The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation

Transcription

1 Journal of ELECTRONIC MATERIALS, Vol. 31, No. 11, 2002 Special Issue Paper The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation W.J. READY 1 and L.J. TURBINI 2 1. MicroCoating Technologies, Atlanta, GA judready@hotmail.com 2. Centre for Microelectronics Assembly and Packaging, University of Toronto, Toronto, ON M5S 3E4 Conductive anodic filament (CAF) formation, a failure mode in printed wiring boards (PWBs) exposed to high humidity and high voltage gradient, has caused catastrophic field failures. This study quantified the effect of flux chemistry, applied voltage (V), spacing (L), and temperature on the failure rate. Test vehicles, which had hole-to-hole spacing of 0.5 mm or 0.75 mm, were processed with one of three water-soluble fluxes (WSFs), and a heated control was also evaluated. The samples were placed in a temperature humidity chamber at 85%RH, at one of three temperatures: 75 C, 85 C, or 95 C. A voltage of 150 V or 200 V was applied to the test vehicle and periodically removed so that a measurement could be taken. A specially designed linear circuit was used to determine when the insulation resistance dropped significantly, indicating a failure. Activation energies were determined. The mean time to failure was a function of L 4 /V 2. Key words: Flux chemistry, printed wiring boards, filament formation PURPOSE The purpose of this research was to investigate the factors that enhance conductive anodic filament (CAF) formation. The variables studied were (1) water-soluble flux (WSF) formulation, (2) conductor spacing, (3) operating voltage, and (4) temperature. Quantification of the effect of each variable was determined through a series of accelerated life tests (ALTs), with each ALT consisting of 32 printed wiring boards (PWBs). An additional goal of this research was to acquire a fundamental understanding of the morphology and chemical composition of CAF. BACKGROUND Beginning in the mid-1970s and continuing through today, the wiring density of electronics has increased rapidly. Electronic equipment is also required to operate in outside areas where elevated temperature and humidity are uncontrolled. While investigating the reliability concerns that might result from these conditions, researchers at Bell Labs discovered a new failure mode. 1 7 This failure mode is characterized by an abrupt, unpredictable loss of insulation resistance (IR) between conductors that (Received February 13, 2002; accepted May 21, 2002) are held at a potential difference. The filament, now termed CAF, is a result of an electrochemical migration (ECM) process that initiates at the anode and proceeds along separated fiber/epoxy interfaces Just prior to this 1976 definition of CAF, Der- Marderosian 12,13 had noted a similar failure mode. They observed that a filament penetrated between two different layers in an MLB from anode to cathode. They termed this failure mode punch-thru. Punch-thru is similar to CAF, except the filament grows between circuit layers rather than along a fiber. The CAF should not be confused with dendrite growth. In dendritic growth, metal ions go into solution at the anode but plate out at the cathode, growing in needle- or treelike formations across the surface of the PWB. In contrast, CAF growth emanates from the anode. Furthermore, CAF contains copper and a halide ion (typically chloride) rather than pure metal as in the case of dendrites. Additionally, dendrite growth is a surface phenomenon, while CAF is a subsurface phenomenon. In a typical PWB, there are surface tracks of copper metallization and copper plated through holes (PTH) that provide component insertion points as well as electrical continuity between the PWBs top and bottom surfaces. A CAF may bridge two surface 1208

2 The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation 1209 Numeric Models of CAF Formation Welsher et al. 5 determined that, when the PWB is exposed to thermal transients, such as during multiple soldering steps, the time to failure associated with CAF decreased by nearly a factor of 2. The application of an intense thermal transient served to speed the debonding of the fiber/epoxy interface as a result of a mismatch between the coefficients of thermal expansion (CTE) of the fiber and epoxy. They also reported that the mean-time-to-failure (MTTF) obeys an Arrhenius relationship in the C range. Furthermore, the humidity dependence seems to be material lot dependent. Time to failure (t f ) results show a strong dependence on relative humidity that roughly obeys Eq. 1 and is inversely proportional to applied voltage according to Eq. 2. t f a(h) b (1) Fig. 1. Schematic representation of CAF pathways (dimensions exaggerated for clarity): (a) PTH to PTH, (b) PTH to track, (c) track to PTH, and (d) track to track. tracks, two PTHs, or a PTH and a surface track. Figure 1 details these pathways graphically. The subsurface deposits of corrosion byproducts emanate from the anode and eventually progress toward the cathode via a separated fiber/epoxy interface. The initial Bell Labs researchers 1 6 detailed a model for the mechanism by which CAF formation and growth occurs. The first step is a physical degradation of the fiber/epoxy interface. When moisture absorption occurs, it creates an aqueous medium along the separated fiber/epoxy interface that provides an electrolytic pathway and facilitates the transport of corrosion products. The second step, electrochemical corrosion, results because the absorbed water acts as the electrolyte, the copper circuitry becomes the electrodes, and the operating voltage serves as the driving potential. The second step occurs sequentially rather than in parallel with the first step. Lando et al. 2 determined that the filaments appeared most often in the PTH-PTH test pattern configuration (Fig. 1a), apparently as a result of the direct contact of copper and fiber/epoxy interface resulting from the hole drilling and plating process. The tracktrack configuration (Fig. 1d) was the least susceptible to filament growth, and increasing the spacing between conductors lowered susceptibility further. 5 The addition of a buttercoat (a layer of epoxy resin without fiber reinforcement between the conductors and the PWB surface) further decreased CAF formation frequency emanating from surface tracks. This reduction is apparently a result of the lack of fiber/copper or fiber/surface interfaces that could initiate debonding and lead to filament growth. Finally, the frequency of filament growth in the track-pth (Fig. 1b) and PTHtrack (Fig. 1c) configurations had intermediate lifetimes compared to the other conductor configurations. t f c a d V b where H is the relative humidity; V is applied voltage; and a, b, c, and d are positive material-dependent constants. [Note: In equations, to prevent confusion with the universal gas constant, R, (8.314 J/K mol), the more common RH designation for relative humidity will be designated simply as H. However, RH will be used in the text to represent relative humidity.] Welsher et al. 4 performed experiments on PTH- PTH patterns with a 75-mil spacing biased at 200 V to ascertain the effect of temperature, humidity, applied voltage, conductor spacing, and substrate material choice. They identified a composite of triazine/glass as a resistant material. Triazine is a thermosetting resin made by reacting diglycedyl ether of bisphenol A (DGEBA) with a cyanogen halide rather than reacting it with dicyandiamide (DICY) as in FR-4. They found that lifetimes were 20 to 30 times greater for triazine as compared to typical FR-4. Welsher et al. 4 performed delayed bias application tests that confirmed the two-step sequential (rather than parallel) process proposed for CAF formation. In particular, the rate-limiting step involved fiber/epoxy interfacial degradation and moisture absorption. This degradation is a function of temperature and relative humidity. The interfacial degradation is followed by a very rapid electrochemical corrosion step that is an inverse function of the electric field (V/L) as well as a function of temperature and humidity. They also understood that the spacing between conductors (L) must be bridged by the filament so that the time for the second sequential step will be proportional to L 2 /V. Their data confirmed that CAF in PWBs, particularly those with closely spaced PTHs, is a potentially serious reliability problem. Combining Eqs. 1 and 2, they developed the model shown in Eq. 3 for the MTTF due to CAF: MTTF a(h) b expa E a b dal2 (3) k b T V b where E a is the activation energy for the process, T is temperature (in Kelvin), V is the applied voltage, L is the spacing between conductors (in mils), k b is (2)

