What began as an ambitious vision for AI-assisted molecular beam epitaxy (MBE) has already resulted in a functioning, lab-integrated platform that combines full MBE control, real-time diagnostics, and autonomous decision-making. This article looks back on the key developments and milestones achieved during the first six months of the collaboration.
The project officially kicked off with a joint technical meeting defining scope, system architecture, software requirements, and FAIR-compliant data handling. From the outset, UnicornOne was designed as a modular and scalable system, capable of being transferred across different MBE tools rather than tailored to a single setup.
From architecture to the lab
An early strategic decision was to focus initial development on PDI’s MBE system M9, enabling rapid deployment and realistic testing under operational conditions. A complete system schematic and detailed documentation were created for M9, establishing a reusable template for future integration on other PDI systems.
Software development advanced along three parallel tracks: FAIR data infrastructure, device configuration and drivers, and AI model development. Within weeks, UnicornOne was communicating with the majority of M9’s hardware components, supported by a flexible debugging and testing interface. In August 2025, the wider PDI community saw UnicornOne for the first time during an internal AI demo, marking a key transition from planning to implementation.
AI workbooks and autonomous workflows
At the core of UnicornOne is the concept of AI workbooks: structured workflows that combine deterministic growth steps with specialised AI models. Over the first six months, these workbooks evolved from simulated prototypes into live, lab-based tools.
Key advances included near-photorealistic RHEED simulation for AI training, benchmarking of low-latency AI models, and the introduction of the “Uni” AI companion, capable of both conversational interaction and direct system commands. Live AI workflows were developed to identify GaN growth regimes, detect spotty RHEED, and calibrate effective growth temperatures.
A recent milestone further underscored the significance of this approach: for the first time, a user-authored UnicornOne plugin was developed directly within the lab, extending the platform with custom growth logic. This achievement demonstrates how UnicornOne’s open and extensible architecture enables researchers to contribute their own methods and innovations, reinforcing scientific autonomy as a core design principle of the system.
By October, UnicornOne and its AI components were installed directly on the M9 control PC. Shortly thereafter, the system was reading data from nearly all connected devices and progressing toward full write control.
From software development to research impact
As UnicornOne began running live on M9 for extended periods, the project crossed an important threshold: the AI workflows started generating new experimental insights, rather than simply validating software performance. Multi-day runs demonstrated that the AI workbook could determine GaN growth conditions on unknown samples in approximately 33 minutes.
A major milestone was reached when the “GaN Live” AI workbook was transferred from M9 to M1 without modification, successfully calibrating the sample and determining growth type. This demonstrated genuine AI interoperability across MBE systems — a key requirement for scalable and transferable automation.
Visibility and outlook
The project has also gained visibility beyond PDI. At the DGKK-DEMBE conference in Magdeburg, Faebian Bastiman presented UnicornOne as an integrated solution combining RHEED analysis, full MBE control, and AI-driven decision-making into a coherent, autonomous workflow.
At the six-month mark, UnicornOne stands as a strong signal of PDI’s role in shaping the future of materials research. Building on the success of the first project phase, PDI and Bizmuth MBE have agreed to expand the collaboration through the end of 2026, firmly establishing UnicornOne as an integral part of PDI’s MBE laboratory infrastructure. The project exemplifies PDI’s capacity to translate cutting-edge ideas into operational research tools, reinforcing the institute’s position as an innovation leader at the interface of materials science, automation, and artificial intelligence.
