J Nondestruct Eval (2015) 34:29 DOI 10.1007/s10921-015-0294-8

Fundamental Study of Microelectronic Chips’ Response Under Laser Excitation and Signal Processing Methods Lei Yang1 · Jie Gong1 · I. Charles Ume1

Received: 11 March 2015 / Accepted: 17 June 2015 / Published online: 19 August 2015 © Springer Science+Business Media New York 2015

Abstract In surface-mount technology, the solder bumps between chips and printed circuit boards are difficult to inspect. The laser ultrasonic-interferometric (LUI) technique proved promising for the non-destructive inspection of solder bump qualities. Previous signal interpretation methods were based on the comparison of signal similarities. A fundamental insight into the chips’ response under the laser excitation was lacking. In this paper, a C-scan procedure was performed to inspect the full surface of chips. This procedure, along with different signal processing techniques such as continuous wavelet transformation, ideal filters and multi-dimensional Fourier transform, established a fundamental understanding of the LUI signals acquired from different chips. Keywords Laser ultrasonic · Multi-dimensional FT · Solder bump

1 Introduction The transition from the traditional through-hole assembly to the surface mount assembly is a significant step in the evolution of electronic packaging. Surface mount devices (SMDs) such as flip chip packages, chip scale packages (CSPs) and ball grid array (BGA) packages are gaining in popularity in the microelectronics industry because they provide high

B

Lei Yang [email protected] Jie Gong [email protected] I. Charles Ume [email protected]

1

Georgia Institute of Technology, Atlanta, GA 30332, USA

density inputs/outputs and better electrical and thermal performance. However, the solder bump interconnections in SMDs are sandwiched between the silicon dies and the substrates or the substrates and the PWBs. The qualities of these solder bumps are difficult to evaluate because they are hidden from view. Solder bumps are one of the most vulnerable parts in electronic products and are subject to various defects introduced during both manufacturing and service. Therefore, inspection of solder bumps has become a crucial process in the electronics manufacturing industry in order to reduce manufacturing costs, improve yield, and ensure product quality and reliability. Currently, electrical testing, X-ray, and acoustic microscopy techniques are the major non-destructive techniques for inspecting solder bump defects. However, each method has its own limitations. A system using the laser ultrasonicinterferometric (LEU) technique has been developed for noncontact, nondestructive quality inspection of solder bumps in microelectronic packaging. This system uses a pulsed Nd:YAG laser to excite ultrasound in the chip packages in the thermo-elastic regime to avoid any damage to the packages. The transient out-of-plane displacement on the package surfaces is measured by a laser interferometer. Solder bump quality is then assessed by analyzing the measured displacement signal, which is expected to change with the presence of solder bump defects. The developed system has been successfully applied to detect solder bump defects, including missing, misaligned, open, and cracked solder bumps in flip chips [1,2], land grid array (LGA) packages, and multilayer ceramic capacitors (MLCCs) [3,4]. Some reliability studies of solder bumps under accelerated thermal cycling have also been reported using this system [5,6]. However, the current signal interpretation methods such as modified correlation coefficient [7] and error ratio [8] are based on the signal similarity, in which the solder bump qual-

123

29 Page 2 of 8

J Nondestruct Eval (2015) 34:29

Fig. 1 Current signal interpretation methods used in LUI solder bump inspection

ity is assessed by comparing the signal with a reference one. Figure 1 shows the basic idea of how these methods work. A fundamental understanding of the received signals is still lacking. It is not known whether the measured out-of-plane displacements are due to the laser excited ultrasound or structural vibration, especially for chips with complex structures. Plus, the fact that the acquired signals vary a lot for different chips requires reference signals for each type of chip to be inspected. Not understanding the chips’ response under the laser excitation makes it difficult to interpret the acquired signals or understand the mechanism how the solder bump defects will affect the acquired signals. This work will perform a C-scan procedure to measure the out-of-plane displacement of the full surface of the chips using the developed LUI inspection system in our lab. From the acquired signals, the transient out-of-plane response of the chips under the laser excitation can be directly visualized. The C-scan procedure, combined with different signal processing methods, will reveal a wealth of information of the chips’ response under the laser excitation and give insight into the acquired signals during the LEU inspection. It is recommended that this procedure be performed on each new chip before the LUI inspection to learn their distinctive behaviors. This prior knowledge can greatly improve the processing and interpretation of the acquired signals during the LEU inspection later.

