Domain Knowledge-Based Automated Analog Circuit Design with Deep Reinforcement Learning

Abstract

The design automation of analog circuits is a longstanding challenge in the integrated circuit field. This paper presents a deep reinforcement learning method to expedite the design of analog circuits at the pre-layout stage, where the goal is to find device parameters to fulfill desired circuit specifications. Our approach is inspired by experienced human designers who rely on domain knowledge of analog circuit design (e.g., circuit topology and couplings between circuit specifications) to tackle the problem. Unlike all prior methods, our method originally incorporates such key domain knowledge into policy learning with a graph-based policy network, thereby best modeling the relations between circuit parameters and design targets. Experimental results on exemplary circuits show it achieves human-level design accuracy (∼99%) with 1.5× efficiency of existing best-performing methods. Our method also shows better generalization ability to unseen specifications and optimality in circuit performance optimization. Moreover, it applies to designing diverse analog circuits across different semiconductor technologies, breaking the limitations of prior ad-hoc methods in designing one particular type of analog circuits with conventional semiconductor technology.

Demonstration of fully integrated parity-time-symmetric electronics

“…Fully integrated PT-symmetric electronics enables new capabilities in the microwave domain compared to the previous electronic platforms [22][23][24][25][26][27][28][29][30][31] . With the unique gain/loss tuning mechanism of PT symmetry, our system shows extended broadband response and improved noise performance for microwave generation over conventional devices.…”

“…In addition, more advanced IC technologies can be used to extend the functional and performance benefit of PT-symmetric systems to the higher millimeter and terahertz frequency range. The study is also expected to motivate further exploration such as PT symmetry in opto-electronics 18 , electro-acoustics 21 , and topological electronics (Supplementary Section 8.2) based on high-dimensional PT-symmetric structures [28][29][30] with standard IC technology, enriching scientific discoveries of non-Hermitian physics. The forward transmission is defined from gain side to loss side.…”

“…Beyond γ EP , C was held fixed at its asymptotic value." Note that machine learning methods [30,31] may be used to achieve an automated manner to adjust the gain-loss contrast for the system. We will explore this method in our future work.…”

“…Its scalability in physical dimension enables ICs to be a powerful platform that covers a wide applied spectra from DC to terahertz. It also supports integration of complex three-dimensional structures 32 , allowing one to extend a core electronic non-Hermitian unit into higher-dimensional structures to study topological electronics [28][29][30] . Despite such intriguing properties, IC technology is yet to be employed to realize PT symmetry, though gain, loss, and their coupling effects do commonly exist there.…”