Topology-Adaptive Design Automation for Ultra-Wideband CMOS D-Band Antennas

To rapidly and efficiently realize the terahertz (THz) on-chip antenna design with many degrees of freedom, this paper proposes for the first time topology-adaptive design automation (TADA) for ultra-wideband D-band antennas, which are complex and unexplored so far. Specifically, TADA (i) simulates random initial topologies to reduce the complexity of the design search space; (ii) utilizes kernel principal component analysis (KPCA) to reduce the dimensionality of high-dimensional inputs while maintaining nonlinear relationships between parameters to accelerate the optimization process; and (iii)integrates multi-agent techniques with a prioritized experience replay strategy to enhance algorithm convergence and improve performance in data-constrained environments. The TADA framework is experimentally validated on a D-band on-chip antenna design in 65nm CMOS, achieving a bandwidth from 100GHz to 177 GHz (≥77GHz) with a peak gain of ≥2dBi and a radiation efficiency of 62% while maintaining a compact area, providing significant performance advantages over conventional designs. This study represents the first known application of machine learning to the design of THz band antennas, which lays the groundwork for fully automated passive device design in future 6G applications.