Digital Twin Method Of High-Power Semiconductor Chips Based On SciML

In our country,where digital twin technology is still in its early stages of development,the digital twins are still known as visual perceptual twins.These twins effectively map virtual reality to reality visually,but they are lack of internal physical laws,it is challenging to ensure its accuracy of the mapping.This thesis proposes a new generation of digital twin technology for large-scale power semiconductor chips,which is based on digital twins for multi-physics simulations.The new generation of Digital twins are a set of computational models,evolving over time to persistently represent the structure,behavior,and context of the corresponding physical asset,from material,component,chip,board and to system.Digital twins also affect and guide the behavior of real physical models through data assimilation and decision iterative optimization.The digital twin model for the semiconductor chip will be created using three steps: high-precision multi-physical field simulation,scientific machine learning driving model reduction,and data assimilation.We will use a power chip as the twin object.The precise work contents to realise the digital twin construction of advanced chips are as follows:(1)High-precision geometric modelling based on commercial software and high-confidence multi-physical field simulation of the chip are performed to realise the twin of visual and physical laws;(2)The advanced scientific machine learning algorithm is introduced to integrate the physical equation with the data science,establish the reduced order model and realise the real-time reality mapping;(3)To achieve information physical fusion,adjust information space parameters in realtime,lower twin model uncertainty and increase mapping accuracy,a data assimilation algorithm is presented.The challenge of the latest generation of digital twin simulations is the simultaneous calculation of complex physical fields.We can now handle the multi-physical field simulation of large-scale 3D semiconductor chips with the use of scientific machine learning method,and we can even map digital and physical space in real time.The development of digital twins can also be made more flexible by the non-intrusive order reduction method,which can be accomplished through physical simulation and can also learn from sensor data.