MPNet: A modular deep learning process TCAD surrogate modeling framework

Qipei Zhang; Pengwei Liu; Wenzhang Fang; Dong Ni; Yuting Kong

doi:10.1088/1674-4926/25100005

J. Semicond. > Just Accepted

ARTICLES

MPNet: A modular deep learning process TCAD surrogate modeling framework

Qipei Zhang¹, Pengwei Liu², Wenzhang Fang¹, Dong Ni¹ and Yuting Kong^1,

+ Author Affiliations

1. College of Integrated Circuits, Zhejiang University, Hangzhou 311200, China
2. College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, China

Author Bio:

Qipei Zhang is an Eng.D. student in the College of Integrated Circuits at Zhejiang University, beginning his studies in 2023. He received his B.S. degree in Microelectronics Science and Engineering from East China Normal University in 2023. His current research focuses on AI for Integrated Circiuts Manufacturing.

Pengwei Liu is a Ph.D. student in the Department of Control Science and Engineering at Zhejiang University, beginning his studies in 2022. He received his B.S. degree in Mathematics and Applied Mathematics from Hefei University of Technology in 2022. His current research focuses on AI for Scientific Computing.

Wenzhang Fang is a Principal Investigator at the College of Integrated Circuits of Zhejiang University (ZJU) and a researcher at ZJU-Hangzhou Global Scientific and Technological Innovation Center (HIC-ZJU). His main research focuses on integrated circuit manufacturing processes, AI-based integrated circuit manufacturing, and 2D material integrated image sensor chips. He has published over 60 papers in high-level academic journals, which have been cited more than 2,800 times.

Dong Ni is a Professor and Doctoral Advisor at the College of Integrated Circuits, Zhejiang University. He received his Ph.D. from the University of California, Los Angeles (UCLA) in 2005. His primary research interests focus on the application of multi-scale systems and artificial intelligence methods to smart manufacturing in advanced fields, particularly integrated circuits. He has published numerous high-impact papers at top-tier AI conferences, including those selected for spotlight presentations.

Yuting Kong is a ZJU100 Young Professor at the College of Integrated Circuits, Zhejiang University. She got her Ph.D. degree from the College of Control Science and Engineering, Zhejiang University in 2021. Her research interests focus on intelligent manufacturing of integrated circuits and automated design of analog circuits.

Corresponding author: Yuting Kong, ytkong@zju.edu.cn

DOI: 10.1088/1674-4926/25100005CSTR: 32376.14.1674-4926.25100005

Abstract: The computational cost of TCAD simulations is becoming prohibitively high with the complexity of advanced process technologies, making simulation acceleration a critical research priority. While end-to-end surrogate models mapping process recipes to device structures and characteristics offer a promising alternative, their application is often limited by poor generalizability and explainability. In this work, we present MPNet, a modular deep learning surrogate modeling framework for process TCAD. MPNet comprises distinct surrogate models for individual process modules, which are assembled into an integrated framework. These modular models employ a novel UNet-attention feature evolution method to capture the complex evolutions of device geometry and doping profiles. Each module can be trained separately on its individual process, after which the modules are cascaded and jointly fine-tuned to minimize error accumulation throughout the cascade. The efficacy of the proposed MPNet framework is demonstrated through a MOSFET integrated process TCAD case study. Results show that MPNet achieves a computational speedup of over 103 times compared to conventional TCAD, while maintaining predictive fidelity exceeding 98%. Finally, to illustrated the application of the proposed framework, MPNet is coupled with a PSO algorithm, showcasing its utility for fast process optimization to meet specific process targets.

