In the complex ecosystem of semiconductor manufacturing, data serves as the lifeblood that enables quality control, yield improvement, and product reliability. Having spent over 30 years in the industry, PDF Solutions has developed comprehensive systems that support diverse data types from initial design all the way through to system-level testing. This blog provides an overview of the landscape of semiconductor manufacturing data, what it means, and why it matters.
The Manufacturing Journey: From Design to System
Before diving into specific data types, it’s important to understand the semiconductor manufacturing flow:
- Design – Where it all begins with chip architecture and specifications
- Fab – Where silicon wafers are manufactured through hundreds of precise steps
- Wafer Sort – Electrical testing of individual dies while still on the wafer
- Assembly – Where good dies are extracted and packaged
- Final Test – Testing at the package or module level
- System Integration – Incorporation into final products like smartphones or computers
Our flagship product, Exensio Manufacturing Analytics (MA), supports data collection and analysis across this entire manufacturing spectrum, serving over 150 global customers including the largest semiconductor companies.
Key Data Types in Wafer Manufacturing
Manufacturing Execution System (MES)
The MES orchestrates all operations in a fab or foundry. Think of it as the central nervous system of the manufacturing plant, tracking materials, processes, and equipment status in real-time. This system generates work-in-progress data that shows exactly where each wafer is located and what processing step it’s undergoing at any given moment.
Automated Material Handling System (AMHS)
Modern fabs feature extensive automation with overhead transportation systems moving wafer cassettes between processing stations. Every wafer has a unique ID code, and the AMHS tracks each wafer’s journey through various tools and chambers. This traceability data is critical for root cause analysis when issues arise.
Fault Detection and Classification (FDC)
FDC data comes directly from the manufacturing equipment. Modern fabrication tools may have 100-200 different sensors monitoring parameters like temperature, pressure, gas flow, and valve positions. These sensors collect data at high frequency during each process run and generate process indicators that help detect any deviations from normal operation.
Metrology Data
Metrology involves measurements of physical characteristics and comes in three primary flavors:
- Shape/3D Geometry – Measures the three-dimensional attributes of structures on the wafer
- Critical Dimension (CD) – Primarily measures the width of lines, which is critical because line dimensions directly translate to transistor performance
- Overlay – Measures the alignment between different layers, ensuring precise layer-to-layer registration across the 2000+ manufacturing steps that occur over a three-month period
These measurements are taken from special targets located in the “scribe lines” (spaces between dies on the wafer).
Defect Data
Because all dies on a wafer should be identical, inspection tools scan wafers to identify anomalies. Defect data comes in two forms:
- Defect Inspection – Records attributes like X-Y coordinates, size, area, and defect type
- Defect Review – For selected defects, high-resolution images are collected and analyzed using Automatic Defect Classification (ADC) to identify the nature of the defect (particle, short circuit, etc.)
Analyzing defect patterns can reveal “spatial signatures” that often correlate to specific equipment issues. For instance, a circular pattern of defects might indicate a problem with a particular chuck in a processing tool – what we sometimes call finding the “smoking gun” in the fab.
Process Control Monitor (PCM) Data
PCM data is collected from specialized test structures in the scribe lines and provides electrical characteristics of the wafer. This is one of the most critical data types because it tells you how close you are to achieving ideal performance. PCM data includes:
- Transistor characteristics (transconductance, leakage, threshold voltage)
- Interconnect measurements (line and via resistance)
- Dielectric breakdown voltage (reliability indicator)
- Junction measurements (sheet resistance, contact resistance, capacitance)
- Application-specific parameters for technologies like RF or image sensors
Interestingly, PCM data is one of the few data types that foundries typically share with their fabless customers. Other data types like defects or metrology are often considered sensitive internal information or “dirty laundry” that foundries prefer not to disclose.
Electrical Wafer Sort (EWS)
EWS data comes from testing each die on the wafer under various conditions, including:
- Standard operating conditions
- High-temperature testing (ensuring chips work at 120°C – think of phones left on car dashboards)
- Low-temperature testing
- Burn-in testing to identify early life failures
The results determine which dies are considered “Known Good Dies” (KGD) suitable for assembly into packages. Dies are sorted into multiple “bins” based on their performance, allowing manufacturers to grade products accordingly. This is why you might see different grades of the same chip (like standard vs. high-performance FPGAs) or image sensors with varying numbers of acceptable defective pixels.
Virtual operations like “ink-out rules” may mark dies as defective even if they pass all tests – for example, if a die is surrounded by failed dies, it might be considered suspect and excluded from further processing.
Beyond the Wafer
Once the wafer sorting is complete and known good dies are identified, the wafer itself ceases to exist as a unit. Dies are picked from the wafer according to a “pick map” and placed into packages, beginning a whole new data journey through assembly and final testing.
The Scale of Semiconductor Data
The semiconductor industry generates terabytes of data daily, with retention requirements extending 5-15 years depending on the customer and application. This translates to petabytes of data that need to be stored, managed, and analyzed, often requiring dedicated data centers.
The ability to effectively harness this data is what separates good semiconductor manufacturers from great ones. By correlating information across different data types, companies can identify yield-limiting factors, improve processes, and ultimately deliver more reliable products to consumers.
Through comprehensive data collection and advanced analytics platforms like Exensio, semiconductor companies can transform raw manufacturing data into actionable insights that drive quality, yield, and profitability in one of the world’s most technologically advanced industries.