3 1210 Ready and Turbini Boltzman s constant ( ev/k), a and b are material dependent constants, and d describes the temperature and humidity dependence. Mitchell and Welsher 6 further developed the model to accommodate different conductor orientations (Fig. 1). If the temperature and humidity dependencies for each step in the CAF process are equal, then they can both be approximated by a single constant a(h) b exp (E a /k b T). They produced a revised MTTF due to the filament formation equation that is given in Eq. 4. Ln MTTF a1 (4) V b H expa E a k b T b where and are material-dependent constants, is a humidity-dependent factor, and n correlates with the orientation of the conductors and has a value of four for the PTH-PTH orientation. Gandhi et al. 9 stated that the voltage dependence is closer to an inverse voltage squared or inverse voltage cubed relationship, rather than the inverse linear relationship suggested by Eq. 2, Eq. 3, and Eq. 4. Lahti et al. 3 showed that for unprocessed PWBs below 60 C, the failure mechanism was not thermally activated and the E a for CAF formation was between 0.0 ev and 0.2 ev. Above 65 C, the E a was between 1.0 ev and 2.5 ev, which is a much stronger temperature dependence than was observed below this temperature. Based on this change in E a, they concluded that PWB lifetimes exceeding twenty to thirty years should be achievable under normal (25 30 C and 40 50% RH) conditions. However, they also discovered that failure rates greatly increased at a critical conductor spacing of 5 mils under their 85 C/85%RH/45 V test conditions. Research at the CALCE Electronic Packaging Research Center used a physics-of-failure approach and found that several factors determined the formation of CAF (CALCE research typically uses the term conductive filament formation [CFF]): the operating conditions (humidity, temperature, and voltage), the laminate choice, and the spacing and geometry of the conductors. They developed a model (Eq. 5) to reconcile these various factors into a single equation: a f (1000 k L)n t f a (5) V m b (M M t ) where M is the percentage moisture content, M t is the percentage moisture content threshold, a is the filament formation acceleration factor, f is a multilayer correction factor, n is a geometry accelerating factor, m is the voltage accelerating factor, and k is the conductor shape factor. A best-fit approach was used to obtain the values for a, f, k, n, and m from their experimental data. They concluded that the formation of CAF is most highly dependent on the operating voltage and moisture content of the laminate. Though it may appear that the Arrhenius temperature dependence was excluded from their study, the temperature dependence has been incorporated into the M t parameter. The CALCE work also noted that hollow e-glass fibers may pose a preferential pathway for the metal migration. 18 A substantial amount of their results focuses on the susceptibility of MLBs to filament formation, particularly under temperature and humidity cycling. The cycling serves to accelerate the debonding of the fiber/epoxy interface. They noted this was particularly deleterious when a PTH was adjacent to the debonding. 19 Augis et al. 7 concluded that the linear accelerating factor extrapolations for the very high stress testing used in earlier work by Bell Labs researchers were not valid. This is because the models based on the linear acceleration factors would have predicted a 6% failure rate of a particular product after 5 years. They observed that this catastrophic failure rate was not even remotely evident and thus the model was wrong if applied under the wrong circumstances and environmental conditions. They used demarcation maps to understand the degradation mode and proved that a humidity threshold exists. They also noted a wide variability between different lots of material. They suggest that there is a relative humidity threshold, below which CAF formation and growth will not occur. They base this suggestion on the many factors identified by the early Bell Labs researchers and additional information such as the reversibility of the hydrolysis of the silane-coupling agent. 24 They found that, for a 50 V circuit operating at 25 C, the critical relative humidity for CAF formation was near 80%. Additionally, they developed a quantitative model shown in Eq. 10 that predicted this critical threshold value: ln(c) 0.9 k B T H ln(v) The constant c (a measure of the rate of CAF formation) was found to be for a failure probability of 0.5%. Voltage-Spacing Influence on CAF Formation A critical factor used in determining CAF susceptibility is the voltage gradient that the assembly experiences. By normalizing the applied voltage over a standard distance, a comparison between different assemblies with different pitch and operating voltages can be established. This normalization procedure is most graphically demonstrated by example. An assembly containing a 20 V power plane with an adjacent PTH biased at 20 V has a potential difference of 40 V. This is not considered a high potential difference for an electronic assembly. However, the 5-mil separation between the power plane and the PTH creates a voltage gradient of 8 V/mil. Figure 2 reveals the circuit failure that occurred when a CAF bridged the 5-mil spacing. 10 Minimizing voltage gradients is essential to reducing CAF (6)

