City University of Hong Kong ... impedance-controlled circuit board, a surface mount low-pass ... interconnects, including solder joints, printed circuit board traces ...
Identification of Interconnect Failure Mechanisms Using RF Impedance Analysis Daeil Kwon2, Michael H. Azarian2, and Michael Pecht1,2 1 Prognostics and Health Management Center City University of Hong Kong 2 Center for Advanced Life Cycle Engineering (CALCE) University of Maryland College Park, MD 20742, USA Phone: +1-301-405-5323, FAX: +1-301-314-9269 Abstract This paper presents RF impedance analysis as a nondestructive indicator of interconnect failure mechanisms. Physical changes on a printed circuit board can be detected using the TDR reflection coefficient as a measure of RF impedance. The test circuit in this study consisted of an impedance-controlled circuit board, a surface mount low-pass filter, and two solder joints providing both mechanical and electrical connections between them. A cyclic mechanical load was applied to generate either solder pad separation failures or cracking of the solder joints. In the case of pad separation, the test results showed that the TDR reflection coefficient gradually decreased as the pad lifted off the circuit board, which was distinguishable from the TDR response due to solder joint cracking. These results are explained using a simple theoretical analysis of the effect of each failure mechanism on impedance. This demonstrates that RF impedance analysis provides for the determination of failure mechanisms as well as early detection of both pad separation and solder joint degradation. The detection sensitivity for solder joint cracking using this technique is also discussed. Introduction In an electronic assembly there are a number of interconnects, including solder joints, printed circuit board traces, component leads, and connectors. During operation, electronics may be exposed to various loading conditions, such as temperature cycling, vibration, and mechanical overstress, that may result in the failure of interconnects or an entire assembly. The consequences of these failures can be disastrous. The failure of everyday electronic devices such as cell phones and laptops would deteriorate the value of those products and the brand equity, which could bring substantial economic loss to the manufacturer. Moreover, for applications that require high reliability, such as aircraft and medical devices, their failure could result in the loss of human life. Therefore, appropriate and accurate reliability assessment of electronics is required not only for economic advantages but also for safety reasons. Traditionally the electronics industry has been using DC resistance to monitor the reliability of electronic assemblies. DC resistance is well-suited for identifying the electrical continuity of an interconnect such as an open or a short circuit, but it is not useful for detecting a partially degraded interconnect or for providing advanced warning of failure. Furthermore, a reliability assessment based on the measurement of DC resistance is not adequate for current and future electronics which operate at frequencies of several
978-1-4244-4489-2/09/$25.00 ©2009 IEEE
hundred MHz or more, and therefore, may be sensitive to early stages of interconnect degradation. As an alternative approach, Kwon et al. [1] introduced RF impedance analysis that provided an early indication of interconnect failures due to the skin effect. They demonstrated the increased sensitivity of RF impedance in detecting interconnect degradation compared to DC resistance. They also showed that a physical crack across a partially degraded solder joint was responsible for the RF impedance increase during fatigue testing by performing failure analysis. This RF impedance analysis uses the time domain reflectometry (TDR) reflection coefficient, which is sensitive to interconnect impedance changes as well as characteristic impedance changes. During their lifetime, electronic products can fail not only because of solder joint cracking but also because of solder pad separation from the circuit board. Since pad separation is associated with a change in the characteristic impedance, monitoring the TDR reflection coefficient can provide an advanced warning of pad separation failures. This paper discusses the ability of RF impedance analysis to predict interconnect failures caused by two different failure mechanisms and to non-destructively identify the failure mechanism by continuous monitoring of the TDR reflection coefficient during fatigue testing. Time domain reflectometry Time domain reflectometry (TDR) measures and localizes impedance discontinuities within a circuit. A TDR measurement is typically reported as a reflection coefficient (ī) and is made by launching an impulse or step into the circuit and observing the reflections caused by impedance mismatches with the characteristic impedance of the circuit [2][3]. The TDR reflection coefficient can be quantified as a ratio of the incident and the reflected voltage and is written as shown in Equation (1):
*
Z L Z0 Z L Z0
(1)
where ZL and Z0 denote the impedance of the device under test (DUT) and the characteristic impedance of the circuit, respectively. As reported at SPI’2008 [1], the TDR reflection coefficient was used to monitor the changes in solder joint impedance during fatigue testing. At the beginning of the test, the TDR reflection coefficient remained almost constant. At the end of the test, however, the TDR reflection coefficient showed a gradual increase followed by a sudden rise, which
SPI 2009
indicated a DC open circuit. In the meantime, a physical crack initiated at the surface of the solder joint and propagated inward, which increased solder joint impedance, ZL, and, therefore, raised the TDR reflection coefficient. This was verified by conducting a failure analysis on the partially degraded sample obtained from a fatigue test halted upon observing a specified increase in the TDR reflection coefficient. The TDR reflection coefficient should also be able to detect physical changes of the circuit associated with the characteristic impedance, Z0. The characteristic impedance is dependent on the design of the circuit board and is normally assumed to be constant. However, pad separation from the circuit board, which is one common failure mechanism in electronics, could increase characteristic impedance and, therefore, decrease the TDR reflection coefficient according to Equation (1). This behavior would be distinguishable from the change in the TDR reflection coefficient during interconnect degradation and can provide an opportunity to identify failure mechanisms in real-time without the need for destructive physical analysis.
