Optical and Auger Microanalyses of Solder Adhesion Failures in Printed Circuit Boards

Transcription

1 Optical and Auger Microanalyses of Solder Adhesion Failures in Printed Circuit Boards K. Kumar and A. Moscaritolo The Charles Stark Draper Laboratory, Incorporated, Cambridge, Massachusetts ABSTRACT Printed circuit boards used in electrical systems have shown solder separation from the copper wiring network with prolonged use at about 325 K (120~176 Optical and Auger microanalyses of boards that have shown such failures have revealed the existence of discrete oxide particles in the form of an oxide layer at the separated interface. High temperature accelerated aging tests have shown that the time-dependent accumulation of the oxide particles at this interface results from rejection of the oxygen out of the reacted zone and into the intermetallic-matrix boundary with the continued growth of (low Pb-containing) CuxSn~ intermetallic compounds. (These compounds are formed as a consequence of the chemical reaction that occurs between the Pb-Sn solder and the underlying copper circuit.) Increased thermal activation leads to increased growth of the intermetauics, and this results in more oxygen exclusion from the reacted zone into the interface region (giving rise to a time-dependent accumulation of the oxide particles at this boundary). Eventually, these particles inhibit the solid-state diffusion reaction between copper and the CuxSny intermetallics and lead to the observed separation of the solder from the underlying copper. Experiments performed in this study have shown that the oxygen (giving rise to oxide-particle accumulation at the interface) is distributed throughout the bulk of the copper plating. Organic addition agents, codeposited with the copper during the formation of the copper wiring network, are probably acting as a source for the observed oxygen. Printed circuit boards are used extensively in the electronics industry as components of larger:electrical assemblies. The printed circuit ~ consists of a wiring network which is produced using a combination of photolithographic and plating and laminating techniques. Both elec~roless and electrodeposition procedures are used to fabricate these circuits. A layer of a preselected Pb-Sn solder composition is electrodeposited on top of the copper network. To achieve proper wetting of the copper by the solder material, a post-deposition fusion procedure is used. In one process, this consists of heating the entire assembly to about 480 K (410~ for close to 15 sec. This results in phase segregation within the solder material and in chemical reaction between the solder and the copper circuit, thereby enhancing the integrity of the completed circuit through increased solder adhesion (due to the chemical forces that result from the soldercopper reaction). A few of the printed circuit boards used in electronic systems at about 325 K (120~176 have shown severe degradation of solder adhesion with prolonged use. This time-temperature dependence of solder layer separation in some of the printed circuits has been known to exist by the circuit industry for a reasonably long time (1). Past attempts at identifying the causes of these failures have met with only limited success (1-3). Our own experience with the use of these boards has shown that circuits obtained in 1977 from vendor A did not display solder adhesion failures. In contrast, samples procured from another vendor (B) in 1978 almost consistently showed this type of deterioration. In most instances, solder adhesion degradation was found to reach a level whereby the solder could be easily removed from the copper surface with a fineedged X-acto blade with a minimum of effort. A detailed study was consequently initiated at this Laboratory to determine the causes of the observed dewetting of the solder. It was hoped that, as a result of such an investigation, further use of these materials could be qualified and also a determination made as to whether Key words: oxide, intermetallics, solder separation, mechanism. newer (1979) material procured from vendor B could be used for future applications. Such an evaluation was deemed desirable in view of the manufacturer's claim that suitable modifications had been made in the processing procedures to produce a 1979-designated product that was superior to the 1978-designated product. This paper illustrates the nature of the problem and identifies the mechanism by which the observed solder separation has most probably occurred. Experimental Procedure Representative samples were mounted (with the copper-solder interface slightly inclined to the plane of polishing) at room temperature in a resin-containing cast and polished for metallographic examination. The selected angular tilt of the copper-solder assembly allowed amplification of the solder-copper interface, upon polishing, by a factor of ten. The samples were photographed, using optical microscopy, and analyzed using a Physical Electronics Auger system in the aspolished condition. The heat-treatments that were performed on several of the selected samples are described in detail in the next section. Results and Discussions The different features observed in microstructures which are typical of those observed in failed samples are shown in Fig. 1. As one traverses the solder-copper interface from the solder side to the copper side (that is from left to right), several layers are observed. The observed layers correspond to a two-phase solder region which contains a high concentration of the Pb-rich phase, two CuxSny intermetallic compounds (containing very low levels of Pb), and a region with a high concentration of oxide particles located between the CusSn intermetallic and the copper. The two intermetallic phases, Cu3Sn and Cu6Sns, observed in this study have earlier been reported elsewhere (I). These correspond to the near-equilibrium phases predicted by the binary Cu-Sn phase diagram (4) indicating that the role of Pb in this ternary system is minimal. An Auger line scan for Pb (shown in Fig. 1), however, in- 379

