Probabilistic Generative Approach for Ambiguity-Aware Parameter Extraction

Artificial neural networks (ANNs) are increasingly used for parameter extraction in semiconductor device modeling. However, in practice, a parameter ambiguity issue arises, where multiple parameter combinations produce identical drain current values (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${I}_{\text {DS}}$ </tex-math></inline-formula>). To address this challenge, we introduce a novel probabilistic generative framework. Specifically, we utilize a conditional variational autoencoder (CVAE) to learn the latent distribution of ambiguous parameters (e.g., threshold voltage and mobility) and generate diverse possible candidate parameter sets conditioned on <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${I}_{\text {DS}}$ </tex-math></inline-formula> characteristics. By validating the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${I}_{\text {DS}}$ </tex-math></inline-formula> characteristics of the candidates, we ultimately select the optimal parameter set. Experiments across different device sizes, process technologies, and operating conditions demonstrate the effectiveness of our model.