Key words: modular surrogate model, deep learning, process TCAD, PSO

References

[1]	Wu T, Guo J. Multiobjective design of 2-D-material-based field-effect transistors with machine learning methods. IEEE Trans Electron Devices, 2021, 68(11): 5476 doi: 10.1109/TED.2021.3085701
[2]	Koshimoto H, Ishimabushi H, Yoo J, et al. Gummel-cycle algebraic multi-grid pre-conditioning for large-scale device simulations. 2020 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD), 2020: 51
[3]	Myung S, Choi B, Jang W, et al. Comprehensive studies on deep learning applicable to TCAD. Jpn J Appl Phys, 2023, 62: SC0808 doi: 10.35848/1347-4065/acbaa6
[4]	Liu P W, Hao Z K, Ren X Y, et al. Papm: A physics-aware proxy model for process systems. Proc Int Conf Mach Learn, 2024, 235: 31080
[5]	Liu P W, Wang P K, Ren X Y, et al. AeroGTO: An efficient graph-transformer operator for learning large-scale aerodynamics of 3d vehicle geometries. Proc AAAI Conf Artif Intell, 2025, 39(18): 18924 doi: 10.1609/aaai.v39i18.34083
[6]	Wu Q X, Liu P W, Ren X Y, et al. MPG: An efficient multi-scale point-based GNN for non-uniform meshes. Machine Learning and Knowledge Discovery in Databases. Research Track. Cham: Springer Nature Switzerland, 2025: 3
[7]	Thomann S, Novkin R, Li J J, et al. ML-TCAD: Accelerating FeFET reliability analysis using machine learning. IEEE Trans Electron Devices, 2024, 71(1): 213 doi: 10.1109/TED.2023.3336305
[8]	Han S C, Choi J, Hong S M. Acceleration of semiconductor device simu- lation with approximate solutions predicted by trained neural networks. IEEE Trans Electron Devices, 2021, 68(11): 5483 doi: 10.1109/TED.2021.3075192
[9]	Fan G X, Ma T L, Sun X G, et al. Graph attention network-based unified TCAD modeling enabling fast design technology co-optimization through transfer learning. IEEE Trans Electron Devices, 2025, 72(1): 474 doi: 10.1109/TED.2024.3493854
[10]	Choi H C, Yun H, Yoon J S, et al. Neural approach for modeling and optimizing Si-MOSFET manufacturing. IEEE Access, 2020, 8: 159351 doi: 10.1109/ACCESS.2020.3019933
[11]	Guo J M, Geng M Q, Ren K, et al. Optimizing plasma etching: Integrating precise three-dimensional etching simulation and machine learning for multi-objective optimization. IEEE Access, 2024, 12: 127065 doi: 10.1109/ACCESS.2024.3444454
[12]	Lukasiak L, Jakubowski A. The influence of nonuniform doping profile on I-V characteristics of MOS transistors. IEEE Trans Electron Devices, 1993, 40(2): 453 doi: 10.1109/16.182527
[13]	Ghulghazaryan R, Piliposyan D, Wilson J. Application of neural network-based oxide deposition models to CMP modeling. ECS J Solid State Sci Technol, 2019, 8(5): P3154 doi: 10.1149/2.0231905jss
[14]	Liu P W, Wu Q X, Ren X Y, et al. A deep-learning-based surrogate modeling method with application to plasma processing. Chem Eng Res Des, 2024, 211: 299 doi: 10.1016/j.cherd.2024.09.031
[15]	Xiao T Q, Ni D. Multiscale modeling and recurrent neural network based optimization of a plasma etch process. Processes, 2021, 9(1): 151 doi: 10.3390/pr9010151
[16]	Xu H Q, Gan W Z, Cao L, et al. A machine learning approach for optimization of channel geometry and source/drain doping profile of stacked nanosheet transistors. IEEE Trans Electron Devices, 2022, 69(7): 3568 doi: 10.1109/TED.2022.3175708
[17]	Han S C, Choi J, Hong S M. Acceleration of three-dimensional device simulation with the 3D convolutional neural network. 2021 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD). Dallas, TX, USA. IEEE, 2021: 52
[18]	Myung S, Kim J, Jeon Y, et al. Real-Time TCAD: A new paradigm for TCAD in the artificial intelligence era. 2020 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD). Kobe, Japan. IEEE, 2020: 347
[19]	Geng M Q, Guo J M, Sun Y T, et al. Accurate and efficient process modeling and inverse optimization for trench metal oxide semiconductor field effect transistors: A machine learning proxy approach. Processes, 2025, 13(5): 1544 doi: 10.3390/pr13051544
[20]	Everingham M, Van Gool L, Williams C K I, et al. The pascal visual object classes (VOC) challenge. Int J Comput Vis, 2010, 88(2): 303 doi: 10.1007/s11263-009-0275-4
[21]	Thung K H, Raveendran P. A survey of image quality measures. 2009 International Conference for Technical Postgraduates (TECHPOS). Kuala Lumpur, Malaysia. IEEE, 2010: 1
[22]	Wang Z, Bovik A C, Sheikh H R, et al. Image quality assessment: From error visibility to structural similarity. IEEE Trans Image Process, 2004, 13(4): 600 doi: 10.1109/TIP.2003.819861
[23]	Synopsys, Inc. Sentaurus device user guide. Mountain View, CA: Synopsys, Inc., 2022
[24]	Silvaco, Inc. ATLAS user’s manual. Santa Clara, CA: Silvaco, Inc, 2023
[25]	Coventor, Inc. Semulator3D user guide. Cary, NC: Coventor, Inc., 2021
[26]	Isola P, Zhu J, Zhou T, et al. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017: 5967
[27]	Kennedy J, Eberhart R, A new optimizer using particle swarm theory, IEEE Int Symp Micro Mach Human Sci, 1995: 39

Fig. 1. (Color online) The framework of MPNet.