2 380 J. Electrochem. Soc.: SOLID-STATE SCIENCE AND TECHNOLOGY February 1981 Fig. 1. Typical Cu-solder interface region in failed board. Photograph shows two CuxSny intermetalllcs and oxide presence at Cu- Cu,~Sn boundary. Auger line scans were made acros~ the entire interface region. dicates that a sma]l amount of Pb does exist in these phases, possibly in solid solution. [Details on Auger spectroscopy and interpretation of spectra can be obtained from Ref. (5).] All of the main features can be easily identified on the Auger line scans (also contained in Fig. 1) that were obtained on this sample. The strong intensity of the oxygen signal at the Cu3Sn-Cu boundary (in the Auger scan) clearty shows that the particles observed are indeed an oxygen-containing species that have collected at this interface. The presence of a two-phase zone with a high concentration of the Pb-rich phase (observed on the solder side of the micrograph as clark precipitates) was a direct consequence of the formation of the (low Pb-containing) Cu-Sn intermetallic compounds between the solder material and the copper network. Formation of these intermetallics resulted in an exclusion of Pb from the reacted regions leading to an enrichment (in Pb) of the solder material close to the solder-cu6sn~ boundary. An Auger survey showed that the only elements which were consistently present in the several samples that were examined (in detectable quantities) were Cu, Pb, Sn, O, and C, all of which were expected. Because of the known differences in the performance of the 1977 and 1978 circuit boards, typica,1 samples representative of those years and one sample, designated as a 1979 product, were selected for further examination. Figure 2 shows the microstructure observed for the as-received condition of these circuit boards. The reacted zone was found to be considerably smaller in the 1977 sample compared to the 1978 and 1979 samples. This also accounted for the near absence of the Pb-rich zone in the 1977 sample in contrast to the 1978 and 1979 samples where its existence was clearly observed. Larger reaction zones mean more overall exclusion of Pb from the reacted area. One, therefore, expected an intense concentration of the Pb-rich phase at the solder-intermetallic boundary in the 1978 and 1979 samp.les in agreement with the observations. The differences in the extent of tbe reaction observed in samples obtained from the two vendors was partly explained by the differences in the solderfusion procedures used by these vendors. Whereas vendor A uses an infrared lamp to cause improved wetting of the copper by the solder, vendor B uses an oil bath in which the entire assembly is immersed. It is possible that the circuits are subjected to more thermal activation by oil immersion than is the case when they are exposed to infrared radiation. Differences in thermal activation can explain the differences in the extent of the reacgon that was observed. The microstructures shown in Fig. 2 were found to be substantially free of the oxide layer that is known to exist at the copper-cu3sn boundary in boards where separation of the solder is known to have occurred. It may be noted, however, that a few oxide particles were nevertheless found to be attached to the copper surface in the 1979 sample even in this condition. Since the degradation in the solder-to-copper adhesion has been observed to be dependent on both time and temperature, samples of as-received boards were subjected to elevated temperature aging treatments in air (so as to accelerate the kinetics of the microstructure changes that are induced during actual service at lower temperatures). Effects related to sampleto-sample variations were avoided by sectioning the as-received boards in two, such that the as-received state of the individual boards could be examined on one piece while the other piece could be examined after it had been subjected to the desired heat-treatment. The microstructures observed for each of the as-received samples corresponded very closely to those shown in Fig. 2. The heat-treatment temperature was chosen as 438 K (165~ and the samples were examined after 4, 75, and 264 hr of exposure. A gradual development of the oxygen-containing species, with increasing time, was observed in the 1978 and 1979 samples in contrast to the 1977 sample where none was observed. The sample microstructures obtained after the 264 hr treatment on these three samp.les are shown in Fig. 3. It is clear that the condition observed for the 1978 and 1979 type samples is most undesirable, since it ultimately results in solder separation from the copper network. That such a direct relationship exists between development of the oxygen-containing particles and degradation of solder adhesion was verified

3 VoZ. 128, No. 2 OPTICAL AND AUGER MICROANALYSES 381 Fig. 2. As-received microstructures of 1977, 1978, and 1979 boards. Extremely narrow reaction zone with little enrichment of the Pb-rich phase in the neighboring solder region is observed in the 1977 sample compared to the 1978 and 1979 samples. in separate experiments. These consisted of soldering individual wire leads to the printed wiring network prior to the 438 K treatment. After the thermal exposure the leads were pulled at known increasing loads until they separated from the circuit board. The decreasing value of load required to cause separation, and the increase in the amount of copper observed after separation (with increase in total thermal exposure), were considered a direct measure of the degradation in solder adhesion. It was determined that these correlated well with the accompanying changes in microstructure which showed a clear development of the oxygen-containing species in samples produced by vendor B. Figure 4 contains an oxygen-line scan which was made across the copper-to-solder interface region in the 1978, 264 hr heat-treated sample. Unlike other line scans that were made on similarly heat-treated samples, this one showed strong oxygen peaks on both sides of the CuxSny intermetauics; that is, at the Cu- CusSn interface, and also at the (Pb-Sn solder)-cu~sn5 interface. Also, within the intermetallics themselves, a lower level of oxygen was detected than that which was found to exist in both the copper plating as well as in the Pb-Sn solder deposit. This observation clearly showed that the intermetallics rejected oxygen on both sides, as they grew, and this accumulated as oxides at both of the intermetallic-matrix interfaces. Oxide accumulation at the solder-cu6sn5 boundary could be a contributing factor to fractures at this interface that appeared in observations made earlier on similar metallic systems (2). The reasonably high level of oxygen in the solder area (of Fig. 4) was found to exist mostly as an oxide of Sn (as was interpreted from the close Fig. 3. Microstructures after 165~ 264 hr air heat-treatment. Severe oxidation observed at Cu-Cu~Sn boundary in 1978 and 1979 samples. None observed in 1977 sample.

