Skip navigation







FIELD AI and Natural Sciences
DATE November 17 (Wed), 2021
TIME 14:00-16:00
PLACE Online
SPEAKER Soogine Chong
HOST Hyeon, Changbong
TITLE Neural compact models - Use of neural networks in semiconductor device modeling
ABSTRACT Transistors, semiconductor devices that turn on and off depending on the input signal, are the main building blocks of semiconductor chips. In order to design a chip, the characteristics of transistors are modeled in the form of computationally efficient “compact models,” so that the operation of a chip can be simulated before hardware implementation. Nowadays, device compact models are mostly physics-based analytical models, which require thorough understanding of the physics involved for their development. Data-based neural network models, or “neural compact models,” have the potential to overcome the limitations of the current method to allow faster and more accurate device modeling. In this talk, I will first give an introduction to semiconductor compact modeling. Then, I will discuss the limitations of the conventional analytical models, followed by benefits of replacing these models by neural compact models. I will also propose a few approaches that can be used to increase the accuracy of the model even when the amount of data is limited. Finally, I will share my view on the future that I foresee with the adoption of neural compact models in the semiconductor industry.
  • list