Neural Electromagnetic Topology Optimization for Sub-6 GHz On-Chip Antennas
DOI:
https://doi.org/10.31838/NJAP/07.03.20Keywords:
On-chip antenna, sub-6 GHz, neural topology optimization, electromagnetic co-design, CMOS RF integration, machine learning in EM design, impedance matchingAbstract
On-chip antenna (OCA) design under 6 GHz has very important issues associated with the radiation efficiency level, substrate loss, and mutual coupling that badly impair the signal integrity of miniature RF systems. Conventional full-wave electromagnetic (EM) solvers and manual tuning of the topology is very taxing in computation and has no scalability to dynamic process changes. The current paper suggests a Neural Electromagnetic Topology Optimization (NETO) model that utilizes the deep neural networks with physics-based EM constraints to design effective on-chip antenna designs at the sub-6 GHz frequency regime. The model acquires parametric EM simulation of the spatial field distribution, material interaction, and impedance properties, which allows quick topology prediction with minimal subsequent correction of the process. The architecture incorporates differentiable Maxwell operators into a neural optimization cycle and reduced up to 40 times in terms of design time than iterative finite-element (FEM) algorithms. With gain of 5.8 dBi, radiation efficiency of 89 percent and impedance matching better at 3.3 to 5.8 GHz, the optimised antenna can be used in the 3.3-5.8 GHz range. Strong consistency between simulated and measured S-parameters is experimentally validated on a 45 nm CMOS substrate which indicates the physical consistency of the neural model. The suggested NETO architecture is an increaseable and extensible design of electromagnetic-conscious neural co-design next-generation RF-SoC integration.
References
1. Zhao, X. (2021). Physics-informed neural models for microwave device optimization. IEEE Transactions on Microwave Theory and Techniques, 69(10), 4221–4235.
2. Park, D. (2022). Deep generative EM modeling for sub-6 GHz systems. IEEE Access, 10, 98211–98222.
3. Nguyen, T. (2021). Compact CMOS on-chip antennas for 5 G applications. IEEE Transactions on Antennas and Propagation, 69(12), 10234–10242.
4. Kim, L. (2022). Topology learning for electromagnetic field synthesis. Nature Communications, 13(7745), 1–10.
5. Singh, R. (2022). Surrogate learning for multi-band antenna design. IEEE Transactions on Antennas and Propagation, 70(8), 6923–6935.
6. Wang, P. (2022). Co-optimization of EM and ML models for antenna arrays. IEEE Transactions on Computer
Aided Design of Integrated Circuits and Systems, 41(5), 1247–1258.
7. Chen, J. (2022). Physics-aware neural networks for EM inversion. IEEE Transactions on Geoscience and Remote Sensing, 60, 561–573.
8. Han, Y. (2022). Differentiable EM solvers for fast antenna design. IEEE Microwave and Wireless Components Letters,
32(7), 485–488.
9. Ito, K. (2022). EM field-driven design automation using deep networks. IEEE Transactions on Electromagnetic Compatibility, 64(3), 703–711.
10. Bose, N. (2023). Hybrid EM–DL co-simulation frameworks. Microelectronics Journal, 122, 105284.
11. Choi, J. (2023). AI-driven topology exploration for low loss interconnects. IEEE Transactions on Very Large Scale
Integration Systems, 31(2), 341–353.
12. Zhang, L. (2023). Physics-informed neural field solvers for microstrip networks. IEEE Access, 11, 54213–54226.
13. Patel, D. (2023). Differentiable EM learning for CMOS antennas. IEEE Sensors Journal, 23(5), 4568–4579.
14. Kumar, V. (2023). End-to-end learning for RF structure prediction. IEEE Transactions on Microwave Theory and Techniques, 71(6), 3455–3465.
15. Rossi, F. (2023). Neural field optimization for broad band antennas. IEEE Transactions on Antennas and Propagation, 71(8), 8765–8775.
16. Singh, A. (2023). Topology-driven design of RF front ends. IEEE Transactions on Circuits and Systems I, 70(4), 1654–1663.
17. Patel, G. (2024). Differentiable neural solvers for EM co-design. IEEE Access, 12, 66421–66433.
18. Lee, M. (2024). Sub-6 GHz antenna miniaturization via topology learning. IEEE Transactions on Antennas and Propagation, 72(2), 1343–1354.
19. Yadav, S. (2024). AI-assisted synthesis of on-chip passive components. Microelectronic Engineering, 257, 111821.
20. Das, R. Neural optimization of EM field topology in SoC antennas. IEEE Transactions on Microwave Theory and Techniques, 73.
