Inspeção de Esferas de Solda

The Visual Inspection of Solder Balls in Semiconductor Encapsulation

Automated inspection of semiconductors has been the focus of numerous research efforts in Microelectronics and Industrial communities in the past few years. The semiconductors inspection has the role of feeding back the methods with information on specific errors, which can be correlated with production problems. The semiconductors analysis includes the assessment on occurrence of failures due to the materials involved in the process or inadequate definitions of the machine parameters.

In the semiconductors context, the production of memory devices represents a great manufacturing challenge, especially due to the small component’s dimensions and the required precision in its operation. Additionally, the market demand for memory devices has increased massively, requiring the expansion of memory production volume, making the manual inspection process critical. 

The use of the conventional visual inspection process, regarding a trained human operator, presents effectiveness between 80% and 90% of cases. However, after the first half working hour, the human operator visual acuity decreases significantly, for the analysis of a single type of defect. In Figure 1, is presented an example of a human visual inspection of silicon wafers, where the human operator should visually run through all the components on the PCB substrate looking for different types of defects.

In this paper, we present an approach to classify solder balls, in the soldering process of a silicon wafer, called die, on BGAs, contained in the PCB substrates. The solder ball is classified into three categories: i) correct; ii) absence; or iii) failure. We also introduce a CNN architecture for supervised classification of solder balls, which learns the main features that represent all approached types of soldering conditions. Experiments in real-world scenarios and simulations show that the obtained results are accurate and applicable in industrial scenarios. 

Our main contribution is to provide an approach, based on deep learning, to detect failures in the die soldering process. Furthermore, the proposed strategy provides a robust solution for a challenging component, in micrometers scale. 

Publications:

Pereira, Paulo V. L. ; Silva, Conceição N. ; Ferreira, Neandra P. ; Meireles, Sharlene S. ; OTANI, MARIO ; da Silva, Vandermi J. ; de Freitas, Carlos A. O. ; OLIVEIRA, FELIPE G. . Automatic Fault Detection in Soldering Process During Semiconductor Encapsulation. Lecture Notes in Networks and Systems. 1ed.: Springer International Publishing, 2023, v. , p. 130-145.

SILVA, CONCEIÇÃO ; FERREIRA, NEANDRA ; MEIRELES, SHARLENE ; OTANI, MARIO ; J. DA SILVA, VANDERMI ; O. DE FREITAS, CARLOS ; OLIVEIRA, FELIPE . The Visual Inspection of Solder Balls in Semiconductor Encapsulation. In: 19th International Conference on Informatics in Control, Automation and Robotics, 2022, Lisbon. Proceedings of the 19th International Conference on Informatics in Control, Automation and Robotics, 2022. v. 1. p. 750-757.

Acknowledgments:

This work was developed with support from Cal-Comp Eletronic through R&D project in Institute of Exact Sciences and Technology of Federal University of Amazonas, Itacoatiara, Amazonas.

Team:

Conceição N. Silva, Graduate Student at Universidade Federal do Amazonas (UFAM)

Neandra P. Ferreira, Engineer at Cal-Comp Eletronic

Sharlene S. Meireles, Engineer at Cal-Comp Eletronic

Mario Otani, Engineer at Cal-Comp Eletronic

Vandermi J. da Silva, Adjunct Professor at Universidade Federal do Amazonas (UFAM)

Carlos A. O. de Freitas, Adjunct Professor at Universidade Federal do Amazonas (UFAM)

Felipe G. Oliveira, Adjunct Professor at Universidade Federal do Amazonas (UFAM)