AI/ML Techniques Supporting the Design Automation of RF/mm-Wave Devices and Circuits

Traditional design methodologies for RF/mm-wave ICs have long been characterized by an iterative process of trial and error, where designers rely heavily on experience and intuition to achieve functional circuits. This approach, while effective in many respects, is becoming increasingly inadequate in the context of the rapidly evolving demands of nowadays high-performance complex systems. In recent years, we have entered the era of Artificial Intelligence (AI) and Machine Learning (ML), and therefore its application to the design of RF/mm-wave ICs is no exception. This workshop will demonstrate how innovative AI/ML techniques can be used to improve the automation in the design of RF/mm-wave circuits and systems. The workshop will cover different design stages, from the device up to the system-level design. Namely, it will be shown how ML techniques can be used to synthesize integrated inductors and transformers with optimal performances in milliseconds, how Artificial Neural Networks (ANNs) can be used to estimate process variability much faster than Monte-Carlo methods for example, how can the layout of RF/mm-wave circuits be automatically performed while improving routing parasitics for example, and how can evolutionary algorithms together with ML models efficiently explore the performance design space to achieve optimal circuit trade-offs. It will be shown how such AI/ML techniques can be used to efficiently automate the design of RF/mm-wave circuits/systems and circuit examples will be shown not only in academic but also in industrial context, with silicon proven automated designs. It will be shown that the design of RF/mm-wave circuits can be performed in days rather than weeks/months if design automation is implemented in the design flow of RF/mm-wave designers. More information: The workshop will be based on novel research, not yet published (measurement results of an LNA and PA operating at 28GHz automatically designed using ML/AI techniques — automatic sizing and layout) and some already published works. Examples of models for integrated inductors and transformers will be shown, with silicon measurements up to 40GHz (for different technologies), with the errors from ML models below 3%. Also, as an example of the circuit design automation, several circuit classes will be shown, such as a VCO @ 2.4GHz, an LNA @ 28GHz and a PA @ 28GHz.