Wafer Maps and Sensors in Semiconductor Manufacturing
Key Terminology (Foundational Concepts)
Wafer
- A thin, round slice of silicon
- Used as the base to build computer chips
- One wafer holds many chips
Die (plural: dies)
- A single chip on a wafer
- Each die becomes one usable chip if it passes testing
Die-Level Testing (Wafer Probe Test)
- Electrical testing done before chips are separated
- Checks if each die works correctly
Yield
- The percentage of dies that work correctly
- Higher yield means lower manufacturing cost
Wafer Map
- A 2D picture of a wafer
- Shows which dies passed or failed testing
Pixel (in Wafer Maps)
- Represents one die
- Stores test results, not colors or brightness
Defect Pattern
- A visible shape formed by failing dies
- Often points to a manufacturing problem
- Examples: center failures, edge rings, scratches
Process Sensors
- Devices inside manufacturing machines
- Measure conditions like temperature, pressure, and power
Process Run / Production Instance
- A single execution of a manufacturing process, often corresponding to the processing of one wafer or lot under a specific set of tool conditions.
Introduction
Semiconductor manufacturing is a highly complex, multi-stage process involving hundreds of tightly controlled fabrication steps. Even minor deviations in process conditions can introduce microscopic defects that significantly reduce product yield and reliability.
To monitor, diagnose, and prevent such defects, manufacturers rely on multiple sources of data collected at different stages of production. Two of the most important and complementary data sources are:
- Wafer map data, which provides a spatial view of defects across a wafer after fabrication and testing
- Process sensor data, which captures physical and electrical conditions during wafer fabrication
While both data sources relate to the same manufacturing process, they capture fundamentally different aspects of it. Wafer maps describe where defects occur, whereas sensor data helps explain why those defects may have occurred.
This presentation explains how wafer map pixels are generated from physical wafers, how process sensors operate in semiconductor manufacturing, and how these two data modalities complement each other in manufacturing analytics.
Die-Level Electrical Testing
Before a wafer is cut into individual chips, it undergoes die-level electrical testing, commonly referred to as wafer probe testing.
During this stage: - A probe card makes physical contact with each die on the wafer - Electrical characteristics such as continuity, leakage current, and voltage thresholds are measured - Each die is classified as passing, failing, or untested
The results of this testing stage provide the raw information used to construct wafer maps.
What Is a Wafer Map?
A wafer map is a two-dimensional grid representation that preserves the physical layout of dies on a wafer.
- Each grid cell corresponds to a single die
- The spatial arrangement reflects the physical location of dies on the wafer
- The circular shape of the wafer naturally emerges from the grid structure
Wafer maps allow engineers to visually inspect defect distributions and identify spatial patterns that are difficult to detect using tabular data alone.
How Wafer Map Pixels Are Generated
In wafer maps, a pixel does not represent light intensity, as in conventional digital images.
Instead: - Each pixel encodes the electrical test outcome of a single die - Pixel values are discrete and symbolic
Typical encoding: - 0 → No die or untested location
- 1 → Die passed electrical test
- 2 → Die failed electrical test
Thus, each pixel corresponds directly to a physical die on the wafer.
Spatial Defect Patterns
Because dies are arranged spatially across the wafer surface, defects often form recognizable geometric patterns:
- Center defects — commonly associated with lithography focus or exposure issues
- Edge-ring defects — linked to temperature gradients or etching non-uniformity
- Scratch defects — caused by mechanical handling or tool contact
- Random defects — typically resulting from particle contamination
These spatial regularities make wafer maps well suited for image-based machine learning models such as convolutional neural networks (CNNs).
Process Sensors in Semiconductor Manufacturing
Unlike wafer maps, which are generated after fabrication, process sensors operate during fabrication.
Sensors are embedded within manufacturing tools such as etchers, deposition chambers, and lithography systems, and they monitor: - Temperature - Pressure and vacuum levels - Gas flow rates - Plasma density - Radio-frequency (RF) power - Chemical concentrations - Tool health and calibration indicators
These sensors provide continuous insight into the conditions under which wafers are manufactured.
Characteristics of Sensor Data
Process sensor data differs fundamentally from wafer map data:
- Tabular structure (rows × columns)
- Continuous numeric values
- Time-dependent measurements
- No spatial information
In datasets such as SECOM: - Each row represents a production instance (e.g., a wafer or manufacturing run) - Each column represents an anonymized sensor measurement - Sensor names are generic (e.g., sensor_0, sensor_1, etc.)
Wafer Maps vs Process Sensors
| Aspect | Wafer Maps | Process Sensors |
|---|---|---|
| Data structure | Image (2D grid) | Tabular |
| Resolution | Die-level (spatial) | Tool-level (global) |
| Timing | Post-fabrication | During fabrication |
| Primary insight | Defect location | Process conditions |
| Key question answered | Where defects occur | Why defects may occur |
Why Both Data Sources Matter
In real semiconductor manufacturing environments:
- Sensors detect abnormal process behavior during fabrication
- Wafer maps reveal how those abnormalities manifest spatially across the wafer
- Engineers combine both perspectives to identify root causes and improve yield
This combined view enables: - More accurate defect detection - Faster root-cause analysis - Data-driven process optimization
Example: Wafer Map Insight (Where Defects Occur)
What we observe
- Many failed dies form a ring near the wafer edge
- The pattern is consistent across multiple wafers
What this tells us
- The problem is spatial, not random
- Defects are linked to wafer position
Conclusion - The issue likely comes from a process that affects wafer edges - Examples: uneven heating or etching near the boundary
Example: Sensor Data Insight (Why Defects Occur)
What we observe
- Temperature sensors show higher values near the end of the process
- Gas flow readings fluctuate outside normal ranges
What this tells us
- The manufacturing process was unstable
- Tool conditions drifted during production
Conclusion
- Abnormal process conditions caused the defects
- These sensor signals explain why the wafer map shows failures
Key Takeaways
- Wafer map pixels represent die-level electrical test outcomes, not optical images
- Each pixel corresponds to a physical die on the wafer
- Process sensors capture fabrication conditions in real time
- Wafer maps show where defects occur
- Sensor data helps explain why defects occur
- Together, they provide a comprehensive view of semiconductor manufacturing quality
References
Kyeong, S., Kim, H., & Kim, S. (2020). Classification of wafer map defects based on deep learning methods with a small amount of data. IEEE Access, 8, 21796–21806. https://www.researchgate.net/publication/339903705
Li, Y., Zhang, X., & Wang, J. (2024). Deep learning approaches for wafer defect pattern recognition. arXiv Preprint. https://arxiv.org/html/2411.11029v1
JMP Community. (2021). How to make a wafer map in JMP in under 30 seconds. https://community.jmp.com/t5/JMPer-Cable/How-to-make-a-wafer-map-in-JMP-in-under-30-seconds/ba-p/348607
MathWorks. (n.d.). Classify defects on wafer maps using deep learning. https://www.mathworks.com/help/vision/ug/classify-defects-on-wafer-maps-using-deep-learning.html
PDF.com. (n.d.). Understanding semiconductor data visualization with wafer maps: An introduction. https://www.pdf.com/understanding-semiconductor-data-visualization-with-wafer-maps-an-introduction/
Wikipedia contributors. (n.d.). Substrate mapping. https://en.wikipedia.org/wiki/Substrate_mapping