From Systematic to Intelligent: Assessing AI-Empowered Optimization Techniques for Analog Building Block Sizing

This paper presents a comprehensive, design-insight-based comparison between an artificial intelligence (AI)-empowered optimization-based analog building block sizing framework and the conventional manual design methodology. Although recent AI-empowered approaches are showing high performance, conventional systematic manual design methods such as the $g_{m}/I_{D}$ -based sizing are still the most widely used methods in the analog IC design community. This raises the necessity of the comprehensive comparative analyses between the two kinds of methods. To fill this gap, this paper compares the optimal designs obtained by a typical method of the former with those obtained by the latter method in the literature/industry. Four case studies, including a comparator, an amplifier (both standard and low power design), and an oscillator, using technology nodes ranging from $0.35~\mu $ m to 65 nm, are presented. Detailed performance evaluations and design insights are presented for each case study, with silicon validation provided for three designs. Our findings highlight that AI-empowered sizing not only meets but often surpasses conventional designs in key performance metrics, while still benefiting from designer interaction to align with design intents.