4 382 J. E~ectrochem. Soc.: SOLID-STATE SCIENCE AND TECHNOLOGY February 1981 Fig. 4. Auger line scans across interface region in ~ 264 hr air heat-treated sample. Oxygen peak~ observed on both matrix-intermetauic interfaces. correspondence between the Sn and the O Auger line scans in the solder region). A low level of oxygen within the intermetallics and the accumulation of an oxygen-containing species at the intermetallic-matrix boundaries clearly implies that oxygen present inside the solder will not migrate to the Cu-Cu3Sn boundary and lead to the observed degradation of the adhesion at that interface. The source of oxygen mustl therefore, be located on the copper side of the reacted assembly. The two sources that could conceivably give rise to the time-temperature accumulation of the oxide in failed samples are the environment (because all of the aging occurs in air) and the copper plating. If the oxygen contribution is from the copper plating, then the source of oxygen must be distributed reasonably uniformly throughout the thickness of the deposit. Surface oxidation of the copper plating, prior to solder deposition, cannot, by itself, account for the timetemperature dependence of oxide development at the interface. If surface oxidation of the copper were the dominant contributor of the observed oxygen, the density of the oxide particles would not have significantly changed with increase in reaction time. The oxide particle concentration at the intermetallic-matrix interface, in such a circumstance, would have either remained constant (after all of the available oxygen was used up to form oxides in the very initial stages of the reaction), or even decreased because the intermetallics do show a finite, albeit low, oxygen solubility. Most of the oxygen contribution to the development of the oxide must, therefore, have originated either from the aging environment or from the bulk of the copper plating. In order to further isolate the source of the oxygen (which leads to ultimate separation of the solder) an additional experiment was performed as part of this study. Sectioned samples, similar to those treated earlier in air, were obtained from the circuit boards and subjected to a 240 hr heat-treatment at:438 K (165~ in an argon atmosphere. Figure 5 shows the microstructures that were obtained on the argontreated samples. As with the air heat-treatments, formation of the oxide layer at the Cu-CusSn interface was observed in both the 1978 and the 1979 samples whereas none was observed in the 1977 sample. Development of the oxide layer from heat-treatment in an inert atmosphere clearly showed that the source of oxygen was located within the copper plating. Organic addition agents, codeposited with the copper during the plating process, are believed to be the contributing source for the oxygen which is detected at the interface (6). This view is indirectly supported by the observation that the 1977 sample (where the copper was deposited in a pyrophosphate bath) did not show development of the oxide, whereas the 1978 and 1979 samples (where copper deposition was performed in a copper sulfate bath) showed extensive oxide formation at the Cu-Cu3Sn interface. The irregularly shaped, jagged Cu-Cu~Sn interface observed for the 1977 sample in Fig. 2 is indicative of copper deposition with limited use of leveling (addition) agents. This contrasts sharply with the smooth matrix-intermetallic interfaces observed for the 1978 and 1979 samples which clearly suggest a greater role for the leveling agents in the copper sulfate bath. The reduced level of irregularity at the Cu-Cu3Sn interface for the 1977 samples in Fig. 3 and 5 was believed to be related to effects resulting from extended high temperature ther- Fig. 5. Microstructures after 165~ 240 hr argon heat-treatment. Oxide development similar to that observec~ in Fig. 3

5 Vol. I28, No. 2 OPTICAL AND AUGER MtCROANALYSES 383 mal exposure of the samples. It is possible that the treatments that were used resulted in decomposition of the organic addition agents and this may have permitted the development of a smoother Cu-Cu3Sn interface with continued growth of the intermetallics. A bulk composition analysis, using Auger instrumentation, of the copper layer in the 1977, 1978, and 1979 samples indicated oxygen and carbon levels of about 3 to 5 atomic percent in the as-received condition for the 1978 and 1979 samples and an undetectable level of oxygen for the 1977 sample in agreement with the above interpretation of observed data. Carbon was not detected in the 1977 sample. All of this supported the view that organic addition agents might well be the source of the observed oxygen. These