2 Inspection System, Test Vehicles and Experimental Process The diagram of the LUI system used in this work is shown in Fig. 2. The sample is fixed to the X–Y motion stage by a vacuum system. Nd:YAG laser pulses are delivered through a fiber and a focusing objective to focus onto the chip surface. The transient out-of-plane displacement of the sample surface is measured by a fiber-coupled heterodyne interferometer. A computer vision system captures fiducial marks on the sample substrate to provide precise alignment with the interferometer and excitation laser. The interferometer is enhanced with an autofocus system to ensure optimal signal strength [9]. The acquired signals are digitized and then

123

Fig. 2 Diagram of LUI inspection system

Fig. 3 Test vehicles: a PB18 flip chip, b PB18 flip chip solder bump layout, c FCBGA package, d FCBGA solder bump layout

averaged to improve the signal-to-noise ratio (SNR). The data acquisition, stage positioning, and vision systems are controlled by a computer during the inspection process. This work used two different test vehicles, as shown in Fig. 3. The first one was a 6.35 mm × 6.35 mm PB18 flip chip with 48 solder bumps in a peripheral layout, as shown in Fig. 3a, b. The second was a 29 mm × 29 mm FCBGA package, as shown in Fig. 3c, d. It had 1152 solder bumps in an area array with a 0.8 mm bump-to-bump pitch in both directions. The top of the FCBGA chip was fitted with a heat spreader made of copper with electroless nickel plating. The heat spreader has a rectangular raised portion in the middle. During the inspection procedure, the pulsed laser was repeatedly firing at the chip center while the laser interferometer scanned over the chip surface to measure the out-of-plane displacements, as shown in Fig. 4. Totally 61 × 61 points

J Nondestruct Eval (2015) 34:29

Page 3 of 8 29

Fig. 5 3-D data matrix

Fig. 4 Diagram of the experimental process: a PB18 inspection pattern, b FCBGA inspection pattern

with a 0.1 mm pitch on the FB18 chip were measured. The rounded edge of the raised portion of the FCBGA chip made it difficult for the laser interferometer to make measurements. Therefore the FCBGA chip was scanned excluding the borders of the raised region, which is a rectangular loop with a width of 2 mm, as shown in Fig. 4b. Over 10,000 points with a 0.25 mm pitch were measured on the FCBGA chip. Fig. 6 Snapshots of FB18’s response after zero padding

3 Chip Response Under Laser Excitation Assuming that the laser source is time consistent, the acquired signals can be treated as if they were measured simultaneously during a single laser pulse. Compilation of the signals based on their acquisition locations forms a 3-D data matrix, as shown in Fig. 5. The first two dimensions correspond to the locations on the chip surface, and the third dimension corresponds to the time. The data on a “slice” perpendicular to the time dimension, as shown in Fig. 5, represent the out-of-plane displacement of the chip surface at a specific moment. Plotting the “slices” continuously in time order helps visualize the transient out-of-plane response under the laser excitation. For the FCBGA chip, locations where no signals were acquired were padded with zeros. For both test vehicles, the amplitudes of signals at locations close to the laser source were found to be abnormally high, which obscured the response at the other locations. Therefore, signals within twice the size of the laser source radius from the chip center were replaced with zeros for both test vehicles.

Fig. 7 Frequency spectrum of a signal acquired from FB18

The snapshots of the FB18 chip’s response are shown in Fig. 6, which shows a vibration with more than one mode. Figure 7 shows the frequency spectrum of one of the temporal signals. Two peak frequencies are 105 and 225 kHz, which correspond to the two different vibration modes. These two modes were isolated by applying an ideal low-pass filter with cutoff frequency 200 kHz and an ideal band-pass filter with bandwidth [200, 250 kHz], as shown in Fig. 8. The laser source primarily excited structural vibrations in the FB18 chip.

