Fig. 1.
(Color online) The overall method flow of the proposed FLR optimization model.
DownLoad:
Article Navigation >
Journal of Semiconductors
>
2026
> Accepted Manuscript
| Citation: |
Cheng Liu, Peng Liao, Jianbin Guo, Yanghao Wang, Qingqing Sun, David Wei, Zhang and, Hang Xu. A termination area optimization method for 650 V IGBT based on DNN and MOPSO algorithm[J]. Journal of Semiconductors, 2026, In Press. doi: 10.1088/1674-4926/26040013
****
C Liu, P Liao, J B Guo, Y H Wang, Q Q Sun, D Wei, Z and, and H Xu, A termination area optimization method for 650 V IGBT based on DNN and MOPSO algorithm[J]. J. Semicond., 2026, accepted doi: 10.1088/1674-4926/26040013
|
A termination area optimization method for 650 V IGBT based on DNN and MOPSO algorithm
DOI: 10.1088/1674-4926/26040013
CSTR: 32376.14.1674-4926.26040013
More Information-
Cheng Liu received his Bachelor's degree in Microelectronics Science and Engineering from Jilin University in 2024. He is currently a Master's student at the School of Microelectronics, Fudan University. His research primarily focuses on the design and optimization of power devices assisted by deep learning -
Hang Xu received his bachelor's degree from Northwestern Polytechnical University and his Ph.D. degree from Fudan University in 2024. He is currently a postdoctoral fellow at the School of Microelectronics, Fudan University, Shanghai. His research interests include power semiconductor devices, silicon-based optoelectronic devices, and advanced CMOS process technologies - Corresponding author: hang_xu@fudan.edu.cn
- Revised Date: 2026-10-10 Available Online: 2026-11-13
-
Abstract
Field-limiting-ring (FLR) terminations require the simultaneous optimization of multiple structural parameters, while conventional TCAD-based design is computationally expensive and often involves a trade-off between breakdown voltage (BV) and termination footprint. In this work, an area-oriented multi-objective optimization framework is developed for the FLR termination of a 650 V Si IGBT, in which BV is maximized while the total termination length (Lt) used as a geometric metric of termination-area consumption, is minimized. A deep neural network (DNN) surrogate model is trained using 420 Sentaurus TCAD samples to predict BV from the FLR structural parameters and is subsequently integrated with multi-objective particle swarm optimization (MOPSO) to construct the BV−Lt Pareto front. The DNN achieves R2 = 0.9988, a MAE of 1.5 V, and a RMSE of 1.9 V on the held-out TCAD test set. Three representative Pareto solutions are further verified by full TCAD simulations. Under nearly identical termination-length conditions, Pareto Solution 1 increases the TCAD-verified BV from 950 to 990 V, whereas under comparable BV conditions, Pareto Solution 3 reduces Lt from 290.0 to 201.9 μm, corresponding to a 30.4% reduction, while maintaining a slightly higher BV. In addition, a fabricated device based on Pareto Solution 2 exhibits close agreement between the measured and TCAD-simulated reverse I−V characteristics. These results demonstrate the effectiveness of the proposed TCAD−DNN−MOPSO framework for BV−termination-footprint co-optimization and indicate its potential extensibility to other power-device design problems. -
References
