not clear.
Now our questions:
Are there any other settings of the registers in the TLK10232 which have impact on the data transmission, especially those for "TI use only"?
What explains the influence of low temperatures on the result?
