Argon plasma cleaning prior to wire bonding is demonstrated to reduce destructive pull strengths for one mil gold wires bonded to thick film gold hybrid substrates. The primary cause of degraded bonding is identified as copper fluoride, concentrated in gold grain boundaries and spreading across grain surfaces. The substrates were apparently fired in the presence of chlorine and plasma cleaned in a chamber containing Teflon.

The controlled bonding and pulling experiment used to quantify bonding degradation included two different pastes fired bye the suspect source, and control substrates fired by one of the paste vendors. Fixed geometry bonds and the same bonding equipment and parameters for the duration of the investigation eliminated potential variations in bond integrity not caused by substrate changes due to plasma cleaning.

Degraded substrates examined in a Scanning Electron Microscope (SEM), exhibited grain boundary nodules which spread across the grain with multiple plasma clean cycles. Auger Electron Spectroscopy (AES) and Electron Spectroscopy for Chemical Analysis (ESCA) identified, copper, and gold fluorides on plasma cleaned, with copper fluoride the key compound causing bonding degradation.

Teflon dust and dendrites observed around Teflon bushings in the plasma cleaning chamber indicated attack bye the argon plasma, which liberated fluoride to react with copper chloride in the gold paste during cleaning. Copper chloride was formed when the paste, which contains copper oxide for substrate adhesion, was screened or fired in a chlorine contaminated atmosphere.

Key Words: Plasma, bond nodules, Fluorides, grains
 

Introduction

A reduction of one mil gold wire bond integrity following argon plasma cleaning of certain alumina substrates in production was reported. Experimentation performed to identify the degraded substrates is discussed in this paper. Efforts to quantify the impact on bond integrity and subsequent analysis performed to identify the nature and origin of the physical phenomenon, are described.

Identification of condition

Production personnel observed reduced average bond pull strength and increased number of crescent lifts for a large percentage of substrates for a certain substrate vendor after multiple plasma cleaning cycles. The same substrates bonded acceptably with no plasma cleaning, while substrates for other vendors bonded acceptably following plasma cleaning. An unknown reaction to plasma cleaning was suspected and investigated.

Bond strength degradation was quantified with a controlled bonding/pulling experiment. All subject substrates were fired by the same substrate vendor, six with paste A, six with paste B. Half of the substrates were screened with conductive epoxy in the appropriate locations for production, resulting in four groups of three substrates listed in Table 1. Two substrates fired by paste vendor B, one with gold on alumina and one with gold dielectric, were used as controls.

Degradation of substrate condition for bonding was of primary concern. Variables in the bonding operation can impact bond strength on identical substrates, so fixing those variables was important.

Table 1. Subject substrates

Epoxy No Epoxy Paste A Serial Numbers     Paste B Serial Numbers   234 242 274   33 96 237 4 55 239   17 112 253   Controls (fired by paste vendor B, not substrate vendor)   1 Gold on Alumina   2 Gold on Dielectric

Bond wire geometry and pull location determine distribution of pull stresses to ball and crescent bonds. Ball-to-crescent separation was fixed at 80 mils, with a loop height of 35-40 mils. Pull location was established at the center point between ball and crescent equalize stress on the bonds. The long ball-to-crescent separation allowed plus or minus 5 mils error in pull location with minimal impact on stress distribution.

One operator used the same bonding, pulling, and plasma cleaning equipment for the duration of the investigation. Bonding parameters were established using a new substrate. Bond failures had occurred most frequently at the second bond so force, time and power levels for the second bond were reduced until crescent lifts occurred, then increased just until crescent lifts were eliminated. The intent was to maximize detection of the effect of plasma cleaning on substrate condition. These initial parameters were maintained for all bonding iterations. All bonds were substrate to substrate.

Fourteen bond were placed on all of the substrates and destructively pull tested. The absence of bond lifts established the good bonding conditions of all substrates. The wire break strengths are not instructive in evaluation of substrate condition. Iterations of production cycle plasma cleaning bonding and pulling followed for two of each three substrates, with the third held in reserve for analysis.

Initial bond pull strengths averaged six to eight gram-force for paste A, with no bond lifts. Paste A substrates with and without epoxy exhibited zero to thirteen crescent lifts after two plasma cleans, and eight to thirteen crescent lifts after three plasma cleans, Results were similar for cycles were required to generate the same number of crescent lifts. The limited sample size did not show a difference caused by the presence of epoxy, and the increase in crescent lifts was not linear with increasing plasma clean cycles. The number of crescent lifts and average crescent lift strength at each plasma clean cycle are presented in Table 2.

No substrates suffered crescent lifts prior to plasma cleaning, and the control samples bonded and pulled the same after six plasma cleans. Crescent lifts after plasma cleaning, therefore indicate some degradation of bonding integrity caused by a change in substrate condition. Degradation varied form barely detectable for serial number 17 to dramatic for serial numbers 4, 55, and 253, indicating that plasma cleaning impacts substrates differently. Subsequent analysis confirmed differences in initial substrate condition which account for the differences.

