For thermal simulations it is important to consider the boundary conditions. In particular you may need to run two simulations. The first representing a measured device, for calibration, and a second representing the device in the expected operating environment. If multiple operating environments are expected, the number of simulations may go up, if needed.
For measurement and calibration, the typical device is usually isolated, although not always. If you consider a typical test structure, on a typical test chip, the scenario is usually a lone device operating in a sea of silicon.
In operation the device being simulated may be isolated, it may be sitting next to a large power device, or what you simulated may be part of a larger device.
Let’s use two examples to look at this closer. The first device could be any typical device, e.g. a nmos transistor, a diode, a bjt, pmos, resistor, etc.You have the device, as part of a test chip, on a wafer, as the sole source of heat. The opposite of this is the same device, but on a product die, sitting in the middle of an array of other operating (heat generating) devices. In the first case most likely the appropriate thermal boundary condition is a Dirichlet BC. For all intents your device sits in an infinite sink. In the second case the boundary condition should be a reflecting, or Nuemann BC. If you are simulating a single device operating in an arrayof active devices.