Deep Learning enabled mmWave PA and Antenna Design

Future mmWave and THz wireless chip-scale systems need to be incorporate complex functionalities including ability to operate over multiple spectral bands in an agile fashion spread across 30-100+ GHz, support spectrum sharing and concurrent multi-band transmission, and allow joint sensing and communication. Design methodologies to enable such challenging features typically suffer from tradeoffs across energy efficiency, bandwidth, linearity and reconfigurability. These trade-offs arise from multiple domains, and primarily from the fact that enabling such features through carefully designed electromagnetic structures, taken from a library of templates, invariantly trades off with loss and efficiency. In this paper, we demonstrate how deep-learning based techniques can allow on-demand rapid synthesis of complex passives structures and antennas. These passive elements are not limited to a library of templates, and therefore, can enable functionalities beyond the capability of human intuition and design insights. Combined with circuits, we demonstrate a deep learning based mmWave PA with 30-94+ GHz P-sat,3dB bandwidth, while supporting concurrent multiband transmission for the first time at mmWave. We also demonstrate how this can enable rapid antenna designs with desired characteristics for these high frequency systems.