Analysis of Substrates

Scanning Electron Microscopy (SEM) was employed to examine surface topography as received and following plasma cleaning and other conditioning steps. Figures 1 and 2 show paste B and control surfaces as received with gold grain structure evident.

The two substrates have the same paste, fired at either the substrate vendor (Figure 1) or the paste vendor (Figure 2) facility. Grain structure is similar, while the control substrate is free of the surface debris seen in figure 1. The difference is attributed to firing conditions. Energy Dispersive Spectroscopy (EDS) detected gold in the grains, and primarily gold, copper, and bismuth in the grain boundaries. These elements are known constituents of the thick film paste. In figure 3, a control substrate appears to be unchanged after six plasma cleans. The surface of serial number 17, in Figure 4, has developed some nodule growth after four plasma cleans. This phenomenon, widespread on the substrate was centered on the grain boundaries. The nodules measured about 0.05 micron across. The nodules were larger and more prevalent on serial number 253 after nine plasma cleans. The irregular shape and approximately 0.1 micron size of the nodules is apparent in figure 5.

An EDS scan on the nodules detected no difference from a clear gold area, indicating the thickness was 0.1 micron or less and therefore not discernable for the underlying material by EDS. Several substrates examined as received, using the SEM, exhibited no evidence of the grain boundary nodules which were observed on all plasma cleaned substrates except the controls.

Accelerated bonding degradation was loosely associated with greater concentration of grain boundary nodules.

Surface analysis was more instructive than SEM?EDS as to the elemental and chemical composition of the gold surfaces. Fixed beam Auger Electron Spectroscopy (AES) was useful in comparing the elemental composition of the surfaces before and after plasma cleaning. Percentages are not provided because of surface sensitivity and varying excitation efficiency concerns. However, comparison of relative concentrations form specimen is valid. Unless otherwise specified, results reported were taken after five seconds sputter etching to remove atmospheric contamination. Table 3 presents a summary of fixed beam AES results. The left column lists the elements detected on the surface prior to plasma cleaning. The right column lists the elements after that substrate's final plasma clean cycle, with notes of how the relative concentration of that element changed from air and even careful handling are detected on most samples. Metallic elements are most likely constituents of the thick film pastes, the exact compositions of which are proprietary.

The control substrate analyzed in Table 3 experienced no crescent lifts after six or more plasma cleans, while crescent lifts were prevalent for the paste A and B substrates. Note that carbon oxygen or both are reduced for all three substrates following plasma cleaning. This is consistent with the purpose of argon plasma cleaning: to remove organic contamination. Notice also that fluorine was detected on both degraded substrates but not on the four good-bonding substrates.

Scanning Auger Microprobe (SAM) was employed to determine the relationship between the grain boundary nodules observed in the SEM and the AES spectra for degraded plasma cleaned gold. The nodules contained copper, chlorine, fluorine, sulfur, carbon and oxygen, Spectra form apparently clean gold grain areas detected the same elements in lower concentrations, plus gold. The nodules are buildups of material which originates in the grain boundaries and spreads across the grain surfaces. SEM examination of substrates after increasing numbers of plasma cleans, presented earlier in Figures 4 and 5, confirms this fact.

Table 3. Auger electron spectroscopy Summary
 

As Received	Plasma Cleaned
Control Substrate     Au     C     O     Cl     Ag     S     Cu     N	Au C O reduced Cl reduced Ag S Cu increased N
Paste A     Au     C     O     Cl     Ag     Cu     F not detected	Au C reduced O reduced Cl increased Ag Cu F
Paste B     Au     C     O     S     Cl     N not detected     Cu     F not detected	Au C reduced O reduced S not deteced Cl increased N Cu F

Electron Spectroscopy for Chemical Analysis (ESCA) identified fluorine and chlorine as the key Elements in substrate degradation. ESCA data presented in Table 4. Elemental percentages are useful here in comparison from substrate to substrate, and in a limited capacity for stoichiometry.

Table 4. ESCA data: surface concentrations - atomic percent trace elements deleted so totals do not equal 100%.
 

Paste A Paste B Element 0 Plasma Clean Post Plasma Clean 0 Plasma Clean Post Plasma Clean Cl F Cu O C Au 2 1 6 37 33 13 1 15 14 33 27 5 3 0 3 25 37 22 1 18 3 28 25 16

Increased levels of fluorine were detected by ESCA over AES because the fluorine compound bonds were broken by the AES electron beam, liberating fluorine to reduce detection. The ESCA x-ray excitation was less damaging to fluoride molecules, so more was detected. Fluorine concentration was the most significant change caused by plasma cleaning. Copper and gold fluorides were significant artifacts on the plasma cleaned, degraded bonding surfaces. Angular Resolve ESCA analyzes 1-5 monolayers, while horizontal ESCA gathers data from 20-30 monolayers. Surface composition data and binding energies obtained using both ESCA techniques facilitated determination of which compounds in what approximate proportions, were present on each surface. The metallic compounds are summarized in Table 5.