DownLoad: Full-Size Img PowerPoint

Fig. 2. (Color online) The module of MPNet, composed of a UNet and a cross-attention-based feature evolution block.

DownLoad: Full-Size Img PowerPoint

Fig. 3. (Color online) Fine-tuned MPNet for integrated process.

DownLoad: Full-Size Img PowerPoint

Fig. 4. (Color online) The step-by-step prediction of the integrated process.

DownLoad: Full-Size Img PowerPoint

Fig. 5. (Color online) The performance of MPNet using MSE/CE as the loss function.

DownLoad: Full-Size Img PowerPoint

Fig. 6. (Color online) The PSO algorithm repeatedly calls MPNet as a fitness function to find the optimal recipes.

DownLoad: Full-Size Img PowerPoint

Fig. 7. (Color online) Visualization of the case MOSFET integrated process before&after PSO

DownLoad: Full-Size Img PowerPoint

Fig. 8. (Color online) The cost decline curve of PSO. (a) Different number of particles (b) Different w (c) Different combinations of C1 and C2 (d) integrated process and different single steps

DownLoad: Full-Size Img PowerPoint

Fig. 9. (Color online) The distribution of the optimized recipe parameters normalized against the ground truth

DownLoad: Full-Size Img PowerPoint

Fig. 10. (Color online) Visualization of target image, TCAD ground-truth image, and corresponding prediction using PSO optimal recipes (a) process module: Poly etch (b) process module: LDD implant (c)Integrated process

DownLoad: Full-Size Img PowerPoint

Fig. 11. (Color online) Comparison of the I-V test results of the devices obtained from the optimal recipe generated by PSO with the truth from TCAD

DownLoad: Full-Size Img PowerPoint

Table 1. SelectedrecipesandtheirrangesfortheintegratedMOSFETprocesscasestudy.

Process Parameters	Max	Min
Lgate (μm)	0.188	0.172
Slice angle (°)	42.0	48.0
Init boron (cm⁻²)	9.0×10¹⁴	1.2×10¹⁵
Diffuse1 temp (°C)	990	1010
PadOx thickness (A˚)	300	700
Drivein temp (°C)	970	1030
PadNT (s)	0.1	0.3
LOCOS mask thickness (μm)	0.2	0.8
LOCOS etch time (s)	1.6	2.0
LOCOS oxide temp (°C)	880	920
TEOS (A˚)	12	28
STI depth (A˚)	0.3	0.7
STI tilt (°)	70	90
STI rate (A˚/s)	0.013	0.017
STI time (s)	1.95	2.0
Vth dose (cm⁻²)	1.00×10¹²	7.00×10¹²

DownLoad: CSV

Table 2. MPNet process modules results (%)

Step	Description	SSIM	IOU	PSNR
1	PAD deposition	99.6	99.9	—
2	LOCOS etch	99.6	99.8	—
3	LOCOS reoxide	99.4	99.2	—
4	LOCOS oxide	98.8	99.6	—
5	LOCOS oxide etch	99.0	99.2	—
6	STI etch	98.2	98.5	—
7	STI oxide deposition	98.9	99.7	—
8	Vth adjustment implant	97.5	—	34.8
9	Poly deposition	99.8	99.9	—
10	Poly etch	99.6	99.7	—
11	LDD implant	97.8	—	36.7
12	Pocket implant	97.2	—	35.1
13	LDD diffuse	97.4	—	49.9
14	Spacer deposition	99.2	99.7	—
15	Spacer etch	99.4	99.9	—
16	SD implant	98.6	—	40.2
17	SD diffuse	98.8	—	47.7
18	Silicide	99.1	99.6	—

DownLoad: CSV

Table 3. Comparison of evaluation metrics (IOU / SSIM / MSE) under various processes (%).

