Intel WGi210IT packet loss at temperatures > 65C

[Jean-Francois]

Hello intel,

I am facing an issue with packet losses on the i210IT (WGI210IT S LJXS, procured from Mouser) at a link speed of 1Gbps only, when operating at ambient temperature between > 65C and <=85C.

This issue has been reproduced on our own proprietary design, as well as 2x commercial off-the-shelf industrial SOM from different companies that use the i210IT. Let's focus on the issue within our own proprietary design.

Here is our configuration:

- External crystal (ECS-250-18-33-JGN-TR), ±30PPM frequency stability, ±20PPM frequency tolerance, with 27pF load capacitance and rated -40/+85C, ESR of max 40R, drive level of 100uW. The load capacitors are C0G/NP0.

- Using the i210IT internal SVR for +1.5Vcc and +0.9Vcc. Considering the DC bias of the capacitors used, we have ~41.2uF effective on +1.5Vcc and ~ 47.1uF effective on +0.9Vcc

- Using a dedicated magnetic (S558-5999-AT-F) rated for -40/+85C and validated already on other designs. Based on the intel reference design guide, the center-tap on the i210IT side are tied together, with 0.1uF per center-tap and 1uF global.

- 100R differential impedance on the PCB layout without any disruptions / traces routed over a plane split. The MDI pairs have a solid GND reference (on L2) and the impedance control has been verified with the PCB manufacturer. The ENET IEEE compliance was executed on the design without any issue for all link speeds (10/100/1000).

- The stackup is 4-Layer, 1oz external copper thickness, 1oz internal copper thickness

The issue:

In a PING configuration with a switch outside the thermal chamber or by doing MAC-to-MAC transactions between 2x WGi210IT, we observe a gradually increasing quantity of packet being lost when exceeding 65C ambient.

Here are the steps taken, to date, to try and diagnose the issue:

1) Our initial design had no heatsinking solution. Upon facing the packet loss issue, we designed and implemented a custom heatsink, and we measured the IC temperature using a thermocouple following Intel recommendation in the i210IT datasheet: at 75C ambient in the thermal chamber, the IC T(case) stabilizes to 98.6C, which is below the max T(case) of 105C, and the ambient surrounding the i210 stabilizes to 82.2C, which is below the max T(amb) of 85C. I was expecting this to solve our issue, but it did not.

2) We have used the lanconf tool to re-perform IEEE compliance test at 85C, and all tests passed successfully at 1Gbps. Thus I don't think the issue we're seeing is related to jitter, impedance mismatch, or other parameters controlled by external components to the i210IT.

3) We have used the lanconf tool with our ethernet card installed in thermal chamber to setup the i210 in internal PHY loopback and external loopback. These tests were executed successfully at 85C for 100M+ packets, there were no "Data verify failures"

4) I have changed the crystal to a precision oscillator (±1.5PPM) without luck, still seeing the same packet loss issue @ 1Gbps at > 65C

5) I have tried reworking our PCBa to power the i210IT using external regulator instead of the internal SVR, without luck, still seeing the same packet loss issue @ 1Gbps at > 65C.

See in the pictures below :

- how the thermocouples were setup to perform T(case) and T(amb) measurements

- how our thermal pad and heatsink are attached to the PCBa

- the date / codes that we have tested so far that all reproduced the issue

- the thermal measurements showing the WGi210IT T(case), T(amb) and packet loss seen in our thermal chamber

I am out of ideas. I'd need to speak to an intel hardware engineer to get feedback on the WGi210IT thermal metrics and proper operating conditions to achieve 0 packet loss.

Much appreciated.

Thank you!

[CarlosAM_INTEL]

Hello, @Jean-Francois:

Thank you for contacting Intel Embedded Community.

Based on the information provided, it seems like you have already taken several steps to diagnose the packet loss issue with the i210IT.

However, there are a few additional things you should consider:

1. Check the power supply stability: Ensure that the power supply to the i210IT is stable and within the recommended voltage range. Any fluctuations or noise in the power supply can affect the performance of the chip.

2. Check for signal integrity issues: Even though you mentioned that the impedance control has been verified with the PCB manufacturer, it might be worth double-checking for any signal integrity issues. This could include checking for any reflections, crosstalk, or other noise sources that might be affecting the signal quality.

3. Verify the thermal design: While you mentioned that you have implemented a custom heatsink and measured the IC temperature, it is worth considering if any other thermal factors might be contributing to the issue. For example, ensure that the thermal dissipation from the i210IT is adequate and that the surrounding components or environment are not affecting the temperature.

4. Consider contacting Intel support: If you have exhausted all possible troubleshooting steps and are still unable to resolve the issue, it might be beneficial to reach out to Intel support for further assistance. They might have additional insights or recommendations specific to the i210IT. Remember to provide them with all the relevant information, including the steps you have already taken and the specific conditions under which the packet loss occurs. This will help them provide more targeted support.

Best regards,

@CarlosAM_INTEL.

[Jean-Francois]

Hello @CarlosAM_INTEL , thank you very much for the detailed feedback. This is much appreciated.

I have initiated the process to get access to Intel premier support.

In the meantime, I just want to share additional information regarding the points you mentioned.

1) As we are using the iSVR, yes we have characterized and validated that the input voltage (+3.3V) and generated voltages (+1.5V and +0.9V) all respect the following criterias as per i210 datasheet section 11.3.1:

- Operational range

- Ripple

- Overshoot

- Transient response

- Brown-in and Brown-out

- Decoupling capacitance / ESR

2) For signal integrity, we have conducted IEEE 802.3 compliance suite of test at all speeds (10/100/1000) at ambient temperature, and again 1000Mbps at 85C, to make sure that there were no signal integrity issues. The compliance tests passed, thus eliminating this option.

3) The thermal design is what we are currently working on. We have performed extensive temperature testing with our custom heatsink solution, and we have determined that at i210IT T(amb) ~ 70C and i210IT T(case) ~ 85C, we are starting to lose packets.
According to the datasheet, i210IT T(case) of 105C is an "Absolute maximum rating" whilst i210IT T(amb) of 85C is the "Recommended operating conditions". However, note 3 of this table mentions "with external heat sink, airflow required for operation in 85C ambient temperature".
Based on the tests we have performed in thermal chamber, it looks like "normal device operation" is not maintained well before the stated 85C ambient temperature.
There are no "heat generators" close to the i210IT, top or bottom side of the PCBa, so the i210IT T(case) and i210IT T(amb) measured and shown in the attached table "packet loss stats vs temperature" shows the i210IT T(case) under bias as well as local i210IT T(amb) without any effect from surrounding components.

Even if we have a properly designed heatsink, is airflow *mandatory* to maintain nominal operating conditions at high temperature?

Thank you for your feedback.

[CarlosAM_INTEL]

Hello, @Jean-Francois:

Thanks for your update.

We have sent an email to the address related to this account with information that may help you.

Best regards,

@CarlosAM_INTEL.