Thermal behaviour, power estimation

Hi there!

I'm currently optimizing my system (using an EP3SL50F780C2N Stratix III device) for minimal power consumption. To compare the PowerPlay power estimation with real life, I not only measure board currents, but temperatures as well. To take away the influence of the FPGA as far as possible, I have a minimal (not clocked) configuration, that just defines the required (output) pins in a good way, which powerplay estimates ~490 mW for. For temperature measurement, I use the FPGA diode to measure the die temperature and an external temperature sensor mounted on the FPGA (center of case). Doing so, I faced some odd behaviour, resulting in following questions:

1. Junction temperature and case temperature in a settled state differ as much as 7-8 degrees C. With a junction-to-case resistance of 0.3C/W this will never fit the total electrical use of <1300 mW for the whole board. What could be the reason? Might there be a heat source just beside the temperature diode on the die? If yes, what (bad surrounding circuitry) could lead to the behaviour?

2. In the same mentioned settled state, I measure an FPGA case temperature of 36.6C at an ambient temperature of 25.0C. With the case-to-ambient resistance of 10.6C/W used by power play this would result in a thermal power of 1140 mW. This is by far higher than the estimated FPGA power of ~500mW. Even if considering additional heating up by adjacent components due to good thermal coupling through power/gnd plates (DDR, power sources), it is quite high: 1140mW radiated by FPGA, where 1330mW electrically supported to the whole board. Seems quite odd to me. Any hint what could go wrong here?

3. To have relevant (relative) temperature measurements, I evaluated the inernal temperature diode in parallel with the external one mounted on the FPGA case. FPGA was otherwise unpowered. Two boards gave similar results. Are there electrical characteristics and temperature calculation formulas available for this diode? I did'nt fînd any so far...

Kind regards,

Peter