coefficient. Thus, TDR reflection coefficient should be sensitive to pad separation as well as solder joint cracking during degradation of a circuit board under stress, and the different trends should allow identification of the relevant failure mechanism. Experimental setup A test circuit was developed to monitor the TDR reflection coefficient compared with the DC resistance during fatigue testing as shown in Figure 2. The test circuit included an impedance-controlled circuit board on which a low pass filter was soldered, two bias-tees for the simultaneous monitoring of the TDR reflection coefficient and DC resistance, a Wheatstone bridge for DC resistance measurement, and a vector network analyzer for TDR reflection coefficient measurement. These components were connected using impedance-controlled RF cables in order to match the characteristic impedance of the test equipment. Vector Network Analyzer (VNA) Bias-Tee RF
Port 1
SMT low pass filter (LPF) on 50 Ohm controlled impedance board
RF+DC
Port 2 DC
LPF
Bias-Tee Wheatstone bridge
Solder joints
DC RF
RF+DC
Figure 2: Schematic of the test circuit Figure 1: Circuit board before soldering The circuit board used in [1] has a coplanar waveguide structure as shown in Figure 1. The characteristic impedance of a coplanar waveguide is a function of the effective dielectric constant (İe), the width of the signal plane (w), and the distance between the signal and the ground plane (s) as shown in Equation (2) [4]:
Z0
30S § 1 § 1 k ' · · ¨ ln ¨ 2 ¸¸ H e ©¨ S ©¨ 1 k ' ¹¸ ¹¸
(2)
where
k'
4 sw 4 w2 s 2w
(3)
According to the equation above, pad separation results in either an increase of the effective distance between the signal and ground plane (s) or a decrease of the effective width of the signal plane (w) projected onto the circuit board, which raises the kƍ value and consequently, the characteristic impedance (Z0) of the circuit board. According to Equation (1), the TDR reflection coefficient decreases as the characteristic impedance increases. From this analysis, pad separation is predicted to lead to a decrease in the TDR reflection
The test circuit board also had a controlled characteristic impedance of 50 Ohms with two SMA connectors on each side. A surface mount technology (SMT) low pass filter was soldered on this circuit board using eutectic SnPb solder paste. The cut-off frequency of the low pass filter was 6.7 GHz, and the monitored frequency span for TDR reflection coefficient was between 500 MHz and 6 GHz. Therefore, the low pass filter acted as a conductor providing two solder joints subjected to fatigue failures. A Wheatstone bridge was incorporated into the test circuit for DC resistance measurement. It not only eliminated environmental noise such as the diurnal changes of ambient temperature but also increased measurement resolution. A Keithley 2010 multimeter was used to monitor the voltage in the Wheatstone bridge circuit, which could be converted into DC resistance. An Agilent E8364A vector network analyzer configured with TDR functionality was used to monitor TDR reflection coefficient during fatigue testing. The DC resistance and the TDR reflection coefficient were simultaneously monitored using the bias-tees that combined and extracted the RF and the DC signals. Test conditions A cyclic mechanical shear force was directly applied to the low pass filter in order to produce interconnect failure either by a separation of the copper pad or by the cracking of the solder joints. The circuit board had two copper pads on which the two end-terminations of an SMT low pass filter were
