[Paper Review] Microwave Integrated Circuits Design with Relational Induction Neural Network

Jie Liu, Zhi-Xi Chen+8 more|arXiv (Cornell University)|Jan 3, 2019

Microwave Engineering and Waveguides24 references5 citations

TL;DR

This paper introduces a Relational Induction Neural Network (RINN) that automates microwave integrated circuit (MWIC) design by learning structural relationships from data without prior knowledge. The method achieves up to four orders of magnitude faster convergence than traditional reinforcement learning, demonstrating strong performance in transmission line, filter, and antenna design tasks while explaining relationships via electromagnetic theory.

ABSTRACT

The automation design of microwave integrated circuits (MWIC) has long been viewed as a fundamental challenge for artificial intelligence owing to its larger solution space and structural complexity than Go. Here, we developed a novel artificial agent, termed Relational Induction Neural Network, that can lead to an automotive design of MWIC and avoid brute-force computing to examine every possible solution, which is a significant breakthrough in the field of electronics. Through the experiments on microwave transmission line circuit, filter circuit and antenna circuit design tasks, strongly competitive results are obtained respectively. Compared with the traditional reinforcement learning method, the learning curve shows that the proposed architecture is able to quickly converge to the pre-designed MWIC model and the convergence rate is up to four orders of magnitude. This is the first study which has been shown that an agent through training or learning to automatically induct the relationship between MWIC's structures without incorporating any of the additional prior knowledge. Notably, the relationship can be explained in terms of the MWIC theory and electromagnetic field distribution. Our work bridges the divide between artificial intelligence and MWIC and can extend to mechanical wave, mechanics and other related fields.

Motivation & Objective

• To address the challenge of automating microwave integrated circuit (MWIC) design due to its vast solution space and structural complexity.
• To develop an artificial agent capable of learning structural relationships in MWICs without relying on external prior knowledge.
• To enable rapid, data-driven design of MWICs such as transmission lines, filters, and antennas through end-to-end training.
• To bridge the gap between artificial intelligence and microwave electronics by grounding learned relationships in electromagnetic theory.

Proposed method

• Proposes a novel neural network architecture, Relational Induction Neural Network (RINN), designed to model structural relationships in MWICs.
• Trains the RINN agent end-to-end using reinforcement learning to optimize MWIC designs without incorporating domain-specific priors.
• Leverages relational inductive bias to infer connections between circuit components based on geometric and electromagnetic properties.
• Uses a reward function based on predefined MWIC performance metrics (e.g., return loss, bandwidth) to guide learning.
• Employs a differentiable architecture that allows backpropagation through structural design decisions.
• Validates the model on three distinct MWIC tasks: transmission line, filter, and antenna design.

Experimental results

Research questions

• RQ1 Can an AI agent learn to design microwave integrated circuits autonomously without prior knowledge of electromagnetic theory?
• RQ2 How does the RINN architecture compare to traditional reinforcement learning in terms of convergence speed and design quality?
• RQ3 Can the learned relationships in the RINN model be interpreted using established microwave circuit theory and electromagnetic field distributions?
• RQ4 To what extent can the RINN generalize across different types of MWICs, such as filters and antennas?

Key findings

• The RINN agent achieved convergence speeds up to four orders of magnitude faster than traditional reinforcement learning methods in MWIC design tasks.
• The model produced strongly competitive designs for microwave transmission line, filter, and antenna circuits, matching or exceeding benchmark performance.
• The learned structural relationships in the RINN were interpretable and aligned with known electromagnetic field distributions and MWIC theory.
• The method successfully automated MWIC design without requiring explicit incorporation of domain-specific prior knowledge.
• The approach demonstrated generalization potential across diverse MWIC types, suggesting broad applicability in microwave and related fields.