What is the thermal coefficient (NOT thermal resistance) between the top of the SoC case (1) and the Thermal Transfer Plate (TTP, 2)? In this picture below, this is the orange piece plus thermal compound that couples the two, including whatever portion of the case and TTP where Nvidia defines the T.case and T.ttp thermal nodes.
This is from Section 3.1.1 in [Nvidia Xavier AGX Series Thermal Design Guide]:
There is no such fixed coefficient as the heat dissipation is complicated, even with same power consumption, it could be different thermal performance because workload of each part of chip is possible different, that’s why we take thermal resistance as the key element. Thermal design guide has listed the method to calculate thermal resistance. Customer should estimate/get the real power consumption of use case and then calculate the resistance by doing tests.
Understood and thank you for the reply. To restate the question: What is the effective thermal resistance between the case and the TTP (θcp)?
The Thermal Design Document v1.1 states a thermal resistance θjp =0.35 to 0.65C depending on load balance, and we are in the process of dressing such balance. In our application we have also implemented our own thermal solution, which interfaces directly with the SOC case/top rather than Nvidia’s TTP. In our design, the SoC case is an intermediate point along θjp, with θjp = θjc + θcp. We define θcp as a the fixed nominal resistance between the case of the SoC and the TTP. In our design we have a similar thermal interface approach at the SoC and we intend to calculate θjc = θjp - θcp, with the variable aspect of your published θjp associated θjc followed by a fixed θcp.
Can you please define an effective θcp? Where θcp is part of the Nvidia design and not part of our design. For this you can assume that our design spreads the heat along the interface at top of the SoC case equivalently to the Nvidia design.
As far as direct measurements go, it would be quite difficult for us to directly measure the top of the case without disturbing this very critical thermal interface without rendering the measurement invalid. Perhaps you have access to test or simulation data that characterizes θcp, which is rather fixed and perhaps simpler to measure, so we can back it out of the design and de-embed θjc. Alternately, if you do have θjc, or an alternate definition of θjc, please provide it as this would be greatly appreciated.
As said in guide, the θ is a calculated result based on the tested data of target use case, there is no fixed value as it might be different heat dissipation on chip even though the power number is same between two use case. As you can see in guide, there is example θjp based on workload that can be a rough reference to you, which also indicates θ is not a fixed value.
If your thermal design is removing TTP + own solution attached on chip directly, then the key target it to let Tj less than threshold.
Thanks for your prompt reply, I understand that θjp was determined empirically and not a fixed value but the TTP is not really a consumable interface for a new design with a new thermal solution. What I am seeking is the fixed portion of the NVIDIA design that is between the case and the TTP, represented as points 1, 2, and 3 in Figure 3.3 of the thermal guide, which I am calling θcp (case to ttp). I know that much effort went into designing this portion of the thermal solution, and being fixed in nature it lends itself to being defined as a fixed value.
I acknowledge that lumping a complex model into a single value is a bit of an over simplification, where the case temperature may not be uniform, but it could be simplified to a mean or effective temperature for this purpose. With a value for θcp, we can move the interface point back to the case where our thermal solution natively mates.
Could you please pass on this question to the thermal group to see if they do have test data, a model, assumption, or guidance for θcp?
Please refer to thermal DG, that is the full thermal solution for all customers. That is validated for customers who made their own thermal solution.
Seems this is a similar question to another topic, so add the linkage here for reference: Temperature and lifetime data - #9 by Trumany