123

29 Page 4 of 8

J Nondestruct Eval (2015) 34:29

Fig. 8 Snapshots of 105 and 225 kHz mode vibration

Compared with the FB18 chip, the FCBGA’s response is much more complicated. Figure 9 shows that two waves of different speeds originated from the laser source. The wave fronts of the faster one quickly reached the chip boundaries, and their reflections constructed an interference pattern due to the structural symmetry. The wave fronts of the slower one, however, didn’t cause any reflections when hitting the boundaries, as shown in Fig. 9e–h. The structural vibration is not obvious to observe from Fig. 9. The continuous wavelet transformation (CWT) was applied to one of the temporal signals, and the resultant time-frequency domain representation is shown in Fig. 10 [10], which shows that high-frequency signal components are present for a very short time and that it takes some time before they appear. Oppositely, signal components below 0.1 MHz exist from the very beginning and last for a longer time, whose amplitudes attenuate with time. The behavior of the former matches well with the wave propagation and the latter with the damped structural vibration. The structural vibration was not visualized previously probably because it was obscured by the strong wave propagation. The presence of the structural vibration is confirmed by applying two ideal filters with bandwidth [0, 50 kHz] and [50, 65 kHz] to isolate the presumed vibration in the low frequencies. Figures 11 and 12 show their snapshots, which plot only the signals acquired within the raised portion to exclude the discontinuity caused by the padded zeros. Figure 11 shows a simple vibration mode, and Fig. 12 shows a mixture of vibration and wave propagation. It can be concluded that both the structural vibration and the wave propagation were excited in the FCBGA chip. The structural vibration fell in the low frequency range with a much lower amplitude than the wave propagation.

123

Fig. 9 Snapshots of FCBGA’s response after zero padding

Fig. 10 Time-frequency domain representation of a temporal signal

The types of the excited waves in the FCBGA chip were identified by performing a B-scan procedure along a path, as shown in Fig. 13a, with a finer pitch of 0.05 mm to achieve

J Nondestruct Eval (2015) 34:29

Page 5 of 8 29

Fig. 11 Snapshots corresponding to frequency below 50 kHz

Fig. 13 a B-scan path, b Frequency-wavenumber representation of B-scan signals

Fig. 12 Snapshots corresponding to frequency range [50, 65 kHz]

a higher spatial resolution. The obtained signals in the timespace domain were converted to the frequency-wavenumber domain using the 2-D FT [11,12], as shown in Fig. 13b. The two straight lines in the frequency-wavenumber domain indicate that the excited waves have constant speeds, which eliminates the possibility of dispersive Lamb waves [13]. The two speeds which were calculated using the reciprocal of the slopes are 342 and 1734 m/s. The slower one matches the speed of sound, which explains why the wave fronts of the slower wave didn’t cause any boundary reflections in Fig. 9. The sound was generated when the laser irradiated onto the chip surface and caused a very thin layer of the top material to vaporize, which could be heard during the experiment. Interestingly, the airborne sound was picked up by the laser interferometer as well. This test demonstrates that the laser source can excite the structural vibration, the wave propagation or even both

depending on the chip’s structure. For simple structures like the FB18, the laser excites only the structural vibration. For complicated structures like the FCBGA, it excites both the structural vibration and the wave propagation. The structural vibration is dominant at low frequencies. In the middle frequency range, the vibration and the wave propagation may coexist. At high frequencies, the wave propagation is dominant.

