A New Compact MOSFET Model Based on Artificial Neural Network With Unique Data Preprocessing and Sampling Techniques

A new compact MOSFET model based on artificial neural network (ANN) is developed for analog circuit simulation. MOSFETs with various widths and lengths are fabricated using 0.18- $\mu \text{m}$ analog process. With the novel preprocessing of the measured drain voltage and current data, a high-precision ANN model covering the whole drain current range is developed. By adopting the Latin hypercube sampling algorithm, the amount of data required for training can be significantly reduced without apparently degrading the fitting result. The result shows that the ANN model has a better fitting capability than the BSIM4 model which is the state-of-the-art model in 0.18- $\mu \text{m}$ analog process. For the transconductance and output resistance simulation, the ANN model also has a higher precision. The amplifier and low-dropout regulator (LDO) built by the ANN model are revealed to have good convergence in dc simulation, indicating that the ANN model has a great potential for analog circuits simulation.