Version française
BlueKaizen Product Quality
BlueKaizen designed BK Product Quality, a generic algorithmic approach targeted to answering problems related to yield management, a key concept in manufacturing industries defined as the ratio between the number of non defective parts and the total number of manufactured parts.
BlueKaizen Product Quality uses the BlueKaizen Machine Learning Toolkit to provide statistical learning solutions to problems, such as accurate yield prediction and fast detection of defective products, or the control and improvement of manufacturing processes. The latter implies that the quality of the products is measured not only at the end but also during the process.
In the last decades, the fast evolution of manufacturing and non-intrusive product measurement technologies, associated to the exponential growth in the volume of data to acquire, transmit and store, have made it possible to:
- Measure many more products in the course of the manufacturing process in order to have a better knowledge of the production state at any time. The introduction of statistical tools for yield prediction can, from now, be considered.
- Have a per-product history of the measurements done during the manufacturing process, which makes it possible to automate most of the work necessary to find the root cause of a defect. The unsupervised classification algorithms seem to play a significant role here.
- Have a permanent overall vision of the production's global quality, which makes it possible to have, indirectly, an overall vision of the manufacturing process. An intelligent modeling of this information allows for earlier detection of slow drifts in the production line's behavior.
- Early detection of defective products in the production line.
- Analysis of the quality measurements taken throughout the manufacturing process, and their correlation with the expected final yield.
- Classification of the defective products in order to speed up the search of root causes.
Applications
BK Product Quality can be applied to various types of problems:
- Early defect detection
The aim is to detect, during the manufacturing process, products with a strong probability of being defective. The products whose quality parameters are globally far from the model, which represents normality - as inferred from a sample of production data representative of this stage of the manufacturing process - are considered potentially defective. The model is estimated by means of multivariate density estimation and quantile estimation algorithms. The corresponding approach is complementary to more traditional approaches, which consist in analyzing parameters one by one, trigerring an alarm when one or several parameters get out of specifications.
For more information about our software solution for the microelectronics and semiconductor industries, we invite you to have a look at BlueKaizen WaferQuality.
- Defect classification
The aim is to organize the products supposed to be defective into groups with common features and, if the corresponding information is available, to determine to which previously known type of defect a group may be associated. The solutions suggested for these problems call upon clustering and commonality search algorithms, in the unsupervised case, and classification algorithms in the supervised case. The latter correspond to the situation where the types of defects are known in advance.
For more information about our software solution for the microelectronics and semiconductor industries, we invite you to have a look at BlueKaizen WaferFit.
- Product reliability improvement
The aim is to identify, at the end of the production line, the products that have a high probability of early failure, among those products for which the test results are good (i.e. products that are good but unreliable). The solutions suggested for this problem consist in building a multivariate model of the test results, applying density estimation and supervised classification algorithms.
For more information about our software solution for the microelectronics and semiconductor industries, we invite you to have a look at BlueKaizen DeviceReliability.
- Slow drift detection
The aim is to determine if a distribution of products, and in particular those described as normal, characterized by a certain number of measured parameters, resembles a reference history or is drifting away from it. The slow drifts observed in product quality measurements are often the consequence of changes or problems in the manufacturing processes involved. Their early detection helps prevent yield crises. The solutions suggested for these problems call upon density estimation and quantile estimation algorithms.
For more information about our software solution for the microelectronics and semiconductor industries, we invite you to have a look at BlueKaizen WaferQuality.