4 Remove Unwanted Signal Components The ultimate goal is to investigate how solder bump defects will affect the chip response under the laser excitation. However, the airborne sound and the boundary reflections, which are not informative, are very strong in the signals acquired from the FCBGA chip. Their presence makes the investigation difficult. The multi-dimensional Fourier Transform has proven effective to separate waves of different directions [11,12,14]. The previous applications dealt with situations where the wave source is beyond the inspected wave field. In this work, the wave source is at the center of the inspected wave field. What makes this work more challenging is that the airborne sound has the same propagation direction as

123

29 Page 6 of 8

J Nondestruct Eval (2015) 34:29 Table 2 Wave directions in four quadrants Quadrant

Incident waves (X direction, Y direction)

Reflected waves (X direction, Y direction)

(1)

(+, +)

(−, +), (+, −), (−, −)

(2)

(−, +)

(+,+), (+, −), (−, −)

(3)

(−, −)

(+,+), (+, −), (−, +)

(4)

(+, −)

(+,+), (−, +), (−, −)

Fig. 14 3-D FFT coefficients in kx − k y − ω domain Table 1 Mapping between wave directions and 3-D FFT coefficients Directions in x–y plane

3-D FFT coefficients

Positive x, positive y

Octants III and V

Positive x, negative y

Octants II and VIII

Negative x, positive y

Octants IV and VI

Negative x, negative y

Octants I and VII

Fig. 16 kx − k y − ω domain filtering for signals in four quadrants of chip surface

Fig. 15 Chip surface divided into four quadrants

the excited ultrasonic wave. A 3-D FT-based algorithm will be implemented to remove both the airborne sound and the boundary reflections. The 3-D FT converts the C-scan signals from the space and time domain to the kx −k y − ω domain, as shown in Fig. 14. The 3-D FT coefficients in the kx −k y − ω domain are conjugate symmetric about the origin. Different octants correspond to waves of different directions. The mapping relationships are showed in Table 1. In Fig. 15, the chip surface is divided into four quadrants. The directions of the incident and the reflected waves in each quadrant are listed in Table 2. Only the first-time reflections from the local boundaries are considered here. The multiple-time reflections and the reflections from boundaries in other quadrants are ignored because it is assumed that they become very weak due to damping. Because the

123

directions of the desired and unwanted waves are different in each quadrant, it is necessary to process the four quadrants separately. The incident wave in quadrant (1) has a positive x and positive y direction, which corresponds to the coefficients in octants III and V after the 3-D FT based on Table 1. The reflections in quadrant (1) were removed by replacing the 3-D FT coefficients in all other octants except III and V with zeros, as shown in Fig. 16a, and converting back using the inverse 3-D FT. The same procedure was applied to signals in the other quadrants. Figure 16 summarizes the kx −k y −ω domain coefficient selection for all the four quadrants. Figure 17 shows the comparison before and after the filtering in quadrant (1), which shows that the boundary reflections were successfully removed. The airborne sound is still present because it has the same propagation direction as the incident ultrasonic wave. It will be removed from the kx − k y − ω domain as well based on its different speed from the ultrasonic wave. Figure 18a shows the remaining 3-D FT coefficients of quadrant (1) after removal of the reflections. Two planes of coefficients in Fig. 18a are plotted in Fig. 18b, c. The sound and the incident

J Nondestruct Eval (2015) 34:29

Page 7 of 8 29

Fig. 18 Remove airborne sound in kx − k y − ω domain

Fig. 17 Comparison of signals in quadrant (1) before and after removing boundary reflections

ultrasonic wave have different slopes in Fig. 18b and different radii in Fig. 18c. In fact, they correspond to two conical surfaces of different angles in the kx − k y − ω domain, as shown in Fig. 18d. The airborne sound can be removed by simply replacing the coefficients on its own conical surface with zeros. The modified coefficients are then converted back using the inverse 3-D FT to reconstruct the signals in the time and space domain. Figure 19 shows the signals in quadrant (1) after the airborne sound was removed. Compared with Fig. 17, the airborne sound is successfully removed.

5 Discussions The laser source created a local thermo-elastic strain zone at the center of both test vehicles, and the signals acquired there

Fig. 19 Signals in quadrant (1) after removing sound

had abnormally high and random amplitudes. The radius of the strain zone was about twice of the laser incident point’s radius, which puts a limit to the smallest chip that the LUI technique can inspect. The airborne sound was not observed in the signals acquired from the FB18 chip. This is because the FB18 chip exposes its silicon die, which has a very high reflection rate and hardness. The laser source couldn’t vaporize the surface material. In fact, no sound was heard during the inspection of the FB18 chip.

