Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PRIME: Physics-Related Intelligent Mixture of Experts for Transistor Characteristics Prediction

Zhenxing Dou, Yijiao Wang, Tao Zou, Zhiwei Chen, Fei Liu, Peng Wang, Weisheng Zhao

TL;DR

This work tackles the difficulty of accurately predicting transistor I-V characteristics across multiple operating regions by introducing PRIME, a physics-related intelligent mixture of experts. PRIME replaces a static, single-model approach with a three-expert mixture whose gating network dynamically assigns region-specific importance, yielding substantial improvements (60%–84%) over state-of-the-art models on gate-all-around transistors. By incorporating high-order derivative regularization and leveraging TCAD-derived data, PRIME demonstrates robust generalization across shapes and holds promise for more accurate and efficient circuit simulation via SPICE-like analysis. The approach bridges physics knowledge and data-driven learning to deliver precise, region-aware transistor modeling suitable for design and fabrication workflows.

Abstract

In recent years, machine learning has been extensively applied to data prediction during process ramp-up, with a particular focus on transistor characteristics for circuit design and manufacture. However, capturing the nonlinear current response across multiple operating regions remains a challenge for neural networks. To address such challenge, a novel machine learning framework, PRIME (Physics-Related Intelligent Mixture of Experts), is proposed to capture and integrate complex regional characteristics. In essence, our framework incorporates physics-based knowledge with data-driven intelligence. By leveraging a dynamic weighting mechanism in its gating network, PRIME adaptively activates the suitable expert model based on distinct input data features. Extensive evaluations are conducted on various gate-all-around (GAA) structures to examine the effectiveness of PRIME and considerable improvements (60\%-84\%) in prediction accuracy are shown over state-of-the-art models.

PRIME: Physics-Related Intelligent Mixture of Experts for Transistor Characteristics Prediction

TL;DR

This work tackles the difficulty of accurately predicting transistor I-V characteristics across multiple operating regions by introducing PRIME, a physics-related intelligent mixture of experts. PRIME replaces a static, single-model approach with a three-expert mixture whose gating network dynamically assigns region-specific importance, yielding substantial improvements (60%–84%) over state-of-the-art models on gate-all-around transistors. By incorporating high-order derivative regularization and leveraging TCAD-derived data, PRIME demonstrates robust generalization across shapes and holds promise for more accurate and efficient circuit simulation via SPICE-like analysis. The approach bridges physics knowledge and data-driven learning to deliver precise, region-aware transistor modeling suitable for design and fabrication workflows.

Abstract

In recent years, machine learning has been extensively applied to data prediction during process ramp-up, with a particular focus on transistor characteristics for circuit design and manufacture. However, capturing the nonlinear current response across multiple operating regions remains a challenge for neural networks. To address such challenge, a novel machine learning framework, PRIME (Physics-Related Intelligent Mixture of Experts), is proposed to capture and integrate complex regional characteristics. In essence, our framework incorporates physics-based knowledge with data-driven intelligence. By leveraging a dynamic weighting mechanism in its gating network, PRIME adaptively activates the suitable expert model based on distinct input data features. Extensive evaluations are conducted on various gate-all-around (GAA) structures to examine the effectiveness of PRIME and considerable improvements (60\%-84\%) in prediction accuracy are shown over state-of-the-art models.
Paper Structure (14 sections, 5 equations, 6 figures, 2 tables)

This paper contains 14 sections, 5 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. PROBLEM DEFINITION AND DEVICE INTRODUCTION
  3. Problem Definition
  4. Device Introduction
  5. PROPOSED PRIME METHOD
  6. Physical Property Classification Neural Network
  7. Physics-Related Intelligent Mixture of Experts
  8. Loss Function
  9. RESULTS AND DISCUSSION
  10. Experimental Setup
  11. Model Parameters and Evaluation Metrics
  12. Performance Evaluation
  13. Physical Applications
  14. CONCLUSION

Figures (6)

  • Figure 1: Calibration of the transfer characteristics ($I_{\mathrm ds}-V_{\mathrm gs}$) of the GAA nanowire transistor with the experimental data.
  • Figure 2: Comparison of MLP, PPC-Net and PRIME. In Figures (a)-(c), black arrows indicate data flow. In Figure (c),the multiplication sign in the circle represents multiplication, and the double solid lines represent the gating network output weights.
  • Figure 3: GAA transistor structure and three types of cross-section view of GAA transistors. (a) Entire 3D schematic of the device structure. (b) Cross section of entire transistor. (c) Three types of cross sectional shapes: circular, triangular, and rectangular.
  • Figure 4: The predicted current characteristics of GAA transistors with triangular, rectangular, and circular cross sections by the PRIME model. The hollow circles represent TCAD data, while the solid lines represent the prediction results of the PRIME model.
  • Figure 5: Inverter simulation results.
  • ...and 1 more figures