Deep Learning-Based Fast BSIM-CMG Parameter Extraction for General Input Dataset

Ashai, Aasim; Jadhav, Aakash; Behera, Amit Kumar; Roy, Sourajeet; DasGupta, Anirvan; Sarkar, Biplab

doi:10.1109/ted.2023.3278615

Cited by 6 publications

(2 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Zhang et al (2020) tackle the issue of wear and tear effects and process variations in FinFET SRAM cells by integrating deep neural networks (DNNs) and evolutionary algorithms (EAs) (2) . Ashai et al (2023) tackle the challenge of extracting parameters for the BSIMCMG semiconductor device model using a deep learning architecture that combines a convolutional neural network (CNN) and GPT-3.5 (3) . Wang et al (2022) aid in the contrary design of materials with specific characteristics by employing High Throughput Virtual Screening (HTVS), Global Optimization (GO), and Generative Models (GM) (4) .…”

Section: Fig 1 Workflow Of the Projectmentioning

confidence: 99%

See 1 more Smart Citation

Machine Learning-based Inverse Design Model of a Transistor

Sahoo,

Patel

2024

IJST

View full text Add to dashboard Cite

Objectives: To develop an inverse design model for transistors, utilizing machine learning algorithms to predict key design parameters specifically, the length and width based on specified gain and bandwidth requirements. And to conduct a comprehensive comparative analysis with existing literature, evaluating the efficacy and novelty of the proposed model in the context of semiconductor engineering challenges and methodologies. Methods: The comprehensive dataset, comprising 30,000 values generated through LTspice simulations, forms the basis for training the machine learning model. Utilizing a Random Forest regressor as the base model and a multi-output regressor as the main model, the project involves extensive data analysis, model development, and iterative fine-tuning. Findings: The outcomes demonstrate the efficacy of the developed model in accurately predicting transistor dimensions. Performance metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-squared, highlight the precision of the model in fulfilling the specified objectives. Novelty: This study introduces a novel approach to semiconductor device design optimization, showcasing the potential of machine learning to streamline the inverse design process. The use of a multi-output regressor, feature engineering, and fine-tuning through log transformation contribute to the innovative nature of the developed model. Keywords: Machine Learning (ML) model, Random Forest regressor, multioutput regressor, Feature engineering, Finetuning

show abstract

Section: Fig 1 Workflow Of the Projectmentioning

confidence: 99%

“…Ashai et al (2023) introduced a deep learning architecture combining CNN and GPT-3.5 for parameter extraction in the BSIMCMG semiconductor device model (3) . Although their focus differs, our study echoes the spirit of innovation by employing machine learning to address transistor optimization challenges.…”

Section: Fine-tuningmentioning

confidence: 99%