Results Figure 3 shows the results of a fatigue test in which failure was due to pad separation from the circuit board. The total duration of the test was 407 min. During the test, the TDR reflection coefficient at the solder joint and the DC resistance converted from a node voltage in the Wheatstone bridge were collected every 30 s. At the beginning of the test both measurements remained close to their initial values. The TDR reflection coefficient showed a gradual decrease towards the end of the test, while the DC resistance did not indicate any anomalies. The test was stopped when the contact between the force transducer and the component was lost due to pad separation. In spite of the pad separation, the solder joint still provided a mechanical and electrical connection between the component and the pads. Therefore, in this test neither measurement exhibited a sudden rise at the end of the test. In contrast, during tests which produced solder joint degradation, the TDR reflection coefficient showed a gradual increase towards the end of the test as the solder joint degraded by physical cracking, while the DC resistance remained at its initial value until complete separation of the solder joint. This coincides with the results reported in [1]. After the pad separation test, the circuit board was inspected with a scanning electron microscope (SEM) to locate the physical damage that was responsible for the TDR reflection coefficient changes. As shown in Figure 4, the SEM
revealed that the copper pad was torn off of the circuit board, which was responsible for the decrease of the TDR reflection coefficient at the end of the test. Also, the SEM showed the solder joint on top of the separated pad providing an electrical connection, which indicated that the decrease of the TDR reflection coefficient was caused by the increase of the characteristic impedance. The same behavior between the pad separation and the TDR reflection coefficient was observed over multiple trials. These results indicate that the TDR reflection coefficient provides an early indication of pad separation during fatigue testing, and, therefore, it is capable of predicting and distinguishing interconnect failure by the failure mechanisms of either pad separation or solder joint degradation. 8 TDR reflection coefficient at the solder joint
40
7
30
6
20
5 DC resistance
10
4
0
DC resistance (Ohms)
50 TDR reflection coefficient (mU)
soldered. Preliminary test results showed that the shear strength of the copper pad was about 70 N. In order to induce a wear-out failure through gradual pad separation, a shear force was applied using an MTS Tytron 250 with a constant offset force of 45 N and an oscillatory force of 10 N superimposed on the offset force at a frequency of 0.25 Hz. The offset force maintained the contact between the component and the force transducer throughout the entire fatigue cycle, and the sinusoidally varying oscillatory force produced a cyclic loading condition leading to fatigue failure. Alternatively, when the intention was to produce a gradual cracking of the solder joints as described in [1], the 10 N sinusoidal shear force was superimposed on a smaller offset force than that applied for the pad separation case. A strip of alumina was inserted between the metal tip of the force transducer and the component to avoid electrical connection at the point of contact. During fatigue testing, instrumental control software was used to instruct the multimeter and the vector network analyzer to collect the DC resistance and the TDR reflection coefficient, respectively, every 30 seconds. Each set of TDR measurement data contained a collection of reflection coefficient values over the partial signal path of the circuit board, collected at a particular instant during the fatigue test. In order to monitor changes to the interconnect over time and allow comparison with DC resistance, the TDR reflection coefficients at the location of a specific solder joint were extracted and displayed in a plot as a function of test duration. During each experiment the TDR responses and the DC resistance were monitored until the cyclic stresses resulted in either a separation of the copper pad or the cracking of the solder joints.
