Abstract: This three-part talk series deals with challenging problems of the modern hi-tech manufacturing industry (electronic and memory products), such as (1) Screening for Reliability, (2) Detecting Systematic Defects, and (3) Test for Yield Learning. The offered solutions conform to the systematic data-driven Six Sigma methodology which is based on setting extremely high objectives, collecting and deep analysing comprehensive production data with the aim to defect elimination toward the level below the six standard deviations between the mean and the nearest specification limit in any process. In particular, the following successfully developed real-world industry-originated projects will be discussed and generalised for extended implementation and application of the obtained results:
1. Eliminating the Burn-in Bottleneck in IC Manufacturing
Reliability screening is one of several types of testing that are performed at different stages of the IC manufacturing process. It plays an important role in controlling and ensuring the quality and consistency of integrated circuits. One of the most popularly used forms of reliability test is burn-in testing (i.e., accelerated testing performed under elevated temperature and other stress conditions). Burn-in is normally associated with a long test time and high cost. As a result, the burn-in testing is often a bottleneck of the entire IC manufacturing process, limiting its throughput. It is no surprise therefore, that much attention and efforts have been dedicated towards possible reduction or even elimination of the burn-in testing.
This presentation offers a step-by-step methodology for the burn-in test time reduction of up to 90% based on the extended use of the High-Voltage Stress Test (HVST) technique. The Weibull statistical analysis is used to model the infant mortality failure distribution.
2. Defect Cluster Recognition for Fabricated Semiconductor Wafers
Many systematic failures in the wafer fabrication (so-called frontend process) can only be caught during the IC manufacturing (i.e., during the backend process). Thus, there is a need for a simple yet accurate system to perform a wafer defect cluster analysis based on fast knowledge extraction from the production test data. The talk will cover the design and development of an automation tool to carry out this task - Automatic Defect Cluster Analysis System (ADCAS). It is aimed at supporting the backend initiated efforts, such as defect root-cause identification, die-level neighbourhood analysis as well as yield analysis and improvement. It is suitable for a plug-and-play type application on semiconductor production databases while providing an excellent trade-off between the simplicity of implementation and high accuracy of the analysis.
3. Automatic Media Inspection in Magnetic Memory Drive Production
In the modern high-volume hard disk drive production process, if an assembled product fails the final test it is normally not discarded, but instead it is sent for so-called Teardown. There it is disassembled to the constituent components. These components are thoroughly examined and retested for their individual functionality. If found to be in a good operational condition, they are redeployed in new products. To retest the magnetic disk (or media) the Laser Doppler Vibrometry (LDV) has been traditionally employed. Unfortunately, LDV test is normally lengthy thus causing a bottleneck in the Teardown, and thus reducing the overall manufacturing efficiency. In order to address the problem, manual visual inspection is often performed as a preliminary filtering step. Such an arrangement is not optimal as it is open to human error factor. It still could be costly and has throughput limitations.
In this part of the talk series, the factors influencing successful and rapid image acquisition of micrometer level defects on a specular surface are explored, namely, camera spatial resolution, spectral properties, image system Signal-to-Noise Ratio and lighting methods. A detection as well as classification scheme is offered to classify four major types of commonly occurring media defects.