123

29 Page 8 of 8

Different chips behave differently under the laser excitation. The affecting factors include dimension, structure, material, and solder bump layout, etc. Therefore, the LUI signals are greatly chip-dependent. Conducting the C-scan inspection procedure establishes a good understanding of the responses of the test vehicles under the laser excitation, which will greatly benefit the signal interpretation later. In the kx − k y − ω domain, waves of different directions and speeds are automatically separated. Therefore, the 3-D FT provides an effective method for removing unwanted components in the C-scan signals.

6 Conclusion This paper experimentally investigated the response of microelectronic chips under laser excitation. A C-scan inspection procedure was performed using the developed laser ultrasonic-interferometric system. Test vehicles presented fundamentally different behaviors due to their different structures. The C-scan procedure combined with different signal processing methods revealed a wealth of valuable information about the chip response under the laser excitation. This procedure, which is recommended for all unknown chips, will greatly benefit the future interpretation of inspection signals.

References 1. Yang, J., Ume, I.: Laser ultrasonic technique for evaluating solder bump defects in flip chip packages using modal and signal analysis methods. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 57(4), 920–932 (2010) 2. Yang, J., Ume, I.C.: Quality inspection of flip chip solder bumps using integrated analytical, numerical, and experimental modal analyses. J. Electron. Packag. 130(3), 031009–031010 (2008)

123

J Nondestruct Eval (2015) 34:29 3. Erdahl, D.S., Ume, I.C.: Online-offline laser ultrasonic quality inspection tool for multilayer ceramic capacitors—Part I. IEEE Trans. Adv. Packag. 27(4), 647–653 (2004) 4. Erdahl, D.S., Ume, I.C.: Online–offline laser ultrasonic quality inspection tool for multilayer ceramic capacitors—part II. IEEE Trans. Adv. Packag. 28(2), 264–272 (2005) 5. Yang, J., et al.: Board-level solder joint reliability study of land grid array packages for RF application using a laser ultrasound inspection system. J. Electron. Packag. 132(2), 021006 (2010) 6. Lizheng, Z., et al.: Study of flip chip solder joint cracks under temperature cycling using a laser ultrasound inspection system. IEEE Trans. Compon. Packag. Technol. 32(1), 120–126 (2009) 7. Zhang, L., et al.: Detection of flip chip solder joint cracks using correlation coefficient and auto-comparison analyses of laser ultrasound signals. IEEE Trans. Compon. Packag. Technol. 29(1), 13–19 (2006) 8. Liu, S., Ume, I.C.: Digital signal processing in a novel flip chip solder joint defects inspection system. J. Electron. Packag. 125(1), 39–43 (2003) 9. Randolph, T.W.: Development of Automated Method of Optimizing Strength of Signal Received by Laser Interferometer, in Mechanical Engineering 2009, Georgia Institute of Technology 10. Rogge, M.D.: In-process Sensing of Weld Penetration Depth Using Non-contact Laser Ultrasound System. PhD Thesis, Georgia Institute of Technology.: ProQuest Information and Learning Company (2009) 11. Michaels, T.E., Michaels, J.E., Ruzzene, M.: Frequencywavenumber domain analysis of guided wavefields. Ultrasonics 51(4), 452–66 (2011) 12. Wu, T.-Y., Ume, I.C.: Fundamental study of laser generation of narrowband Lamb waves using superimposed line sources technique. NDT & E Int. 44(3), 315–323 (2011) 13. Achenbach, J.D.: Wave Propagation in Elastic Solids. NorthHolland series in applied mathematics and mechanics, vol. 16. North-Holland Pub. Co, American Elsevier Pub, Co, Amsterdam (1973) 14. Wu, T.-Y., Ume, I.C.: Measurement of Weld Dimensions of Butt Joint Welding in Thin Plates Using Superimposed Laser Sources. in ASME 2010 International Mechanical Engineering Congress and Exposition. ASME (2010)