Process	Method	IOU	SSIM	MSE
Film	RTT^[18]	96.8	96.2	4.2
	Pix2pix^[26]	97.5	98.1	2.8
	DeeplabV3+^[7]	97.2	97.9	3.1
	MPNet	99.1	99.7	1.2
Etch	RTT^[18]	96.5	95.9	4.5
	Pix2pix^[26]	97.8	98.2	2.6
	DeeplabV3+^[7]	97.4	98.0	2.9
	MPNet	99.2	99.4	1.3
Imp	RTT^[18]	/	95.5	4.9
	Pix2pix^[26]	/	97.5	3.2
	DeeplabV3+^[7]	/	97.2	3.5
	MPNet	/	98.8	1.7
Integrated	RTT^[18]	95.6	95.1	4.8
	Pix2pix^[26]	96.9	97.2	3.6
	DeeplabV3+^[7]	97.5	97.8	3.0
	MPNet	98.5	98.7	2.1

DownLoad: CSV

Table 4. Comparison of 50 integrated process runs predict- ing time(s)

Step	TCAD	MPNet
PAD deposition	52	0.0115
LOCOS etch	8	0.0036
LOCOS reoxide	40	0.0112
LOCOS oxide	21	0.0031
LOCOS oxide etch	47	0.0030
STI etch	11	0.0113
STI oxide deposition	57	0.0104
Vth adjustment implant	62	0.0553
Poly deposition	9	0.0223
Poly etch	17	0.0202
LDD implant	18	0.0358
Pocket implant	50	0.0432
LDD diffuse	103	0.0221
Spacer deposition	11	0.0058
Spacer etch	17	0.0056
SD implant	47	0.0233
SD diffuse	127	0.0228
Silicide	42	0.0032
Integrated	749	0.3134

DownLoad: CSV

Table 5. Comparison of parameter quantity before and after parameter sharing (M)

Modules	Model Num	Before	After
Etch	5	8.75	1.76
Film	7	12.25	1.76
Implant	6	10.5	1.78

DownLoad: CSV

Table 6. MPNet’s etch module SSIM accuracy before and after parameter sharing (%)

Models	Before	After
LOCOS-Etch	99.6	99.6
LOCOS-OX-Etch	99.0	98.8
STI-Etch	98.2	97.9
Poly-Etch	99.6	99.4
Spacer-Etch	99.7	99.6

DownLoad: CSV

Table 7. SSIM of ablation studies for different feature evolution methods (%)

Methods	Film	Etch	Implant	Full - Loop
Early	98.4	98.2	96.2	97.2
Late	98.8	98.8	97.8	98.1
Concatenate	98.8	98.5	97.2	98.2
Dot Product	99.2	99.1	97.6	98.6
MPNet	99.4	99.1	98.2	98.7

