With this stress, I noticed that the bare sensor was reporting an increasing temperature in the first minute of operation: it got into a "warm-up" phase which caused a wrong reading of the actual temperature. Almost as high as +1°C.
Well, the device measures its own die temperature, so if it is stressed and left in open-air without a heatsink it probably does heat up. Power consumption according to the datasheet is 5V*9mA = 45mW. Room temperature was lightly above 25°C.
In order to better understand this useful tiny device I will try and report three approaches:
- keep the 12-bit resolution and add a small heatsink
- decrease resolution to 9-bit and keep the duty cycle
- at 9-bit resolution keep the 2" loop