This paper presents the comparison of simulation and experimental output with the neural network results of different materials like Si and GaAs-based IMPATT diode at 94 GHz window frequency. The experimental results are nearly equal to simulation results and show a quantitative agreement between theoretical values and experimental output. The experimental results in terms of efficiency and RF power are almost 80-85% of the simulation results. Rest has 15-20% discrepancy between theory and experiment due to several causes like temperature, small-signal approach, parasitic effects, impedance matching problem and proper heat-sink arrangements. The developed neural network model is used to optimise performance of the diode by taking other wide band gap semiconductor material-based IMPATT diode. The model and the results will be very useful for applications in radar and guided missile technology.