DownLoad: CSV

[1]	Wu T, Guo J. Multiobjective design of 2-D-material-based field-effect transistors with machine learning methods. IEEE Trans Electron Devices, 2021, 68(11): 5476 doi: 10.1109/TED.2021.3085701
[2]	Koshimoto H, Ishimabushi H, Yoo J, et al. Gummel-cycle algebraic multi-grid pre-conditioning for large-scale device simulations. 2020 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD), 2020: 51
[3]	Myung S, Choi B, Jang W, et al. Comprehensive studies on deep learning applicable to TCAD. Jpn J Appl Phys, 2023, 62: SC0808 doi: 10.35848/1347-4065/acbaa6
[4]	Liu P W, Hao Z K, Ren X Y, et al. Papm: A physics-aware proxy model for process systems. Proc Int Conf Mach Learn, 2024, 235: 31080
[5]	Liu P W, Wang P K, Ren X Y, et al. AeroGTO: An efficient graph-transformer operator for learning large-scale aerodynamics of 3d vehicle geometries. Proc AAAI Conf Artif Intell, 2025, 39(18): 18924 doi: 10.1609/aaai.v39i18.34083
[6]	Wu Q X, Liu P W, Ren X Y, et al. MPG: An efficient multi-scale point-based GNN for non-uniform meshes. Machine Learning and Knowledge Discovery in Databases. Research Track. Cham: Springer Nature Switzerland, 2025: 3
[7]	Thomann S, Novkin R, Li J J, et al. ML-TCAD: Accelerating FeFET reliability analysis using machine learning. IEEE Trans Electron Devices, 2024, 71(1): 213 doi: 10.1109/TED.2023.3336305
[8]	Han S C, Choi J, Hong S M. Acceleration of semiconductor device simu- lation with approximate solutions predicted by trained neural networks. IEEE Trans Electron Devices, 2021, 68(11): 5483 doi: 10.1109/TED.2021.3075192
[9]	Fan G X, Ma T L, Sun X G, et al. Graph attention network-based unified TCAD modeling enabling fast design technology co-optimization through transfer learning. IEEE Trans Electron Devices, 2025, 72(1): 474 doi: 10.1109/TED.2024.3493854
[10]	Choi H C, Yun H, Yoon J S, et al. Neural approach for modeling and optimizing Si-MOSFET manufacturing. IEEE Access, 2020, 8: 159351 doi: 10.1109/ACCESS.2020.3019933
[11]	Guo J M, Geng M Q, Ren K, et al. Optimizing plasma etching: Integrating precise three-dimensional etching simulation and machine learning for multi-objective optimization. IEEE Access, 2024, 12: 127065 doi: 10.1109/ACCESS.2024.3444454
[12]	Lukasiak L, Jakubowski A. The influence of nonuniform doping profile on I-V characteristics of MOS transistors. IEEE Trans Electron Devices, 1993, 40(2): 453 doi: 10.1109/16.182527
[13]	Ghulghazaryan R, Piliposyan D, Wilson J. Application of neural network-based oxide deposition models to CMP modeling. ECS J Solid State Sci Technol, 2019, 8(5): P3154 doi: 10.1149/2.0231905jss
[14]	Liu P W, Wu Q X, Ren X Y, et al. A deep-learning-based surrogate modeling method with application to plasma processing. Chem Eng Res Des, 2024, 211: 299 doi: 10.1016/j.cherd.2024.09.031
[15]	Xiao T Q, Ni D. Multiscale modeling and recurrent neural network based optimization of a plasma etch process. Processes, 2021, 9(1): 151 doi: 10.3390/pr9010151
[16]	Xu H Q, Gan W Z, Cao L, et al. A machine learning approach for optimization of channel geometry and source/drain doping profile of stacked nanosheet transistors. IEEE Trans Electron Devices, 2022, 69(7): 3568 doi: 10.1109/TED.2022.3175708
[17]	Han S C, Choi J, Hong S M. Acceleration of three-dimensional device simulation with the 3D convolutional neural network. 2021 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD). Dallas, TX, USA. IEEE, 2021: 52
[18]	Myung S, Kim J, Jeon Y, et al. Real-Time TCAD: A new paradigm for TCAD in the artificial intelligence era. 2020 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD). Kobe, Japan. IEEE, 2020: 347
[19]	Geng M Q, Guo J M, Sun Y T, et al. Accurate and efficient process modeling and inverse optimization for trench metal oxide semiconductor field effect transistors: A machine learning proxy approach. Processes, 2025, 13(5): 1544 doi: 10.3390/pr13051544
[20]	Everingham M, Van Gool L, Williams C K I, et al. The pascal visual object classes (VOC) challenge. Int J Comput Vis, 2010, 88(2): 303 doi: 10.1007/s11263-009-0275-4
[21]	Thung K H, Raveendran P. A survey of image quality measures. 2009 International Conference for Technical Postgraduates (TECHPOS). Kuala Lumpur, Malaysia. IEEE, 2010: 1
[22]	Wang Z, Bovik A C, Sheikh H R, et al. Image quality assessment: From error visibility to structural similarity. IEEE Trans Image Process, 2004, 13(4): 600 doi: 10.1109/TIP.2003.819861
[23]	Synopsys, Inc. Sentaurus device user guide. Mountain View, CA: Synopsys, Inc., 2022
[24]	Silvaco, Inc. ATLAS user’s manual. Santa Clara, CA: Silvaco, Inc, 2023
[25]	Coventor, Inc. Semulator3D user guide. Cary, NC: Coventor, Inc., 2021
[26]	Isola P, Zhu J, Zhou T, et al. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017: 5967
[27]	Kennedy J, Eberhart R, A new optimizer using particle swarm theory, IEEE Int Symp Micro Mach Human Sci, 1995: 39

Search

GET CITATION

shu

Export: BibTex EndNote

Article Metrics

Article views: 22 Times PDF downloads: 4 Times Cited by: 0 Times

History

Received: Revised: Online: Accepted Manuscript: 17 March 2026

Article Navigation > Journal of Semiconductors > 2026 > Accepted Manuscript

Qipei Zhang, Pengwei Liu, Wenzhang Fang, Dong Ni, Yuting Kong. MPNet: A modular deep learning process TCAD surrogate modeling framework[J]. Journal of Semiconductors, 2026, In Press. doi: 10.1088/1674-4926/25100005 ****Q Zhang, P Liu, W Fang, D Ni, and Y Kong, MPNet: A modular deep learning process TCAD surrogate modeling framework[J]. J. Semicond., 2026, accepted doi: 10.1088/1674-4926/25100005