[1] Yang Y H, Dorn-Gomba L, Rodriguez R, et al. Automotive power module packaging: Current status and future trends. IEEE Access, 2020, 8: 160126 doi: 10.1109/ACCESS.2020.3019775[2] Lorenz L. Power semiconductor devices-development trends and system interactions. 2007 Power Conversion Conference-Nagoya, 2007: 348[3] Goreci L S, Popescu M, Tilă I. The choice of the IGBTs and their cooling in electric traction converters for autonomous vehicles. 2021 International Conference on Applied and Theoretical Electricity (ICATE), 2021: 1[4] Navarro D, Pesic I, Morikawa Y, et al. Investigation of 4H−SiC insulated-gate bipolar transistor turn-off performance for achieving low power loss. Jpn J Appl Phys, 2016, 55(4S): 04ER12 doi: 10.7567/JJAP.55.04ER12[5] Binder A T, Steinfeldt J, Allerman A A, et al. Optimization of step-etched junction termination extensions for vertical GaN devices. IEEE Trans Electron Devices, 2024, 71(3): 1541 doi: 10.1109/TED.2023.3344057[6] Huang C Y, Guo Y F, Zhang J, et al. Field plate-adaptive doping: A novel surface electric field optimization technique for SOI LDMOS with gate field plate. IEEE Trans Electron Devices, 2022, 69(1): 291 doi: 10.1109/TED.2021.3130837[7] Onose H, Oikawa S, Yatsuo T, et al. Over 2000 V FLR termination technologies for SiC high voltage devices. 12th International Symposium on Power Semiconductor Devices & ICs. Proceedings. Toulouse, France. IEEE, 2000: 245[8] Yuan J, Cheng Z J, Guo F, et al. A bottom charge-modulated field limiting rings termination with N−P−N sandwich epitaxial wafers for 4H−SiC devices. IEEE Electron Device Lett, 2025, 46(8): 1297 doi: 10.1109/LED.2025.3581587[9] Ji X Q, Yue J Y, Wang J J, et al. Edge-field-suppression for improved breakdown and leakage in vertical β-Ga2O3 Schottky barrier diode through trench field limiting rings. IEEE Trans Electron Devices, 2024, 71(12): 7696 doi: 10.1109/TED.2024.3484356[10] Zhao Y S, Li Z H, Xia Z M, et al. Enhanced filed limiting rings for improving breakdown voltage stability. 2022 IEEE 16th International Conference on Solid-State & Integrated Circuit Technology (ICSICT), 2022: 1[11] Ng M H, Kwok Y, Zhou W J, et al. AI-based general solution for automated TCAD parameter calibration: Case studies on TFT devices. IEEE Trans Electron Devices, 2025, 72(12): 6757 doi: 10.1109/TED.2025.3616083[12] Li J Y, Zhao H Y, Yuan H, et al. An innovative and efficient approach for field-limiting ring design of power devices based on deep neural networks. IEEE Electron Device Lett, 2024, 45(3): 488 doi: 10.1109/LED.2024.3352041[13] Hu L B, Wei X F, Shu L, et al. EMP injection for post-fabricated RON-BV optimization in bulk LDMOS and AI-assisted mechanism analysis. IEEE Electron Device Lett, 2026, 47(3): 586 doi: 10.1109/LED.2025.3649721[14] Yee N, Lu A, Wang Y F, et al. Rapid inverse design of GaN-on-GaN diode with guard ring termination for BV and (VFQ) −1 co-optimization. 2023 35th International Symposium on Power Semiconductor Devices and ICs (ISPSD), 2023: 143[15] Cao Z, Sun Q, Peng Q W, et al. Machine learning-based modeling and BV optimization for LDMOS with multifloating buried layers. IEEE Trans Electron Devices, 2025, 72(2): 550 doi: 10.1109/TED.2024.3517578[16] Luo R D, Sun B T, Hou X L, et al. Optimal design of 100−2000 V 4H−SiC power MOSFETs using multi-objective particle swarm optimization algorithms. IEEE Electron Device Lett, 2024, 45(5): 786 doi: 10.1109/LED.2024.3382004[17] Synopsys, Inc. Sentaurus Process User Guide, Version R-2020.09. Mountain View, CA: Synopsys, Inc. , 2020[18] Carrillo-Nuñez H, Dimitrova N, Asenov A, et al. Machine learning approach for predicting the effect of statistical variability in Si junctionless nanowire transistors. IEEE Electron Device Lett, 2019, 40(9): 1366 doi: 10.1109/LED.2019.2931839[19] Sola J, Sevilla J. Importance of input data normalization for the application of neural networks to complex industrial problems. IEEE Trans Nucl Sci, 1997, 44(3): 1464 doi: 10.1109/23.589532[20] Zhou Y T, Zhang O, Xia X, et al. Application of MOPSO algorithm to optimization of energy efficiency and spectral efficiency trade-off in massive MIMO system. 2022 IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB). Bilbao, Spain. IEEE, 2022: 1 -
Proportional views
