In this work, a methodology for developing a memory compact model is described using recurrent neural network (RNN). Compared to traditional modeling approaches, it is flexible and able to develop an accurate model based on physical data before the material physics is fully understood. A simple ReRAM-type memory model is demonstrated using Nonlinear Autoregressive with External Input (NARX) machine learning approach. To enable the neural network to capture the resistive switching characteristics of ReRAM, the output at a range of cycling voltages was captured and used as training data. The trained model is used for DC prediction under different voltage amplitudes, and the prediction results are consistent with the physical data, with MSE values below 10-9. The accuracy of the RNN-assisted ReRAM model during the read, write and erase operations is also evaluated to show the validity of the approach.

This letter proposes a machine learning (ML)-based method for statistical device modeling from a data-oriented perspective. A mapping from the process variation to the selected variational neurons is derived by the proposed algorithm and backward propagation of variation method (BPV) to capture the non-homogeneous effects of model parameters. The same size as the nominal ML model is used, with fewer variation parameters rather than numerous physical parameters, in favor of circuit simulation speed. In addition, a secondary classification of the selected variational neurons is applied to model the correlation between n- and p-type devices. The ML-based statistical modeling framework has been well implemented and verified on the GAA devices and circuits, which indicates its great potential in modeling of emerging device technology.

A non-quasi-static model for ferroelectric capacitance is developed in this letter. A state transition in the voltage and time domains between two polarization states is formulated first. The quasi-static model is derived from the state transition of voltage domain, and supports the minor loops. Different from the Preisach model, an initial state is supported, and the modulated coercive voltages are responsible for minor loops. The non-quasi -static model is then derived with the state transition in the time domain, similar to a relaxation approximation in MOSFET modeling. The non-quasi-static model reproduces the saturation loop, minor loops, the frequency-dependent characteristics of measured ferroelectric capacitances, with their origins explained from polarization switching relaxation. The pulse width dependent switching is well reproduced with the model.