Figure 1.

Schematic of the diffusion and retrodictive diffusion processes, and the overall deep learning structure. (a) The image shows the thermal evolution of a chip under load captured by a thermal imaging camera. The right image depicts the thermal map evolution of the chip during thermal diffusion. (b) The thermal evolution data of the final state is fed into the TRO network to retrodict the initial thermal field. (c) Using mesh information, the thermal diffusivity is estimated based on the thermal field evolution data, as shown in the TE-Net diagram. The inferred thermal field is then compared with the actual field, and the obtained thermal diffusivity is subsequently fed into the TRO to retrodict the initial temperature field.