MIT Spinout FabuBlox Introduces FabAgent for Semiconductor Process Engineering

FabuBlox has introduced FabAgent, an AI collaborator designed to help semiconductor engineers build, analyze, and emulate fabrication processes before running them in a cleanroom. The MIT spinout is connecting that agent to structured process designs and facility knowledge, aiming to help engineers reason through the many dependencies that determine whether a fabrication flow will work.

The company’s central proposition is that process intelligence needs to understand the fab as well as the process. A sequence of fabrication steps must fit the equipment, materials rules, recipes, and operating limits of the facility that will execute it. FabAgent’s connection to FabuBlox Facility OS is intended to make that context available during engineering decisions.

From Process Instructions to a Visible Device Structure

FabAgent sits within a platform that already organizes fabrication knowledge. The company’s Process Design Suite combines standardized process descriptions, cross-section visualization, layer-stack emulation, version control, and shared repositories. Engineers can describe fabrication intent and use the AI assistant to help construct flows, configure individual steps, and explore tools and recipes while retaining review of the proposed process.

The visual layer gives those decisions a physical reference. FabuBlox says its ray-tracing algorithms emulate how successive steps transform the device’s layer stack, including film coverage and sidewall profiles. Engineers can select a cutline from a GDS layout file to connect the fabrication sequence to a cross-section of the intended device. Changes to the flow propagate through the visualization as the design evolves.

That combination is useful because a process step changes the starting conditions for later steps. Reviewing the evolving structure alongside the instructions can help a team spot integration problems earlier. The usefulness of the emulation still depends on how faithfully its parameters represent the equipment and process conditions, making calibration and experimental measurements an important part of the engineering loop.

Facility OS Gives the Agent Context About the Fab

Facility OS represents the equipment and constraints behind a process. Its structured facility model includes tool capabilities, supported resources, material restrictions, parameter ranges, recipes, and calibration data. Engineers can search a tool library, while AI-assisted matching is designed to help identify equipment suitable for a requested operation.

Recipes are linked to particular tools and their characterization data. As recipes evolve, the platform highlights changes so engineers can selectively synchronize them with process flows. This preserves the relationship between a proposed process and the settings and measured behavior of the equipment expected to carry it out.

Contamination control is another important part of that model. Facility OS combines material definitions and layer-stack information with tool-specific restrictions to support compatibility decisions during planning. An agent evaluating a process can therefore consider whether a step fits the facility’s permitted material and equipment combinations, rather than treating each operation as an isolated instruction.

The company also describes workflows for reviewing incoming fabrication requests and estimating costs from configured process steps and facility pricing. Those capabilities extend the same structured knowledge into the coordination work surrounding fabrication.

A Company Built From MIT’s Fabrication Experience

FabuBlox’s roots are in the knowledge-sharing problems its founders encountered at MIT. An MIT Microsystems Technology Laboratories biography says Jan Tiepelt and his co-founders created the platform in 2023 to support shared database building for the MIT.nano and MTL community. Tiepelt completed his MIT electrical engineering PhD in May 2024 after research on degradation in blue OLEDs.

MIT’s Microsystems Technology Laboratories lists FabuBlox among startups that emerged from its community. Investor Pillar VC identifies the founding team as MIT PhDs Jan Tiepelt and Josh Perozek, together with engineers Jack Muller and Eyal Perry.

The company joined MIT.nano’s START.nano program, according to a January 2025 MIT announcement. The program supports hard-tech ventures through discounted facility access, technical advice, and connections to the institute’s innovation network. MIT’s description of FabuBlox at that time emphasized its process-design interface and GitHub-like repositories for organizing fabrication knowledge.

That history helps explain the current product direction. Versioned process flows, shared repositories, and structured equipment records give an AI assistant an engineering context it can work with, while preserving the history of how a design developed.

Learning From Wafers, With Autonomous Fabs on the Roadmap

FabuBlox’s third platform component, Fab Run Studio, connects process-flow versions and recipes to physical wafers, execution records, and metrology. The company labels it as early beta. Its described workflows include tracking branched wafer routes and parameter splits, comparing outcomes, and using wafer maps and statistical process control to identify variation.

The intended feedback loop connects an observed result to the wafer, route, recipe, and tool that produced it. FabAgent can then help engineers consider a revised flow, an updated calibration, or the next experiment. Keeping those relationships together is particularly relevant in research and high-mix fabrication, where a process may change frequently.

The wider industry is also investing in shorter learning cycles, as seen in collaborative semiconductor process R&D that brings materials, equipment, and integration teams closer together. FabuBlox is addressing the software and knowledge-management side of that challenge.

Its longer-term vision goes further: specialized agents for areas such as lithography, deposition, and etching would learn from physical outcomes and help coordinate experiments and process changes. Fab-specific world models and increasingly autonomous manufacturing remain development goals. Facility OS likewise describes live equipment signals and predictive maintenance as part of its future direction.

For engineers evaluating FabAgent now, the immediate opportunity is a collaborator grounded in the structure of a process and the constraints of a facility. The next step is connecting that reasoning to traceable experimental outcomes, so each fabrication run can inform a better engineering decision.