4 The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation 1211 Fig. 2. Catastrophic failure due to a CAF blowout of layers 6 and 7 of a 14-layer MLB. formation. Using the physics-of-failure approach, CALCE researchers showed 20 that the velocity of the migrating ions is proportional to the ionic mobility and the voltage gradient. Polyglycol and WSF Influences on PWBs and CAF Formation It has been shown that polyglycols diffuse into the epoxy during soldering and degrade insulation resistance This absorption occurs when the PWB is above its T g, where elevated temperatures and a more open structure facilitate diffusion. This polyglycol absorption has been shown to reduce performance by increasing moisture uptake by the substrate. 28 Jachim et al. 29 were the first to link the use of polyglycols in soldering fluxes and fusing fluids to increased susceptibility to CAF formation. Furthermore, Ready et al. 10 detail a field failure, which occurred on only certain production lots. They show that this failure resulted from the use of a polyglycol containing hot air solder leveling (HASL) fluid during production. This fluid also contained hydrobromic acid that diffused into the brominated epoxy substrate resulting in an increased bromide concentration in the PWB. Both of these constituents increased the assembly s propensity for CAF formation by enhancing moisture absorption and providing an appropriate anion for the electrochemical reaction. Diffusion of polyglycols into the PWB occurs during soldering. Since the diffusion process follows an Arrhenius behavior, the length of time the PWB is above the T g will have an effect on the amount of polyglycol absorbed into the epoxy and that, in turn, will affect its electrical properties. Brous 26 linked the level of absorbed polyglycol in a PWB to surface insulation resistance (SIR) measurements. Jachim et al. 29 reported on WSF treated PWBs that were prepared using two different thermal profiles. Those which experienced the thermal profile with higher temperatures exhibited a SIR level that was an order of magnitude lower than those processed under less aggressive thermal conditions. It is clear that the higher the soldering temperature, the greater the polyglycol absorption. Similarly, for each thermal excursion that occurs, the fiber/epoxy interfaces within the PWB are weakened due to different thermal expansion characteristics of these two materials. Likewise, the PTH and the PWB laminate material will experience stresses due to a CTE mismatch, which may induce delamination. Zado 25,27 was the first to identify the deleterious nature of unremoved WSF residues. In 1983, he reported on the effect of various constituents of WSFs on SIR. He noted that for WSFs, PEG was the most common flux vehicle at the time. Additionally, he determined that polyglycol esters and ethers used to process the PWBs produced a lowered SIR value. Furthermore, he proposed that PEG became hydrogen bonded to the epoxy when the PWB was heated to soldering temperatures. Upon cooling, the PEG became locked within the polymeric structure. He showed that the hydrophylicity of the epoxy increases when PEG remained in the PWB. He also showed that a change in relative humidity from 55% to 75% resulted in a slight degradation of SIR, but an increase from 75% to 85% decreased SIR by two orders of magnitude. He did not see a similar effect when polypropylene glycol (PPG) was used. He concluded that the steric hindrances between PEG and the epoxy were the primary reason that PEG exhibited these SIR decreases while PPG did not. Work done by Brous 28 at Alpha Metals (Jersey City, NJ) extended the base of understanding of the deleterious effect that WSF residues posed to PWB SIR. His research showed that SIR is affected by the (1) degree of absorption of flux constituent, (2) hygroscopicity of absorbed flux constituent, (3) volatility (vapor pressure) of absorbed flux constituent, (4) effectiveness of cleaning process, (5) temperature and humidity, and (6) presence of water soluble ions on the PWB surface. He tested several materials frequently used in WSF formulations: glycols, polyglycols, glycol ethers, polyglycol surfactants, polyols, and glycerine. He found that polyglycols (particularly those with lower molecular weights) are strongly hygrosopic and that significant amounts of moisture absorption can occur within 24 h if a critical humidity level is exceeded. Brous 28 confirmed that polyglycols diffused into the PWB during the elevated temperature of soldering operations. Furthermore, he confirmed that (in general) lower molecular weight polyglycols are more highly absorbed into the substrate. His investigations of various polyglycols showed that PWBs treated with them gave SIR values that were initially low ( at 35 C/90%RH). For many polyglycols, these SIR values rose appreciably (by an order of magnitude) over short time periods as the polyglycols diffused out of the epoxy. However, polyethylene glycol (PEG) SIR values did not increase appreciably. He believed this was indicative of a trapping of the PEG within the polymeric backbone, which prevented its diffusion out. Brous 28 further stated that if hygroscopic contaminants are present on the surface, a conductive condensed film of water will deposit across the surface, possibly initiating electrolytic oxidation at the anode and reduction of the metal ions at the cathode. He found that when the epoxy/glass laminate absorbed polyglycols, the laminate became hygroscopic. He also

