ORCID Identifier(s)

0000-0001-9651-2382

Graduation Semester and Year

2016

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Materials Science and Engineering

Department

Materials Science and Engineering

First Advisor

Choong-Un Kim

Abstract

This dissertation presents experimental observations that may assist the understanding of electromigration (EM) failure mechanism active in Cu wire and Al thin film pad (wire bond). This study is motivated by an ongoing industrial effort to adapt Cu as wire bonding material for interconnection. Traditional material used for the wire bond has been Au; however, its high cost combined with its susceptibility to reliability failure caused by excessive growth of IMCs (IMC) makes Cu to be an attractive replacement. Though, since an application of Cu to wire bond technology is relatively new, many of reliability failure mechanisms are unknown especially the ones related to EM reliability as it is increasingly serious in limiting useful life of wire bond. This makes the investigation on EM failure mechanism in Cu to be necessary. One of the most challenging difficulties of EM study is the isolation of the EM from any other effects on the failure process such as an increase in temperature by Joule heating. The Joule heat effect is of particular concern because it is expected to be considerable at wire bond configuration, making failure by EM to proceed concurrently with other failure mechanisms. Our investigation then begins with design of test structure that can prove that wire bond failure is indeed induced by EM, and progresses towards understanding microscopic mechanism by which wire bond becomes failure prone by EM. Therefore, this study proposes the use of a special configuration sample where two interfaces can be tested at the same temperature and current density conditions. The design of the sample includes a two-level interconnect structure pattern on Si substrate where a short strip conductor allows flow of test current from one pad to another. The short distance combined with heat-conduction through Si substrate assists the minimization of temperature gradient between the two interfaces. This design helps to isolate the EM effect over other factors that could contribute to the interface degradation such as Joule heating. The samples are then subjected to various testing conditions of temperature and current density and it is indeed found that there exists a difference in the failure kinetics when the interfaces are compared: the pad where current flows from pad to wire (mass flow is from the wire to pad) always fails faster than the opposite case. If it is considered that both interfaces have been tested at the same temperature and current density, the only factor contributing to this difference will be the EM because it is directional. The dependence of failure rate on the pad in respect to the direction of current flow is also found in Au-Al wire bond which is also tested for comparison purpose. This result indicates that the A-Al and Cu-Al wire bond shares the same failure mechanism when electric current is applied and provides further evidence that EM plays a critical role in inducing degradation in the interface integrity. When the EM failure kinetics of the first failing pad of samples are collected, and analyzed, they are found to follow the classic "Black's equation" that relates the failure rate to current density and test temperature; the failure rate shows current density dependence with exponent close to 2 for both Cu-Al and Au-Al wire bonds and temperature dependency with an activation energy of ~1.2eV for Cu-Al and 0.9eV for Au-Al. While Au-Al and Cu-Al wire bonds do not show typically different EM failure parameters, overall life of Cu-Al wire bond is notably higher (nearly one order of magnitude). This suggests that EM life enhancement in Cu-Al is mainly due to slower diffusion rate of species at the interface, which is consistent with other observations reporting slower IMC growth rate in Cu-Al than in Au-Al wire bond interface. In order to understand the EM mechanism leading to failure, microstructure of the interface is closely inspected after EM testing. This study shows that the growth of IMC is enhanced at the failing pad where EM forces IMC to grow into Al pad. Both Cu-Al and Au-Al shows the same trend, that is that the growth of IMC is substantially enhanced by EM when it is directed from the wire to the Al pad. This is consistent with a common expectation that the growth of IMC layers leads to the interface failure and EM makes the failure to be accelerated by speeding up their growth. However, it is noticed that there exists a distinctive difference in failure morphology between Au-Al and Cu-Al wire bond interface. The difference is that the damage at Au-Al interface appears as an extended voiding, while such voiding is absent in case of Cu-Al interface. Instead, it appears that the damage proceeds by the growth of cracks. It is believed that slow interdiffusion rate in Cu-Al interface makes the Kirkendall voiding to be suppressed from its formation while other factors like interface strain becomes primary factor responsible for interface failure. Although EM failure mechanism in Cu-Al wire bond interface is microscopically understood by the use of the theory that EM forces IMC growth to be accelerated when it directs atomic flux towards Al pads where short-circuit diffusion path is available, the difference in activation energy between EM failure and thermal growth rate of IMC phases appears to disagree with the theory. However, EM failure involves multiple processes such as crack initiation and growth in addition to IMC growth, it is possible that its activation energy may not necessarily be the same to that of thermal IMC growth. It is our belief that the cracking is delayed at higher temperature because of active stress relaxation process while such process is suppressed at lower temperature. This can make the activation of EM failure to be higher than that of IMC growth. On the other hand, the microscopic mechanism of EM failure established in this study also provides a reasonable explanation on an abnormal failure behavior at the interface. As is indicated, the failure always occurs faster at the pad where EM flux is directed toward Al pad. However, the rate of failure development is actually the opposite in the beginning EM testing. Initially, the failure rate measured by the interface resistance change is higher at the pad where EM flux is directed toward wire. With progressing of EM, its rate decreases and is eventually slower than the opposite case. This crossover in failure rate does not appear in case of Al-Au wire bond interface but is unmistakably existing in Cu-Al wire bond. This is found to be related to the interplay between the type of IMC phase forming at given time and its contribution to the interface resistance. In the pad where EM is directed toward Cu wire, the growth of high resistance phase is initially more active because EM flux is against the direction of Cu diffusion. As EM progresses, with slow migration of Cu into Al pad, the growth of IMC phases in Al pad is generally suppressed, leading to slower increase in interface resistance. This agrees well with the microstructural mechanism found in our study. So as to better understand EM mechanism, the growth of IMC phases without EM load is investigated by subjecting the samples to only thermal condition. This study is done to compare IMC growth behaviors with and without EM, and also to determine the fundamental mechanism of IMC growth in Cu-Al; wire bond. This study leads to the conclusion that there are three IMC phases possible to grow at interface, and they are , and . Interdiffusion makes the phase to form first in Al pad because Al provides fast diffusion path for Cu due to abundant grain boundaries. This gradually consumes Al pad while a portion of is replaced for . Finally, Cu rich phase, , forms at Cu wire side but its growth is very sluggish due to lack of diffusion short-circuit. As consequence, the three IMC layers form at the interface, mostly consuming Al pad with time. The growth rate of the total IMC layer thickness, all phases combined, follows an ideal diffusion controlled growth kinetics showing dependence. The activation energy for the growth is observed to be ~0.5 eV, which is far lower than that of EM failure. While our investigation has yielded reasonably self-consistent EM failure mechanism active in Cu-Al wire bond, there still exists a number of subjects that await further investigation. Among many, the question as to the type of IMC phases possible to form at Cu-Al interface requires careful and extensive investigations. The source of stress that initiates the cracking and the type of IMC phase affecting such progress is an unanswered but critical question not only for understanding the fundamental mechanism of interface structure but also for finding ways to improving EM reliability of Cu/Al wire bonds.

Keywords

Electromigration, Intermetallic compound, Interdiffusion

Disciplines

Engineering | Materials Science and Engineering

Comments

Degree granted by The University of Texas at Arlington

Download

Share

COinS