Researchers have developed a new method using deep learning to design complex millimeter-wave and terahertz circuits in minutes, a process that previously required years of human expertise and extensive simulations. This breakthrough, published in Nature Communications, could revolutionize the development of future wireless networks and high-resolution sensing applications.

The traditional design process for these circuits involves co-designing and optimizing active circuit elements and passive electromagnetic structures. This iterative process is time-consuming and complex, often relying on intuition, pre-selected templates, and ad-hoc parameter sweeps. The new method employs deep learning models to predict the scattering and radiative properties of arbitrary planar electromagnetic structures. This eliminates the need for time-consuming simulations and opens up a vast design space of potential circuit configurations.

The researchers demonstrated the versatility of their approach by designing various structures, including filters, resonators, power splitters, combiners, quadrature hybrids, diplexers, and antennas. They also designed and fabricated a broadband millimeter-wave amplifier chip, showcasing the potential of this method for end-to-end circuit design.

This deep learning-enabled design approach offers several advantages over traditional methods. It significantly reduces design time, from weeks to minutes, and allows for the exploration of a much larger design space. This can lead to the discovery of novel circuit topologies with improved performance and functionalities. The researchers believe that this method will pave the way for the rapid and automated synthesis of complex RF/sub-THz circuits and systems, enabling the development of next-generation wireless technologies.

©️ The Rocky Mountain Dispatch LLC. 2025