A Deep Learning-Based Approach for Predicting the Performances of CMOS Voltage-Controlled Oscillator with Optimized Component Sizes

While design automation plays a crucial role in contemporary large-scale digital systems, the automation of the transistor-level circuit design process continues to pose significant challenges.Recent studies indicate that deep learning algorithms may be utilized to determine optimal transistor dimensions in compact circuitry, such as voltage-controlled oscillators.However, achieving robust and efficient analog circuit design automation in integrated circuit field remains challenging.A deep neural network architecture is introduced for the automatic sizing of analog circuit components, specifically targeting radio frequency applications within the 2 to 5-GHz range.A novel deep learning model designed to simulate voltage-controlled oscillators for microwave applications.This work introduces four algorithms: DNN, CNN, RNN, and SCINet.Two characteristics have been evaluated: output power and phase noise.The models achieved an accuracy of 96%-97% and exhibited a loss ranging from 0.0024 to 0.0036.The prediction of the required features demonstrates outstanding performance across all utilized models.We aim to determine the most effective deep learning model suitable for a specific dataset and computational setting.