Automated Sizing for Complementary Cross-Coupled LC-VCO Using Deep Deterministic Policy Gradient
Year: 2026
Type: conference-paper
Abstract: Modern analog integrated circuit design, particularly for Voltage-Controlled Oscillators, faces significant challenges due to high-dimensional design spaces and complex tradeoffs between coupled performance metrics. Traditional manual resizing methods are often time-consuming and struggle to achieve optimal results in advanced technology nodes. This paper proposes an automated sizing framework leveraging the Deep Deterministic Policy Gradient (DDPG) reinforcement learning algorithm to optimize a complementary cross-coupled LC-VCO. By integrating a simulation environment, the agent learns an optimal sizing policy through reward feedback based on the circuit’s Figure of Merit. Implemented in a 65 nm CMOS process at a 1 V supply, the proposed framework achieves a phase noise of $-\mathbf{1 1 2 . 5 ~ d B c / H z}$ at 1 MHz offset and a power consumption of only 0.42 mW. The resulting figure-of-merit of $\mathbf{- 1 8 6 . 5 ~ d B c} / \mathrm{Hz}$ validates the effectiveness of DRL as a robust EDA support tool for analog circuit optimization.
Institution: Ho Chi Minh City University of Technology
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204114267
