Abstract Preview

Here is the abstract you requested from the imaps_2019 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.

Early detection and instantaneous cause analysis of defects in interconnects by machine learning (ranking-CNN) of scattering parameter patterns
Keywords: Prognostics and Health Monitoring, Scattering parameter, Machine Learning
This paper presents a novel method not only to detect defects in interconnects early but also to determine the root causes instantaneously by using machine learning of scattering parameter(s-parameter) patterns. Recently, there has been growing interest in monitoring the ongoing health of products and systems in order to predict component failures in advance of the catastrophic one. When it comes to electronic products, interconnect reliability has become crucial and affected the performance, power consumption and reliability of the whole circuit as the IC undergoes the continuous downscaling. Thus, it is fundamentally important to forecast the failure of interconnects to achieve high reliability of the system. Researchers, so far, have focused on means to assess the extent of deviation or degradation from expected normal operating to do the task above. However, it is also important to find out the root cause of the defect to reduce the life cycle cost of equipment by decreasing inspection costs, downtime, inventory and logistical support of fielded and future systems. We paid attention to the certain s-parameter patterns of each defective interconnect to provide information about both the severity of defects and the causes. S-parameters describe the electrical behavior of electrical networks when undergoing various steady state stimuli by electrical signals. Many electrical properties of networks of components (inductors, capacitors, resistors) may be expressed using S-parameters. Also, the S-parameter measurements using network analyzers are the most basic work of RF engineering. Thus, it would be highly convenient if the defects could be detected and analyzed by the s-parameter measurement itself. This study has dealt with bond wires, ITO and graphene electrodes as representative examples of conventional, transparent and next-generation interconnects respectively. Different kinds of defects were made quantitatively on a set of interconnects to investigate their effects on the S-parameter patterns. First, s-parameter measurements on normal bond wires showed patterns of typical inductors. It is simply because inductance of a wire increases with increasing frequencies. Thermal shock testing on bond wire samples was performed to pose defects caused by CTE mismatch between substrates and bond wires. It is investigated that the inductive pattern became prominent on defective bond wires. Second, the graphene specimens were precisely defected with 3 - 7 macroscale cracks and microscale ones introduced by 2.5% - 7.5% strain. The measurement on bare graphene materials show the inductance characteristic which is represented by decrease in S21 parameters with increasing frequencies. Then, the RF analysis on the cracked specimens reveals that S21 values in the low frequency region drop as much as increase in the DC resistivity. Moreover, the S21 parameter of the cracked specimens decreases more with increasing frequencies than that of the bare specimens, which means that the defects affect not only resistivity but also inductance of graphene materials. This trend becomes more prominent with microscale cracks than macroscale ones. Inductance depends mostly on the shape of cracks. Third, the s-parameter patterns of ITO electrodes with cracks and photo-chemically degraded ones were obtained and compared to the patterns for the normal electrodes. Interestingly, it is found out that defective interconnects showed certain and unique s-parameter patterns. If the cause of the defect is fixed, severity of the defects changes the level of the pattern without affecting its shape. Furthermore, we applied our own machine learning algorithm to achieve early detection and instantaneous root cause analysis. In the literature of machine learning research, ranking-CNN is the first work that uses a deep ranking model for facial age estimation. We modified this algorithm to decide both severity and causes of defects in interconnects from s-parameter patterns. Each basic CNN in ranking-CNN can be trained using all the labeled data, leading to better performance of feature learning and also preventing overfitting. In result, we were able to distinguish the causes and severity of the defects of interconnect samples at once with a much tighter error bound and faster process time.
Tae Yeob Kang, Senior researcher
Agency for Defense Development, South Korea
Taean, Choongnam
Republic of Korea

  • Amkor
  • ASE
  • Canon
  • Corning
  • EMD Performance Materials
  • Honeywell
  • Indium
  • Kester
  • Kyocera America
  • Master Bond
  • Micro Systems Technologies
  • MRSI
  • Palomar
  • Promex
  • Qualcomm
  • Quik-Pak
  • Raytheon
  • Rochester Electronics
  • Specialty Coating Systems
  • Spectrum Semiconductor Materials
  • Technic