5 1212 Ready and Turbini found that under exposure to high temperatures in the soldering process, many of the polyglycols appear to be capable of penetrating the surface sufficiently to be retained by the PWB through the cleaning step. Zado 25,27 studied the effect of nonionic WSF residues, such as PEG, on SIR. Test PWBs treated with PEG and cleaned in an aqueous medium showed reduced SIR values, while PWBs treated with the other flux constituents gave SIR levels equivalent to unprocessed PWBs. He concluded that the cleaning step was ineffective in removing the polyglycol. Since the polyglycol rapidly diffuses into the epoxy at temperatures above T g, diffusion out of the epoxy at temperatures below the T g is expected to take much longer. Brous 26,28 studied several solvent combinations for removing PEG and other polyglycols from PWBs and determined that soaking the PWB in acetonitrile for 24 h provides the best removal process. Ionox FCR (Kyzen) has also been successfully used to remove polyglycol residues from PWBs processed with HASL fluids. Some laboratory tests 30,31 have used 75% isopropyl alcohol and 25% deionized water at 80 C for 1 h to remove polyglycols and other ionic contaminants from processed PWBs. Bent et al. 32,33 studied PWBs processed with PEG and several other polyglycols and noted that no CAF was observed with PEG treated PWBs. Chemical Composition of Conductive Anodic Filament Formation Augis et al. 7 noted that the filaments always contain copper and sometimes contain chlorine or sulfur. Research by Ready et al. 10,34 indicated that CAFs generally contain copper and chloride but may also contain bromide when a bromine containing WSF is used. Meeker and LuValle 35 also identified that other anions could be present in the CAF. Most notably, they identified bromine and sulfur ions as being present in some specimens. CALCE research 17 mistakenly identified an x-ray peak (Figure 6 of their work) as a CAF containing aluminum, when in fact the CAF most likely contained bromine. This misidentification has occurred in other research 36 and arises because the energy of the bromine L 1 x- ray line is at kv, while the aluminum K 1 x- ray line is at kv. This difference is so small that it is virtually impossible to distinguish the two elements. In order to differentiate between these two elements, it is necessary to look for the bromine K 1 line at kv. Improved Measurement System to Detect CAF The SIR measurement method presently specified in J-STD is an ECM test, which should be able to detect both surface SIR degradation and subsurface CAF formation. The test method has two major limitations. 38 (1) The test method takes SIR measurements at discrete and widely separated intervals (i.e., 24, 96, and 128 h). (2) It does not provide for a prompt termination of bias when ECM causes an electrical short circuit. This allows dendrite or CAF blow-outs (similar to Fig. 2) to occur destroying the short and preventing characterization. A linear circuit (LC) has been developed 39 to overcome the deficiencies of J-STD-004. Figure 3 is an electrical schematic of the LC. The IR of the test pattern (i.e., interdigitated comb, PTH-PTH, etc.) is depicted as a variable resistor (R ECM ). The LC is essentially an inverting operational amplifier or opamp (Fig. 4). 40 In the LC, the feedback resistor (R f ) is 1.0 M and the input resistor (R 1 ) is equal to 4.7 k plus R ECM and the additional resistors and capacitors are filters used to reduce high and low frequency noise. The theoretical gain (V out /V in ) of the LC is plotted in Fig. 5. A test pattern with good resistance characteristics will have a high initial IR (R ECM ). As ECM forms and bridges between the anode and cathode, the value of R ECM decreases. The value of R ECM will drop to near 0 at the instant of shorting. This drop (and resulting decrease in op-amp gain) causes the LC to reduce the current flowing through the ECM so that the dendrite or filament does not blow out. EXPERIMENTAL PROCEDURE Samples A specialized PWB (Fig. 6) was created for this research. The nominal distance between the points of closest approach between the two PTH barrels was 0.5 mm or 0.75 mm. The PWB is manufactured with 1 oz. of copper and the thickness of the copper surface features is approximately 70 m. Fig. 3. Electrical schematic of linear circuit. Fig. 4. Inverting op-amp.

6 The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation 1213 Fig. 5. Theoretical gain of linear circuit (Fig. 3) as a function of R ECM. Fig. 6. PWB with variable PTH-PTH spacing. A Zero Ion System containing a mixture of 75% isopropyl alcohol and 25% deionized water at 25 C was used to pre-clean the PWBs to a resistivity of 150 M /cm 2. The PWBs were subsequently handled with latex gloves to minimize any new contamination. Three WSF formulations were tested. The WSFs were selected based on their propensity to form CAF during SIR testing. 32,33 Table I outlines the composition of each WSF. Eight PWBs were prepared with each WSF and eight control PWBs were included in each ALT (i.e., 32 PWBs per ALT). The four cathodic test sites on each PWB (PTHs on right in Fig. 6) were ganged together with a connecting wire so that a failure at any single site would register as a failure for the entire PWB. For each ALT, a pipette was used to dispense 100 L of flux per PTH-PTH test point (i.e., 400 L per PWB). The PWBs were reflowed in a bench-top OK Industries (Mt. Vernon, NY) model JEM-310N convection reflow oven with a profile that reached a maximum temperature of 205 C. The post-reflow PWB cleaning procedure consisted of a vigorous tap water rinse at room temperature for 1 min. This rinsing was followed by a 5-min ultrasonic cleaning with deionized water at 65 C in a Branson (Danbury, CT) 5210 ultrasonic bath. A final deionized water rinse completed the cleaning process. This manual method was used due to the unavailability of the common, industrial scale aqueous in-line cleaner. Consistency of the manual cleaning process from sample to sample and from batch to batch will most likely be lower than that found in the automated system. Accelerated Testing Teflon-coated connecting wires were attached to each PWB using rosin-cored solder. The solder joints for the wires were placed well away from the WSF test points so that the rosin-cored solder would not impact the results. The 32 PWBs were then placed in a Thermotron model SM-4S-SH temperature/ humidity chamber. The connecting wires were passed through a side access port of the chamber and connected to the data acquisition and switching apparatus described below. The relative humidity in the chamber was maintained at 85% 2% throughout the ALT. The temperature was also maintained at a constant level ( 0.5 C) throughout the ALT, but it was varied between each ALT. Temperature settings of 75 C, 85 C, and 95 C were used. Table II details the temperature settings used for each ALT. To prevent condensation, the chamber was ramped from room temperature to the desired ALT temperature. Once the temperature had stabilized at the desired level, the relative humidity was slowly ramped over about 2 h to 85%. In addition, the PWBs are allowed to equilibrate in the stabilized chamber for 24 h prior to commencing the electrification of the PWBs. Automated Measuring System Two Hewlett Packard (Palo Alto, CA) model E3612A power supplies were connected in series to apply the high voltage necessary to cause CAF growth. Each power supply can generate a maximum Table I. Flux Formulation Matrix* Code Polyglycol Type (always 20 wt.%) HCl, 37.4% HBr, 48% IPA PG3-2 Poly(ethylene/propylene) Glycol (Avg. mol. wt. 1,800) PG3-4 Poly(ethylene/propylene) Glycol (Avg. mol. wt. 1800) PG7-3 Modified linear aliphatic polyether *All components are listed in weight percent. HCl and HBr are calculated to provide halides as chlorides equal to 2 wt.%

