I was able to reproduce your error based on the model you posted. In this case there are some changes to the model settings (solver and Steady-State Analysis) that are required for this to converge at the proper temperatures.
First, you should be first tuning the Relative Tolerance setting to a smaller value and not the Max Step Size, to improve speed and accuracy. The step size setting only limits the largest step the solver can take, but not influences its performance in between any two steps, whereas the tolerance setting instructs it how to check and determine accuracy of each step. I have changed these from what you had of Max step size = 1e-7 and Relative tolerance = 1e-3 (the default value) to Max step size = 1e-3 (back to the default value) and Relative tolerance = 1e-5, as shown below:
Further, because the initial thermal states (temperature) are so far from the final values, unless you use initial temperature values closer to the final ones (than 50C like you have now), you need to add significantly more Initial cycles. I found that 200 worked well:
With these changes to the model the Steady-State Analysis should now work as you expect, but I attached my version of your model for reference.
Let me know if you have any questions.