RF Analog Circuit Parameter Generation Using GNN and Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analog circuit design lacks effective automation techniques, requiring extensive human effort and manual derivation of device parameters due to complex interrelations between device parameters and circuit specifications, limiting design accuracy and efficiency, especially in high-frequency RF circuits.
Innovation Solution
A deep reinforcement learning framework using a graph neural network (GNN) and fully connected neural network (FCNN) to model the relations between device parameters and circuit specifications, enabling automated design of analog and RF circuits across various semiconductor technologies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual derivation of device parameters is used, then design accuracy can be maintained through human expertise, but design time and labor effort increase significantly
Solution Approach 1:
The patent replaces manual human derivation of device parameters with an automated deep reinforcement learning system. The RL agent learns optimal parameter mappings from circuit specifications through training, eliminating the need for human experts to manually calculate parameters while maintaining high design accuracy and significantly reducing design time.
Solution Approach 2:
The patent creates a virtual copy of the human designer's expertise through the trained reinforcement learning model. The model captures the essential knowledge and decision-making patterns of expert designers, allowing it to replicate their parameter derivation process automatically without requiring actual human intervention in each design case.
2Extent of automation
If conventional ML methods are used, then design automation is achieved, but accuracy and efficiency are limited due to treating P2S optimization as a black box
Solution Approach 1:
The patent introduces domain knowledge as an intermediary component between the reinforcement learning agent and the circuit design process. This domain knowledge module provides explicit guidance on the relationships between device parameters and circuit specifications, enabling the RL agent to make more accurate and efficient design decisions by incorporating physical and circuit-level constraints.
Solution Approach 2:
The patent segments the design process into distinct components: the reinforcement learning agent handles optimization, while the domain knowledge module handles the physical and circuit-level constraints. This segmentation allows each component to specialize in its function, improving overall accuracy and efficiency compared to treating the entire P2S optimization as a black box.
3Extent of automation
If existing ML methods are applied, then automation is achieved for low-frequency analog circuits, but applicability to RF circuits and other semiconductor technologies is limited
Solution Approach 1:
The patent designs the reinforcement learning system with universal applicability across different circuit types and semiconductor technologies. The domain knowledge module is formulated in a technology-agnostic manner, allowing the same RL framework to handle low-frequency analog circuits, RF circuits, and various semiconductor processes by simply changing the input specifications and device models without retraining the core architecture.
Data Source
AI summary
A computer-implemented method is provided for generating device parameters of circuits using a pretrained reinforcement learning (RL) agent composed of a graph neural network (GNN) and a fully connected neural network (FCNN). The method is performed by steps including acquiring inputs with respect to a set of desired specifications or one desired specification of a circuit, device parameters, a fixed topology of the circuit and providing the inputs to the RL agent. The desired circuit description includes a graph modeling the topology of the circuit and device parameters of the circuit, and the desired specifications include gain, bandwidth, phase margin, power consumption, output power and power efficiency. The pretrained RL agent performs steps including transmitting an action selected from a set of actions to an environment module, updating the device parameters of the circuit according to the selected action using a data processor of the environment module, obtaining a current specification of the circuit by simulating a netlist of the circuit, acquiring a reward from the environment module, and generating the updated device parameters of the circuit.