7 1214 Ready and Turbini Table II. Temperatures for Each ALT Test ALT Maximum Reflow Number Temperature Temperature 1 75 C 205 C 2 85 C 205 C 3 95 C 205 C of 120 V at 0.25 A. Voltage noise was less than 200 mv and current noise was less than 200 ma. In order that multiple test points could be measured with a single linear circuit (LC), a switching apparatus consisting of 1 National Instruments (Austin, TX) SCXI-1001 chassis, 8 SCXI-1160 relay modules (each with 16 double-pole, single-throw electromechanical relays), and 8 SCXI-1324 wiring blocks was constructed. This chassis also contains one SCXI data acquisition (DAQ) module that makes voltage measurements of the LC output. Figure 7 schematically illustrates the switching arrangement that was applied to 32 test points (four per SCXI-1160 relay module). The 1 M resistor is known as a limiting resistor, because it limits the Fig. 7. Switching arrangement for automated LC measurement system. The figure on the left details the switch arrangement for the high bias phase of the test, while the figure on the right illustrates the switch settings for the LC test phase. amount of current flowing through the CAF at the instant of bridging. The control software, especially developed with LabView (National Instruments, Austin, TX), can display the data as a web page for remote monitoring of system and test status. In addition, the system obtains and automatically records the failure times of the various test points as tab delimited data in an Excel (Microsoft, Redmond, WA) spreadsheet. This automation facilitates the analysis of data and allows for an accurate determination of the failure distribution and MTTF. Also, since the system will automatically remove the high voltage test bias once a failure occurs, the filament is preserved for future analysis. Design of Experiment Four factors were varied in this study: (1) WSF composition, (2) voltage, (3) PTH-PTH spacing, and (4) ALT temperature. The software program CARD Pro (S-Matrix, Eureka, CA) v5.1.2 by S-Matrix was used to design the experiment and analyze the MTTF results. It also randomized (1) the PWB preparation order, (2) placement within the environmental chamber, and (3) data connection port on the DAQ equipment. If all combinations of variables were run, then 36 separate experiments would have been required (not including control PWBs). Through the use of the DOE and a quarter-fraction factorial design, this number was reduced to nine (not including control PWBs). The fractional design matrix used is presented in Table III. Eight PWBs were run at each test condition. Electrical Characterization Electrical observations consisted of failure time data obtained by the LC. During each of the ALTs, eight PWBs for each WSF chemistry and eight control PWBs (32 PWBs per ALT) were simultaneously tested to failure. The failure times were then fit to a Weibull distribution. The scale parameter was used to estimate the MTTF for each PWB set. Microscopic Characterization An Olympus SZ-40 optical microscope (magnification range of 6.7 to 40 ) was used for transmission Table III. Fractional DOE Matrix Flux Spacing (mm) Voltage (Volts) Temperature ( C) Test Number Control PG PG PG Control PG PG PG Control PG PG PG

8 The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation 1215 optical observations of the shadows of the CAF. Images were recorded with a charge coupled device (CCD) camera attached to a video capture card inside a computer. The colors of the CAF were observed using reflected light. A Hitachi (Tokyo, Japan) S-800 scanning electron microscope (SEM) with a cold field emission source was used to observe CAF morphology on polished sections of the samples. Backscattered electrons were used because compositional contrast is superior to that of secondary electrons. Energy-dispersive x-ray spectroscopy (EDS) was used to analyze the composition of the CAF. An SEM accelerating voltage of 20 kv was selected to obtain the most precise EDS data possible. A Hitachi HF-2000 transmission electron microscope (TEM) with a 200 kv field emission source was used for TEM observations. The HF-2000 is equipped with a thin window EDS for compositional measurements. The sample was prepared as an acetate replica by dipping cellulose tape into acetone and laying the softened tape across a cross-sectioned CA. Once the tape dries, it is carefully peeled from the surfaces and pulls material from the CAF. The removed material is transparent enough for an electron diffraction pattern to be obtained using standard methods. The measured radius (r) of the polycrystalline diffraction rings is related to the d spacing of the crystalline compound through L r d (7) where is the wavelength of the incident radiation ( m) and L is the camera length (1.2 m). The experimental values of d are compared to a powder diffraction database to determine the phase. RESULTS Electrical Observations In order to determine the MTTF for the PWBs due to CAF formation, an analytical technique that incorporates the entire distribution of failure times is required. The Weibull distribution contains a value, the scale factor ( ), which details the point at which 63.2% of the distribution has failed for any value that the shape factor of the distribution may take. Therefore, it accommodates all the failure points on a weighted basis according to the distribution and allows a much more accurate gauge of the failure rate than would be achieved using the mean, the median, or the mode. The scale factor is also known as the characteristic life of the PWB population. Table IV details the characteristic life ( ), while Table V details the shape factor for the specimens at 85% relative humidity and at the various WSF, voltage, spacing, and temperature conditions for the DOE ALTs. The uncertainties are given as the values and correspond to the standard error based on the Weibull distribution. Activation Energy To quantify the degree to which polyglycols, polyethers, or halides enhance CAF formation, a determination of the E a for this failure mode was made. Augis et al. 7 has shown that there is a minimum RH required for CAF formation. Thus, 85%RH (which is above the minimum RH value) was the humidity chosen for use throughout this study. By maintaining a constant relative humidity throughout the tests and using the same lot of material for all ALTs, the pre-exponential humidity-dependent factor in Eq. 3 becomes a constant, C, as shown in Eq. 8: MTTF C expa E a b dal2 (8) k b T V b Mitchell and Welsher 6 showed that the L 2 dependence above, for track-track CAF, was actually better represented by L 4 for the PTH-PTH conductor configuration. Additional researchers 9 found that the 1/V term may also be modeled as 1/V 2 or 1/V 3. Therefore, the d (L 2 /V) term should be better represented as d (L 4 /V m ), where m is a constant between one 6 and three. 9 Equation 9 depicts this modification to Eq. 8: MTTF C expa E a L4 b da k b T V mb (9) Table IV. Characteristic Life Comparison Flux Voltage (V) Spacing (mm) 75 C (h) 85 C (h) 95 C (h) Control PG PG PG Table V. Shape Factor Comparison Flux Voltage (V) Spacing (mm) 75 C 85 C 95 C Control PG PG PG