analyses were performed at a depth of about 1000A to reflect actual bulk composition. (Depth profiles of the copper showed that the composition values stabilized after penetration by the Auger beam of a few hundred angstroms of thickness from the surface.) Examination of the composition after the 264 hr treatment in air (also at a depth of 1000A) showed a much lower level of oxygen and almost no carbon in the 1978 and 1979 samples suggesting possible migration of these elements from regions close to the interface to the reacted zone under the influence of thermal activation. Considerations such as solder composition, carbon contamination of the interface, and quality of the copper, can all, conceivably, to some extent, be expected to influence the rate of degradation of solder adhesion. Investigations at this Laboratory, however, suggest that the degradation mechanism itself will not be significantly altered by these several variations. The role of solder composition will be limited mainly to influencing the kinetics of growth of the various compounds and, therefore, to the accumulation rate of the oxygen-containing species. A detailed kinetics study showing the activation energy variations, for the several compounds, with changes in solder composition, is the subject of a separate publication (7). As for carbon contamination of the interface, this is expected to manifest itself in a manner similar to that observed for oxygen in this study. It may be noted, however, that even though almost equal amounts of carbon and oxygen were detected for the bulk composition of the as-received 1978 and 1979 boards, the distribution of these elements in the interface region was markedly different. Whereas oxygen was found to exist over the entire interface, where the oxygencontaining particles were distributed, high concentrations of carbon were detected only in a few discrete locations. This indicated that while carbon was also being rejected from the reaction zones in a manner similar to the oxygen (accounting for the observed buildup in concentration) its role in the observed degradation was minimal, albeit finite. The role of quality of the copper, however, is somewhat more complex. In addition to solder separation occurring from buildup of occluded impurities, such as those discussed in this paper, microstructural details, in particular the grain size, can be expected to also influence the rate of solder separation. The kinetics of growth of the several compounds can be expected to be dependent on grain size variations especially if this is the slow stage in the overall growth process. Conclusions This study shows that every effort should be made to eliminate the source of oxygen in the copper plating to avoid solder-adhesion failures in printed circuit boards. There is reason to suspect that organic addition agents that are codeposited with copper are acting as sources of the observed oxygen. The oxygen accumulates as discrete oxide particles at the matrixintermetallic boundaries because of oxygen exclusion from the reacted zone with the continued growth of the intermetauics. Increase in oxide-particle concentration leads to solder separation, possibly through inhibition of the Solid-state diffusion reaction across the Cu-Cu, Sn interface even while the diffusion reactions at the Cu3Sn-Cu6Sn~ and the Cu6Sns-(Pb-Sn) solder interfaces continue to proceed. Acknowledgments The authors wish to thank M. Seilonen for help in obtaining some of the Auger analyses and A. Lattanzi for continued encouragement of this effort. The keen interest of G. Ives in this activity is deeply appreciated. Manuscript submitted June 3, 1980; revised manuscript received Aug. 29, Any discussion of this paper will appear in a Discussion Section to be published in the December 1981 JOURNAL. All discussions for the December 1981 Discussion Section should be submitted by Aug. 1, Publication costs of this article were assisted by The Charles Stark Draper Laboratory, Incorporated. REFERENCES 1. E. Toledo, S. Bonis, J. Breen, and F. Oberin, Circuits Manufacturing, p. 64 (January 1970). 2. L. Zakraysek, in Froceedings of llth Annual Reliability of Physics Conference, p. 6 (1973). 3. R. B. Keyson, Paper presented at 5th Contamination Control Seminar, Hughes Aircraft Company, E1 Segundo, California, October "Metals Reference Book," 5th ed., C. ft. Smithells, Editor, Butterworths (1976). 5. "Handbook of Auger Electron Spectroscopy," Physical Electronics Industries, (Perkin-Etmer) (1976). 6. T. Fulton, E. Toledo, and J. Breen, Tech. Report No. SR , Raytheon Company, Equipment Division, January 1970 (Unpublished). 7. K. Kumar, A. Moscaritolo, and M. Brownawell, Unpublished.