Researchers Unveil AI System for Assembling Graphene Transistors

A team of researchers has launched Qumus, an innovative AI system aimed at conducting experiments with quantum materials. This development, which has been detailed in a recent arXiv publication pending peer review, marks a significant advancement in autonomous scientific research.

The Qumus platform merges robotics, computer vision, and a multi-agent system to operate within a compact robotic laboratory. It autonomously manages the entire experimental workflow, including planning protocols, analyzing results, and generating reports.

In a notable achievement, Qumus successfully extracted graphene flakes using mechanical exfoliation and created nanoscale devices, notably a graphene field-effect transistor utilizing a van der Waals structure. The AI system optimized conditions to produce a graphene flake exceeding 200 µm² by varying temperature, contact time, pressing cycles, and tape peel-off speed, ultimately generating a flake of 245 µm² after multiple iterations.

Qumus also demonstrates error correction capabilities. For example, it identified the absence of a silicon chip and adjusted its procedures to perform exfoliation on a different substrate. Additionally, it rectified a misclassification of hexagonal boron nitride as graphene during its operations.

In assembling the transistor, Qumus executed a complex 90-minute dry transfer, which involved 30 physical operations and 18 decision points. The laboratory's operations are overseen by a structured network of agents based on large language models, while computer vision technology, specifically YOLOv8, is employed for object and inventory tracking.

Concerns regarding the implications of AI and quantum technologies for current security frameworks were highlighted by Darren Guccione, CEO of Keeper Security, earlier this year.