9 1216 Ready and Turbini Since all PWBs of a given WSF were run at the same voltage and spacing, an iterative methodology (Eqs ) was used to determine optimum values of E a and C based on the MTTF values at the DOE temperatures (converted to degrees Kelvin). These values were in turn used to determine optimum values of d and m. Table VI presents the summary of these iterative results. Based on the derived values in Table VI, Eq. 9 can be simplified further to Eq. 13: Microscopic Characterization of Control PWBs An SEM image of a two sequential cross sections of a CAF that formed on a control PWB is shown in Fig. 9. The micrograph shows the filament to be confined to a thin region at the separated fiber/ epoxy interface. The EDS spectrum in Fig. 10 shows that the CAF is copper and chlorine containing. The calcium, aluminum, and silicon peaks are artifacts from the interaction volume of the MTTF 1 C expa E a k b T 1 b da L4 V mb MTTF 2 C expa E a k b T 2 b da L4 V mb (10) (11) MTTF 1 MTTF 2 C c expa E a k b T 1 b expa E a k b T 2 bd MTTF C expa E a k b T b a L4 V 2b (12) (13) A plot of the MTTF results for each WSF (normalized to each other with respect to the different voltages and spacings via Eq. 13) on a logarithmic scale versus inverse temperature is shown in Fig. 8. The error bars indicate the uncertainty based on the Weibull assessment of the MTTF. The equation of the best-fit line for each data series is given. Also, the goodness of fit (R 2 ) value for each best-fit line is given. Table VI. C and E a Values Activation C Energy D (h) (ev) (V 2 /mm 4 ) m Control PG PG PG Fig. 9. SEM image of two sequential cross sections of CAF on control PWB. (a) approximately 100 m from the anodic PTH. (b) approximately 50 m from the anodic PTH. The individual E-glass fibers are labeled to aid in visualization. Fig. 8. Activation energy determination for all DOE PWBs. The dotted line in each plot indicates the relative position of the control line. Fig. 10. EDS spectrum of CAF on control PWB shown in Fig. 9.

10 The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation 1217 electron beam and the e-glass fiber. The bromine, carbon, and oxygen peaks are due to the FR-4 epoxy. The gold peak is due to the conductive coating placed on the sample to facilitate SEM analysis. Microscopic Characterization of PG3-2 PWBs Figure 11 is an SEM image of a CAF that formed on a PWB processed with PG3-2. Note the stratified morphology and the crack associated with the uppermost fiber. The accompanying EDS elemental map reveals that the CAF contains copper and chlorine. The occurrence of the stratified copper and chlorine containing morphology was universally present in all PWBs processed with PG3-2 that were observed with the SEM. Microscopic Characterization of PG3-4 PWBs Figure 12 is an SEM image showing the initiation of a CAF at the PTH barrel on a PWB processed with PG3-4. Figure 13 is an SEM image showing the initiation of yet another CAF at the PTH barrel. Note the corrosion of the copper anode adjacent to the CAF initiation point. The accompanying EDS data reveal that the CAF does not contain bromine as might be expected due to the bromide containing WSF, but rather is predominantly Fig. 12. SEM image of CAF initiation point at PTH on the PG3-4 PWB. The portion labeled Copper Land is actually the PTH Nailhead on PWB surface. Fig. 11. SEM image (a) and EDS elemental map (b) of CAF on a PG3-2 PWB. Fig. 13. SEM image (top) and EDS elemental map (bottom) of CAF initiation point at PTH on PWB processed with PG3-4. The portion labeled copper land is actually the PTH Nailhead on PWB surface.

11 1218 Ready and Turbini copper and chlorine as in earlier cases. Figure 14 shows another example of the stratified CAF morphology that was universally present on PWBs processed with PG3-4 that were observed with the SEM. Microscopic Characterization of PG7-3 PWBs Figure 15 is an SEM image of a CAF that formed on a PWB processed with PG7-3. Note the striated morphology. Other PWBs showed CAFs that had spatially varying chlorine concentrations (Fig. 16). Another CAF was seen at the initiation point with the PTH (Fig. 17). Again, note the striated morphology. Fig. 16. SEM image and EDS elemental map of CAF on PWB processed with PG7-3. Fig. 17. SEM image of CAF on PWB processed with PG7-3. The portion labeled Copper Land is actually the PTH Nailhead on PWB surface. Fig. 14. SEM image of stratified CAF on PWB processed with PG3-4. Fig. 15. SEM image of CAF on PWB processed with PG7-3. Chemical Nature of Conductive Anodic Filament The SEM/EDS data show that the CAF observed in this study is copper and chlorine containing. In an attempt to determine the exact crystalline composition of copper-bearing CAF, TEM was used to obtain electron diffraction data 41 on a CAF studied earlier. 36 When indexed (Table VII), the micrographs and electron diffraction patterns (Fig. 18) showed good correspondence to that of synthetic atacamite (2CuCl 2 5Cu(OH) 2 H 2 O). Indexing of d spacings above 2.7 Å is not possible due to overexposure of the film by the transmitted electron beam. When present in nature, atacamite is transparent to translucent and deep green in color. It displays a slender, striated, and acicular to fibrous rhombic or orthorhombic crystal structure Synthetic atacamite, or copper chloride hydroxide hydrate, is detailed in card number from the International Centre for Diffraction Data (ICDD). DISCUSSION Shape Factor Analysis In the first detailed study of CAF by Boddy et al., 1 it was found that the failure mechanism followed a log-