Table 5. Metallic Compounds on Substrates
 

Metal	Cu	Au
Paste A   0 Plasma      Clean   Post      Plasma      Clean   Paste B   0 Plasma      Clean   Post     Plasma     Clean	60% CuO    40% Cu2O    60% Cu2O    40% Cu (FOH)2      63% Cu2O    37% CuO    60% CuO    40% Cu (FOH)2	Metallic Metallic       Metalic   AuF3

Bonding degradation was caused by copper fluoride for both pastes. Considering AES and Table 4. Paste A was formulated so that more copper was present on the surface than on paste B, so ultimately  copper fluorides dominated the plasma cleaned surface. Paste B exhibited gold and copper compounds. The literature on copper fluorides does not include standard ESCA spectra for comparison to identify compounds. Two likely candidates, copper difluoride dihydrate and copper hydroxyl fluoride, were therefore synthesized for comparison. The material on the substrates was thereby identified as copper hydroxyl fluoride (Cu(FOH)₂). Gold fluoride standards, also missing from the literature, were impractical to synthesize, so existence was inferred from energy shifts I the gold fluorine peaks, and by the percentage of fluorine remaining after accounting for the copper fluoride.

Discussion

The thick film gold pastes contain 1-2% copper oxide, which is concentrated in the grain boundaries upon firing. Compounds grew out from the grain boundaries during plasma cleaning, confirming the grain boundary oxides as the root source of copper in the copper fluoride. Teflon bushings and rack supports in the plasma cleaning chamber exhibited growing Teflon dendrites and Teflon dust on the floor of the chamber below, indicating the argon plasma attacked the Teflon, releasing free fluorine into the chamber during cleaning.

Copper oxide is present in all the substrates, yet all are no degraded by plasma cleaning: therefore copper fluoride is not formed by a reaction of fluorine with copper oxide. Fluorine will, however react with copper (cuprous) chloride to form copper difluoride, which forms copper hydroxyl fluoride when exposed to atmosphere. Referring back to Table 3, one can see that Chlorine concentration is reduced by plasma cleaning for the control  substrate and increased by plasma cleaning for paste A and paste B substrates. The inferred source of chlorine is, therefore hydrochloric acid in the atmosphere at the substrate vendor. Chlorinated solvents in contact with humidity readily form hydrochloric acid, as has been observed in vapor degreasers. Cupric chloride (CuCl₂) is formed at room temperature, and converts to cuprous chloride (CuCl) during the high temperature firing. Cuprous chloride may oxidize on the surface , resulting in the ESCA spectra obtained, while remaining chloride below the surface. Electrostatic forces generated by the plasma cause the chloride to migrate to the exposed grain boundary surfaces where reaction with free fluorine form the Teflon forms copper fluoride.

Conclusions

The substrates degraded by plasma cleaning bonded acceptably as received, which is how they were evaluated in incoming inspection. This is common in production environments, so that vulnerability to plasma cleaning or other processes is not revealed before production yields are impacted. Incoming inspection associated with any process is of value only if materials are evaluated consistently with their intended eventual use. Failure to appropriately evaluate incoming materials will ultimately result in reduced production yields.

Atmospheric chlorine contamination during substrate processing resulted in the formation of copper chloride in thick film gold pastes formulated with copper oxide. The copper chloride, exposed to energetic free fluorine during plasma cleaning, formed copper fluoride, which is sufficient quantity impeded gold wire bonding. Teflon inside an argon plasma cleaning chamber was attacked by the plasma, which liberated free fluorine to form the fluoride compounds. An anomaly in the substrate processing caused a latent condition in the thick film gold. Argon plasma cleaning, a normally benign process, provided the activation stimulus for the latent substrate condition, and bond integrity was reduced.

A complex mechanism such as  the one described here often cannot be understood through the use of one analytical tool. Statistical examination of the controlled experiment data confirmed plasma cleaning as part of the mechanism. The detection of fluorine by AES indicated a trace element of the degraded substrates. Following up with ESCA revealed fluorine to be a key element on the degraded substrates. Following up with ESCA revealed fluorine to be a key element in bond strength degradation. SEM imaging identified the grain boundary origin of the fluoride nodules, strengthening the conviction that components of the gold paste were migrating to react on and degrade the surface.

Acknowledgments

Several other Raytheon employees contributed to the bonding investigation and subsequent analysis:
 Steve Hein of Research Division
 Dr. Carl Miller of Equipment Division
 Larry Pitman of Missile Systems Division
 Bob Zalanskas of Missile Systems Division

Home | News | Applications | Lab | Plasma Equipment | Service | Contact | Site Map