Calibration of Air Electrostatic Discharge Circuit Model with Long Short-Term Memory Network

Air-electrostatic discharge (Air-ESD) has a significant impact on the performance of electronic devices and components. Simulating the effects of air-ESD and performing electrostatic discharge testing are essential for evaluating this impact. However, existing tests are influenced by environmental factors, and the discharge effects vary across different testing methods, which results in low feasibility of the simulated ESD behavior under specific discharge conditions. In this study, we propose a calibratable air-ESD system to minimize the influence of testing environments and methods, which yields a consistent air discharge currents. In this system, a calibratable circuit model is established, including the ESD generator, discharge arc, and measurement setup sub circuit. Two key parameters will be tuned to reveal the impact of environmental factors and testing methods, which are the arc length in discharge arc model and the inductance in measurement sub circuit. Based on this model, an ESD behavior dataset is developed, containing circuit models covering common testing scenario and the corresponding discharge currents. To infer key parameters of the circuit model, we propose a Long Short-Term Memory (LSTM) based neural network as a calibration model that captures the time-domain characteristics of the current. Experimental results demonstrate that the prediction error of the calibration model is less than 0.95 %. In real-world experiments, the calibration model successfully reproduced the ESD current waveform, with the reproduced waveform closely matching the actual waveform. This provides a reliable reference for system-level air ESD modeling.