AI powered design approach to prototyping Power Amplifiers (PAs) in GaN

GaN is a high efficiency semiconductor with huge potential in upcoming technologies for 5G and 6G telecommunications networks. However, it is still hampered by material issues
that result in trapping and self-heating and ultimately poor linearity. To resolve this requires a sound understanding at the device and circuit levels. The traditionally approach to designing a PA requires (i) an accurate large signal model of the transistor. Subsequently, (ii) a known topology of network with physical dimensions is tuned for realising matching impedances. Both steps, involve time-intensive iterative processes that rely on the experience of the designer.

Deep convolutional neural network (CNN)-based EM emulators have transformed this design process in recent years, to allow rapid prototyping of the passive network [K. Sengupta 23, Beach
23], while relatively little is demonstrated by way of device modelling [ Guo 23]. In the latter work, the approach is based on a hybrid physical model that still relies on parameter extraction. Our goal is to explore new developments in AI such as reservoir computing for exploring large signal and matching networks. The model should be sophisticated to predict challenges of thermal and trapping constraints that currently affects GaN technology.