No abstract

Traditional TCAD simulation has succeeded in predicting and optimizing the device performance; however, it still faces a massive challenge -a high computational cost. There have been many attempts to replace TCAD with deep learning, but it has not yet been completely replaced. This paper presents a novel algorithm restructuring the traditional TCAD system. The proposed algorithmpredicts three-dimensional (3-D) TCAD simulation in real-time while capturing a variance, enables deep learning and TCAD to complement each other, and fully resolves convergence errors.

TCAD simulation has incessantly solved many complex problems, but it becomes demanding that alternatives be found because TCAD simulation cannot provide the precise and fast prediction in the nano-scale era. With the success story on deep learning in research area, many big data companies have attempted to introduce deep learning to support or replace TCAD simulation. The reason is deep learning models has great potential that solves the problems of the TCAD simulation in terms of execution time, coverage. This paper aims to describe various scenarios of deep learning applicable to TCAD. We firstly describe an application that supplies TCAD data to the deep learning model although TCAD simulation is not calibrated. We then review various approaches that mimic TCAD simulation itself. We finally introduce an application that deep learning model automatically calibrates TCAD models to the measurement without experts. In each scenario, we review the related papers and compare pros and cons.

No abstract

Inductive transfer learning aims to learn from a small amount of training data for the target task by utilizing a pre-trained model from the source task. Most strategies that involve large-scale deep learning models adopt initialization with the pretrained model and fine-tuning for the target task. However, when using over-parameterized models, we can often prune the model without sacrificing the accuracy of the source task. This motivates us to adopt model pruning for transfer learning with deep learning models. In this paper, we propose PAC-Net, a simple yet effective approach for transfer learning based on pruning. PAC-Net consists of three steps: Prune, Allocate, and Calibrate (PAC). The main idea behind these steps is to identify essential weights for the source task, fine-tune on the source task by updating the essential weights, and then calibrate on the target task by updating the remaining redundant weights. Under the various and extensive set of inductive transfer learning experiments, we show that our method achieves state-of-the-art performance by a large margin.