12 The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation 1219 Table VII. Electron Diffraction Pattern Indexing Tabulated d Spacing for Measured Calculated Synthetic Ring Radius d Spacing Atacamite (mm) (Å) (Å) Fig. 18. TEM micrograph (left) and electron diffraction pattern (right) of CAF. normal distribution for the central portions of the distribution. When the shape factor for the Weibull distribution is 2.5, the Weibull distribution very closely approximates the lognormal distribution. Table V details the shape factor data that results when the DOE ALT data are fit to a Weibull distribution. With the exception of the control, all the data values contain 2.5 within the standard error limits. Thus, the present data agree with the earlier lognormal distribution found by Boddy et al. A likely reason that the control PWBs do not follow a similar distribution as that observed by Boddy et al. is that the control PWBs were heated in the reflow oven for these data, while they were unheated in the Bell Labs research. Activation Energy The majority of previous CAF research was performed on unprocessed PWBs. The results from Bell Labs researchers showed an E a for CAF formation to be between 0.0 ev and 0.2 ev below 60 C and between 1.0 ev and 2.5 ev above 65 C. 3 Calculating the E a for unprocessed PWBs may yield interesting data. However, since all real PWBs must undergo a soldering step, it is more useful to determine the E a for CAF formation that results when various WSFs are used during PWB processing. This is particularly true since it has been shown in previous research 10,29 that the occurrence of CAF can be enhanced by using WSFs that contain certain ingredients such as halides and polyglycols. Table VI details the E a results. The effect of using PG3-4 is virtually indistinguishable from the control and does not appear to be appreciably deleterious to MTTF. The graphical depiction of these normalized MTTF results (Fig. 8) suggests that the addition of bromide in this formulation (PG3-4) actually improves CAF resistance as compared to the nonbromine containing version of the same polyglycol (PG3-2). Rubin et al. 45 similarly observed that bromine additions to no-clean fluxes improved SIR performance. It is unclear why this behavior occurs, though the bromide anion may have some competitive effect with the chlorine reaction kinetics, thereby slowing CAF formation. Also, note that all E a values are less than or equal to 0.5 ev, which is indicative of a relative humidity driven corrosion or degradation mechanism. 46 Also of interest is the very low E a for the PG7-3 PWBs. This low value implies that there is a very low energy barrier to CAF formation with PG7-3 and, hence, a much faster rate of formation. A plausible reason for this dramatic difference from the other WSFs is that PG7-3 contains chlorine, whereas other fluxes do not. Chlorine is a key component of CAF. For the cases of the other WSFs, the chlorine must be scavenged from residues within the PWB. However, when using PG7-3, there will be an abundance of excess chlorine available for ionization and incorporation into the CAF. The excessive CAF formation with PG7-3 may also be due to the interaction between the modified linear aliphatic polyether and the polymeric epoxy. To verify that the iterative values were accurate, substitutions into Eq. 13 were made using the given T, V and L values in coordination with the calculated E a

13 1220 Ready and Turbini and C values (Table VI). The calculated MTTF results have a very good fit (Table VIII) to well within the experimental error of the measured MTTF for each WSF. Data from another set of experiments processed at a different reflow temperature were used to verify that the model was applicable to other data sets. Table IX shows that the calculated MTTF results are also comparable to the experimental error. This indicates that the activation process is consistent even for a slightly lower reflow temperature. It was noted that the calculated E a value for the control PWBs was less than that observed by the Bell Labs researchers. 3 The most likely cause of the difference is that the control PWBs experienced the thermal reflow excursion in this study but did not in the Bell Labs work. Thus, the fiber/epoxy interface was weakened due to the CTE mismatch and thermal excursion. In addition, the PTH-PTH spacing used for Bell Labs research was 75 mils (2 mm), versus 20 mils (0.5 mm) or 30 mils (0.75 mm) here. Drilling PTHs or reflowing the PWBs can weaken or separate the fiber/epoxy interface. If the PTHs are spaced closely together, fiber/epoxy separations at one PTH may intersect the separations from another PTH. For the wider spacing in the Bell Labs work, the separations may not intersect. WSF Formulation Effects Control PWBs Identifying CAF on control PWBs is often difficult due to the very fine filament size (i.e., m in breadth). The SEM cross sections of the CAF taken within 100 m of the PTH barrel (Fig. 9) reveal that the CAF is confined to a very small halo region immediately adjacent to the separated fiber/epoxy interface. The size of the CAF does appear to increase slightly; however, this apparent thickening is actually due to a larger crack opening closer to the anodic PTH. The EDS spectrum presented in Fig. 10 shows numerous peaks for a CAF on a control PWB. One reason for the numerous peaks is the fine filament size. Despite being able to precisely select the spot on the CAF where a spectrum is obtained, as the electron beam decelerates in the analyzed material, it forms a teardrop-shaped interaction volume below the spot selected for analysis where x-rays may be excited and detected. Since the acceleration voltage is fairly large, a relatively large interaction volume is generated. The calcium, aluminum, and silicon peaks result from the e-glass fiber adjacent to the spot on the CAF selected for analysis. The carbon, oxygen, and bromine peaks are due to the epoxy matrix that is also adjacent to the CAF. The gold peaks are due to the conductive coating. The peaks due to the actual CAF are copper and chlorine. The source for the copper is obviously the PTH. Chlorine contamination can come from a variety of sources. However, the most likely source for the chloride is from the epichlorohydrin (C 3 H 5 OCl) used in the epoxy manufacturing process. Generally, ppm of total chloride are present in the finished epoxy. However, the amount of available chloride is typically 100 ppm. 47 Table VIII. Comparison between Measured and Calculated MTTF Meas. Std. Calc. MTTF V L C E a Temp. MTTF Error MTTF Difference Flux (V) (mm) (h) (ev) ( C) (h) (h) (h) (h) Control PG PG PG Table IX. Comparison of Measured and Calculated MTTF for Other Experimental Data ALT Meas. Std. Calc. MTTF Temp. V L C E a MTTF Error MTTF Difference Flux ( C) (V) (mm) (h) (ev) (h) (h) (h) (h) Control PG PG PG

