AI × ManufacturingFrom Perception to Autonomy
Star on GitHub
A growing research collection
How does AI move from recognizing a defect to coordinating a factory? Explore the manufacturing tasks, the methods and the decisions that connect them.
01 / Overview
Five manufacturing scopes meet five AI stages. Explore a task, or follow the pipeline from Perception to Autonomy.
From Perception to Autonomy: A Survey on AI-Assisted Semiconductor Manufacturing
* Corresponding authors.
02 / Inside manufacturing
Films are deposited, patterns are transferred, and structures are built layer by layer. Testing and packaging turn a processed wafer into finished chips.
BUILD THE LAYERS
A thin layer of material is added to the wafer. Its thickness and uniformity influence the structures built in later steps.
Fabrication repeats deposition, patterning and material removal to build multiple layers.
03 / Five AI stages
Recognize patterns, estimate outcomes, compare explanations, choose actions, and execute and adjust them through feedback.
L1 / Perception
Recognize patterns, defects and operating states in images or sensor data.
Locate a defect in an inspection image, or recognize a spatial failure pattern on a wafer.
Explore Perception papers ↗04 / The research landscape
Explore how the literature is distributed across manufacturing scopes, AI stages and years.
Explore a manufacturing scope and an AI stage
Circle area = share of scopeExplore a bar to see the research over time.
Studies are grouped by their representative publication’s year.
Entries marked “Counted with” share one study count. Studies may span multiple AI stages; testbeds without stage labels contribute only to scope totals. Collection updated: .
Download data ↓05 / A shared direction
A gain in one task can change what happens in others. Better manufacturing outcomes require coordinated decisions.
01 / Operating loop
World models predict action consequences. Agents coordinate decisions across tasks; controllers and transport systems execute them. New measurements guide the next decision. Physical AI supports adaptive robotic inspection and maintenance.
02 / Learning loop
Operating feedback improves world models and agent workflows. Recurring errors guide more informative experiments and better diagnostic steps.
Recursive self-improvement (RSI) also refines how the system proposes and evaluates improvements.
Our ongoing work
Explore fabxr.cn
Keep exploring
The collection grows with the field. Browse the papers here, or suggest additions on GitHub.
A Survey on AI-Assisted Semiconductor Manufacturing
@unpublished{zhong2026perception,
title = {From Perception to Autonomy: A Survey on {AI}-Assisted Semiconductor Manufacturing},
author = {Zhong, Rui and Huang, Zhikun and Zhang, Xue and Jin, Qian and Li, Yu and Sun, Qi and Gao, Dawei and Song, Yi and Wu, Hanming and Zhuo, Cheng},
year = {2026},
note = {Manuscript}
}