Ferroelectricity-induced negative capacitance in microelectronic devices via phase-field simulations and machine learning

FerroX, a massively parallel 3D phase-field simulation code for modeling ferroelectric materials, was used to study the influence of the thicknesses of the insulator and ferroelectric layers in a metal-ferroelectric-insulator-semiconductor-metal (MFISM) device. Different combinations of the parameters that describe the polarization dynamics of the ferroelectric material were investigated, confirming the existence of negative capacitance. Gaussian process regression was applied to the data, showing that the dielectric thickness and the second-order Landau free energy parameter are the most important design parameters. Neural network models were trained on the data and used to optimize the device geometry and the materials properties. Details about the 3D visualization of the ferroelectric domain walls and other physical quantities will be discussed.