14 The Effect of Flux Chemistry, Applied Voltage, Conductor Spacing, and Temperature on Conductive Anodic Filament Formation 1221 PG3-2 and PG3-4 The chemical composition of CAF that forms with PG3-2 includes copper and chlorine (Fig. 11); the same is true for PG3-4 (Fig. 13). Since PG3-2 does not contain chlorine as an intentional additive, the epichlorohydrin remaining from the epoxy manufacturing process is again a likely source for this anion. Testing of PG3-4 was based on the hypothesis that a brominecontaining WSF would induce bromide-containing CAF as had been observed previously. 10,34 Figure 13 and all other SEM/EDS analyses showed that there was a normal concentration of bromine in the EDS spectra for the PG3-4 CAFs. Since the bromine concentration in PG3-4 (2 wt.%) is lower than that from earlier work 10,34 (approximately 15%), the results suggest that a critical threshold must exist for bromide containing CAF to form. A CAF that bridged the PTHs on a PG3-4 PWB was sectioned at different locations as it approached the anodic PTH (Fig. 19). The SEM micrographs of the two sections (Fig. 20) show that the CAF is larger close to the PTH. The individual e-glass fibers are labeled to aid in visualization. Fig. 19. Optical transmission micrograph showing several CAFs and approximate location of cross section 1 (left of Fig. 20) and cross section 2 (right of Fig. 20). Cross sectioning proceeded from right to left on this PG3-4 PWB. Anode is at left. Since the CAF is growing from the anodic PTH (PTH on left in Fig. 19), the copper ions must travel the length of the filament if it is to be extended toward the cathodic PTH (right PTH in Fig. 19). Migration along the exterior surface of the filament provides the most favorable pathway, since migration along the interior of the already formed CAF would be impeded by the significantly less mobile copper-chloride compound, which is bound within the epoxy structure. Thus, it is reasonable to assume that the filament would be larger near its base than at its tip, since the copper compound must continually advance radially outward if it is to advance toward the cathodic PTH. This type of behavior has also been observed previously. 36 Figure 11 is an SEM micrograph of a CAF that formed on a PWB processed with PG3-2. The separated fiber/epoxy interface is visible, with a crack extending through the epoxy to the upper left. Note that this CAF extends into the epoxy unlike CAF observed on the control PWB. Furthermore, there is a striated appearance of the deposits. The polymeric structure of the epoxy (Fig. 21) has both distinct layers and functional groups where a copper ion can form a complex or chelated structure along the polymeric backbone. Since a high voltage gradient is one of the driving forces for filament growth, the stratified morphology may be due to the alignment of the electric field. PG7-3 Figure 22 shows the locations of the cross sections shown in Fig. 23. It is obvious that the filament is larger closer to the anodic PTH (right of Fig. 23). There is a distinct striated appearance to the CAF. Whereas a stratified appearance is associated with PG3-2 and PG3-4 processed PWBs, the striated (i.e., spotted) appearance for CAF is associated with PG7-3 PWBs (Figs ). PG7-3 contains a modified linear aliphatic polyether with a structure R-(O- CH 2 -CH 2 ) x -R (where R is an alkyl or alkylaryl group, R is a modifying cap, and x is the number of moles of ethylene oxide) and the other fluxes contain a poly(ethylene/propylene) glycol. Zado 25,27 has Fig. 20. CAF cross section 1 (left) approximately 100 m from PTH and cross section 2 (right) adjacent to PTH on PG3-4 PWB. The portion labeled Copper Land is actually the PTH Nailhead on PWB surface. Fig. 21. Epoxy structure. 48

15 1222 Ready and Turbini hours, 50 and this lends credence to the Bell Labs data showing that the electrochemical corrosion step in CAF formation is quite rapid. 4 Equations provide the stoichiometric chemical equations that lead to synthetic atacamite formation. The term R designates a generic cation. 7Cu 7Cu 2 14e (14) 4RCl 4R 4Cl (15) 11H 2 O 10(H) 10(OH) H 2 O (16) Fig. 22. Approximate cross section locations of CAF from a PWB processed with PG7-3. Cross sectioning proceeded from right to left. Fig. 23. SEM image of PWB in Fig. 22 showing cross section 1 (left) located approximately 100 m from the anodic PTH and cross section 2 (right) located adjacent to the PTH. 7Cu 4RCl 11H 2 O 2CuCl 2 5Cu(OH) 2 H 2 O 14e 4R 10(H) (17) In addition, the Pourbaix diagram 51 for the copper-chlorine-water system (Fig. 24) reveals that a compound similar to synthetic atacamite known as paratacamite with the form CuCl 2 3Cu(OH) 2 is stable in the mid to upper left-hand portion of the diagram, a region which corresponds to anodic voltages and acidic conditions. These are the precise conditions required for CAF formation. Equations provide chemical reactions that lead to the formation of the stable product predicted by the Pourbaix diagram. It is obvious that the only differences between synthetic atacamite and the compound in the Pourbaix diagram is the relative stoichiometric proportions of copper, chloride-salt, and water. Furthermore, mineralogists sometimes refer to atacamite as paratacamite or botallacktite, 43,44 so these differences may be a matter of semantics between Professor Pourbaix and mineralogists. shown the effect of polyethylene oxide on epoxy, and this striation behavior is likely attributable to the absorption characteristics of the polyether into the pockets of the epoxy backbone (Fig. 21). The CAFs formed on PWBs processed with PG7-3 also were copper and chlorine containing. However, unlike previous filaments, the chlorine content varied with spatial location within the CAF (Fig. 16). This striation of chloride deposits was only associated with PWBs processed by PG7-3. The additional 2% chlorine in this WSF formulation provides significantly more available anions for inclusion within the CAF than the residual chloride from the epichlorohydrin. The absorption characteristics of the modified linear aliphatic polyether may be instrumental in the formation of these striations. Chemical Nature of Conductive Anodic Filament The qualitative EDS data from the SEM analysis clearly showed that CAF in this study is copper and chlorine containing. The electron diffraction results (Fig. 18 and Table VII) provide evidence that the CAF is synthetic atacamite. Using x-ray powder diffraction, Raffalovich 49 also identified green atacamite as a common corrosion product on PWBs. The formation of atacamite in moist air can be completed in a few Fig. 24. Pourbaix 51 diagram for copper-chlorine-water system at 25 C with 35 ppm Cl. The chemical formulas indicate stable compounds for that particular range of ph and potential.