Citation:

Qipei Zhang, Pengwei Liu, Wenzhang Fang, Dong Ni, Yuting Kong. MPNet: A modular deep learning process TCAD surrogate modeling framework[J]. Journal of Semiconductors, 2026, In Press. doi: 10.1088/1674-4926/25100005 ****

Q Zhang, P Liu, W Fang, D Ni, and Y Kong, MPNet: A modular deep learning process TCAD surrogate modeling framework[J]. J. Semicond., 2026, accepted doi: 10.1088/1674-4926/25100005

Citation:

Q Zhang, P Liu, W Fang, D Ni, and Y Kong, MPNet: A modular deep learning process TCAD surrogate modeling framework[J]. J. Semicond., 2026, accepted doi: 10.1088/1674-4926/25100005

PDF( 9897 KB)

MPNet: A modular deep learning process TCAD surrogate modeling framework

DOI: 10.1088/1674-4926/25100005

CSTR: 32376.14.1674-4926.25100005

1.
College of Integrated Circuits, Zhejiang University, Hangzhou 311200, China
2.
College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, China

More Information

Qipei Zhang is an Eng.D. student in the College of Integrated Circuits at Zhejiang University, beginning his studies in 2023. He received his B.S. degree in Microelectronics Science and Engineering from East China Normal University in 2023. His current research focuses on AI for Integrated Circiuts Manufacturing
Pengwei Liu is a Ph.D. student in the Department of Control Science and Engineering at Zhejiang University, beginning his studies in 2022. He received his B.S. degree in Mathematics and Applied Mathematics from Hefei University of Technology in 2022. His current research focuses on AI for Scientific Computing
Wenzhang Fang is a Principal Investigator at the College of Integrated Circuits of Zhejiang University (ZJU) and a researcher at ZJU-Hangzhou Global Scientific and Technological Innovation Center (HIC-ZJU). His main research focuses on integrated circuit manufacturing processes, AI-based integrated circuit manufacturing, and 2D material integrated image sensor chips. He has published over 60 papers in high-level academic journals, which have been cited more than 2,800 times
Dong Ni is a Professor and Doctoral Advisor at the College of Integrated Circuits, Zhejiang University. He received his Ph.D. from the University of California, Los Angeles (UCLA) in 2005. His primary research interests focus on the application of multi-scale systems and artificial intelligence methods to smart manufacturing in advanced fields, particularly integrated circuits. He has published numerous high-impact papers at top-tier AI conferences, including those selected for spotlight presentations
Yuting Kong is a ZJU100 Young Professor at the College of Integrated Circuits, Zhejiang University. She got her Ph.D. degree from the College of Control Science and Engineering, Zhejiang University in 2021. Her research interests focus on intelligent manufacturing of integrated circuits and automated design of analog circuits
Corresponding author: ytkong@zju.edu.cn
Available Online: 2026-03-17

Abstract

Abstract

The computational cost of TCAD simulations is becoming prohibitively high with the complexity of advanced process technologies, making simulation acceleration a critical research priority. While end-to-end surrogate models mapping process recipes to device structures and characteristics offer a promising alternative, their application is often limited by poor generalizability and explainability. In this work, we present MPNet, a modular deep learning surrogate modeling framework for process TCAD. MPNet comprises distinct surrogate models for individual process modules, which are assembled into an integrated framework. These modular models employ a novel UNet-attention feature evolution method to capture the complex evolutions of device geometry and doping profiles. Each module can be trained separately on its individual process, after which the modules are cascaded and jointly fine-tuned to minimize error accumulation throughout the cascade. The efficacy of the proposed MPNet framework is demonstrated through a MOSFET integrated process TCAD case study. Results show that MPNet achieves a computational speedup of over 103 times compared to conventional TCAD, while maintaining predictive fidelity exceeding 98%. Finally, to illustrated the application of the proposed framework, MPNet is coupled with a PSO algorithm, showcasing its utility for fast process optimization to meet specific process targets.
- modular surrogate model,
- deep learning,
- process TCAD,
- PSO

FullText(HTML)

References(27)