3 0
100
200
300
400
Test time (min)
Figure 3: TDR reflection coefficient (in milliunits, mU) and DC resistance during the fatigue test
Figure 4: SEM micrograph of the damaged sample Sensitivity of the technique Since the use of RF impedance analysis to detect interconnect degradation takes advantage of the surface
concentration of high speed signals, the sensitivity of this technique depends on the skin depth, which is related to the material properties of the solder joint and the monitored frequency ranges. Due to the skin effect, more than 99% of the signal at frequencies of several hundred MHz or more is concentrated within five skin depths from the surface. Considering eutectic tin-lead solder and the lowest monitored frequencies of 500 MHz, RF impedance should be sensitive to crack lengths within 50 micrometers. Research is on-going to identify the minimum crack size that can be detected with RF impedance. Stress tests have been halted upon observing changes in the TDR reflection coefficient using the sequential probability ratio test (SPRT). SPRT is a statistical hypothesis test that provides a decision on whether the test data fall into the probability density distribution of the training data that serve as a base line [5][6]. As reported in [7], SPRT has been used to monitor anomalies in the TDR reflection coefficient. In order to identify the minimum detectable crack size, SPRT hypothesis tests are performed continually during a stress test to detect any anomalies present in the data, upon which an alarm is triggered to stop the test. Failure analysis is then conducted to locate cracks within the solder joint and to measure their sizes. Cracks in the range of 40~70 micrometers have consistently been found in these degraded samples, as expected. Conclusions The ability of RF impedance analysis to non-destructively identify interconnect failure mechanisms during cyclic mechanical loading has been demonstrated. Changing the stress level for fatigue testing produced two different failure mechanisms. This allowed comparison of the behavior of the TDR reflection coefficient over time for each failure mechanism. It was shown that the TDR reflection coefficient gradually increased as the solder joint was degraded by physical cracking, whereas it gradually decreased as the solder pad was separated from the circuit board. This behavior was explained on the basis of a theoretical analysis of the effect of changes in characteristic impedance or impedance of the device under test on the TDR reflection coefficient. Meanwhile, the DC resistance, a traditional measure of interconnect reliability, did not provide any indications of degradation throughout the entire test except at the point when solder joint cracking caused complete separation of the solder joint. Therefore, RF impedance analysis provides a means for early detection of interconnect degradation on printed circuit boards as well as non-destructive identification of failure mechanisms. These results imply that RF impedance analysis can provide reliability assessment of electronics exposed to cyclic loading conditions, such as vibration, repeated mechanical shocks or flexure, or even thermal cycling. Not only can reliability monitoring based on RF impedance be used to sense interconnect degradation prior to destruction of the signal path, but it also allows identification of the physical processes by which failures are being induced even while the electronics are still in service. This real-time diagnostic capability can eventually be used to help prevent catastrophic failures. Although significant practical challenges must be
overcome to integrate this method into functional products, no other technique has shown greater sensitivity to early stages of interconnect degradation or the ability to discriminate between alternative failure mechanisms. It was found that solder joint cracks of the order of 50 micrometers in length were consistently detectable when RF impedance analysis was paired with a SPRT, a statistical technique useful for anomaly detection. As clock speeds and communication frequencies rise, it may be expected that the performance of electronics will be adversely affected even by partially degraded interconnects, due to the skin effect. Thus, the ability to detect physical changes such as small cracks or pad separation early in the degradation process should become increasingly valuable. RF impedance analysis offers great promise for use in prognostics of electronic products. Since the changes in RF impedance leading up to failure were gradual in nature, it should be possible to quantify the damage level associated with solder joint cracking or pad separation. An interconnect failure model can then be created by correlating the changes in RF impedance during stress testing with the level of interconnect degradation. This model can be used to predict the failure of interconnects more accurately and to calculate the remaining useful life of products in the field for which RF impedance is being monitored. Eventually, this technique could enable condition-based maintenance, resulting in improved safety and economic advantages not only to manufacturers but also to users. Acknowledgement This work was funded by the members of the CALCE Electronic Products and Systems Consortium at the University of Maryland. References [1] Kwon, D., Azarian, M. H., and Pecht, M., “Effect of Solder Joint Degradation on RF Impedance,” 12th IEEE Workshop on Signal Propagation of Interconnects, Avignon, France, May 2008. [2] “Time Domain Analysis using a Network Analyzer,” Agilent Application Note 1287-12, Agilent Technologies, 2007. [3] Sander, K. F., and Reed, G. A. L., Transmission and Propagation of Electromagnetic Waves, New York: Cambridge University Press, 1978. [4] Ulrich, R. K., and Brown, W. D., Advanced Electronic Packaging, New Jersey: Wiley-Interscience, 2006, pp. 487-536. [5] Humenik, K., and Gross, K., “Sequential Probability Ratio Tests for Reactor Signal Validation and Sensor Surveillance Applications,” Nuclear Science and Engineering, Vol. 105, 1990. [6] Gross, K., and Lu, W., “Early Detection of Signal and Process Anomalies in Enterprise Computing Systems,” IEEE Int’l Conf. on Machine Learning and Applications, Las Vegas, NV, Jun. 2002. [7] Kwon, D., Azarian, M. H., and Pecht, M., “Early Detection of Interconnect Degradation Using RF Impedance and SPRT,” IEEE Int’l Conf. on Prognostics and Health Management, Denver, CO, Oct. 2008.
