This paper develops a back-propagation neural network (BPNN) and a recurrent neural network (RNN) to predict long-term degradation of CMOS low noise amplifiers due to NBTI and HCI. It thus responds to the challenge of long experimental time caused by low stress voltages. A CMOS low noise amplifier (LNA) is designed and fabricated to test the models. The impacts of NBTI and HCI on the LNA are investigated by the Measure/Stress/Measure (MSM) technique. It is shown that the effects of NBTI and HCI on the LNA are frequency dependent and S-parameters of the LNA are sensitive to the circuit aging. The measured S21, S11, and S22 at 500 MHz are selected as the degradation indicators. The 6-step-ahead, 9-step-ahead, and 12-step-ahead predictions for the LNA aging are developed based on the neural networks. A comparative study of the predicted results obtained from the BPNN and RNN models is carried out to appraise the prediction capability of these models. The results show that the BPNN has a better performance for multi-step-ahead prediction of circuit aging.

中文翻译：

---

使用神经网络对 NBTI 和 HCI 引起的 CMOS 低噪声放大器老化进行多步提前预测

本文开发了一个反向传播神经网络 (BPNN) 和一个循环神经网络 (RNN) 来预测由于 NBTI 和 HCI 引起的 CMOS 低噪声放大器的长期退化。因此，它应对了由低应力电压引起的长实验时间的挑战。设计并制造了一个 CMOS 低噪声放大器 (LNA) 来测试模型。NBTI 和 HCI 对 LNA 的影响通过测量/应力/测量 (MSM) 技术进行研究。结果表明，NBTI 和 HCI 对 LNA 的影响是频率相关的，LNA 的 S 参数对电路老化很敏感。选择在 500 MHz 下测量的 S21、S11 和 S22 作为退化指标。LNA 老化的 6-step-ahead、9-step-ahead 和 12-step-ahead 预测是基于神经网络开发的。对从 BPNN 和 RNN 模型获得的预测结果进行了比较研究，以评估这些模型的预测能力。结果表明，BPNN对于电路老化的多步提前预测具有更好的性能。