Advances in Predictive Reliability Assessment and Failure Prevention for Semiconductor Manufacturing Equipment

Abstract

Semiconductor fabrication relies on hundreds of capital-intensive process tools, each containing subsystems (RF generators,
vacuum pumps, electrostatic chucks, gas delivery modules, ion sources, and cooling loops) whose degradation can scrap
wafers, interrupt cycle time, and erode yield. In recent years, the field of equipment reliability in semiconductor manufacturing
moved decisively from fixed-interval preventive maintenance toward data-driven predictive and prescriptive approaches.
This paper reviews the principal advances to date. It covers (i) deep learning for fault detection and classification (FDC) that
operates on raw, variable-length sensor traces rather than hand-crafted summary statistics; (ii) remaining useful life (RUL)
prognostics catalyzed by the public PHM ion mill etch benchmark; (iii) comparative machine-learning studies of failure
prediction from production sensor data; (iv) the extension of prediction into prescriptive maintenance that couples time-tofailure
models with Bayesian root-cause inference and maintenance planning; (v) enabling data and cloud infrastructure; and
(vi) facility, power, and subsystem reliability. Methodologically, the study is an integrative narrative review of 114 sources
published to date, comprising 14 core semiconductor prognostics sources and 100 related and semi-related studies screened
from a consolidated list of 1,056 records and spanning reliability engineering, energy, offshore asset integrity, industrial safety,
cybersecurity, cloud infrastructure, and supply chain research. A five-layer conceptual model linking detection, diagnosis,
prognosis, and maintenance decisions is proposed together with four testable propositions. The discussion identifies persistent
obstacles, namely class imbalance, label scarcity, maintenance-induced drift, chamber-to-chamber variability, cybersecurity
of connected tools, human factors, and spare-parts supply, and the paper sets out implications for practitioners, researchers,
and equipment suppliers, states the limitations of the review, and proposes directions for future research.