Japanese Journal of Applied Physics 48 (2009) 07GM19
REGULAR PAPER
Room-Temperature Ultrasonic Bonding of Semiconductor Thin-Dies with Die Attach Films on Glass Substrates Sui Yin Wong, Siu Wing Or, Ho Chi Wong, Yiu Ming Cheung, and Ping Kong Choy Department of Electrical Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Received November 20, 2008; revised January 29, 2009; accepted February 25, 2009; published online July 21, 2009 A novel ultrasonic bonding technique for semiconductor thin-dies laminated with die attach films (DAFs) is proposed to improve the process window and to increase the throughput limited intrinsically in the state-of-the-art thermocompression bonding technique. The proposed technique involves the introduction of ultrasonic vibration energy generated from an ultrasonic transducer to the DAFs underneath the thindies so as to adhere the thin-die-DAF laminates onto the substrates. In this paper, a 40 kHz piezoceramic ultrasonic transducer is developed and integrated with a mechatronic test bed to form an automated equipment model for the ultrasonic thin-die bonding. Process studies are conducted to bond 50 mm thick thin-dies with 10 mm thick DAFs on glass substrates using the ultrasonic and thermocompression techniques. The results show that the ultrasonic technique can effectively reduce the process temperature and time as required by the thermocompression technique. Ultrasonic bonding at room temperature (25 C) is achieved with a process time of 2 s and an ultrasonic power of 150 W. Comparable bondability can only be obtained using thermocompression bonding at temperatures in excess of 120 C. # 2009 The Japan Society of Applied Physics DOI: 10.1143/JJAP.48.07GM19
1.
Introduction
The continual evolution of electronic products towards multi-functionality and miniaturization has pressed the manufacturing journey of the electronics industry in the direction of higher density, higher performance, smaller size, and lower cost. The significant reduction in thickness of semiconductor dies from 500 to 25 mm in the recent years not only represents a revolutionary breakthrough in semiconductor technology, but also creates a great tendency towards the major formats of semiconductor packages and their resulting electronic products.1) The impact is indispensable since these semiconductor thin-dies can make packages and products to be even thinner (e.g., smart cards, smart/biological identity cards, biological passports, etc.) on the one hand, and can empower high-density stacking within a limited space or area on the other hand. From the viewpoint of semiconductor packaging technique and process, the conventional die bonding based on liquid adhesives is unable to accommodate semiconductor dies of such thinness due to unavoidable failures caused by adhesive overflow, insufficient coverage, etc.2) By contrast, thermocompression bonding of semiconductor thin-dies that are pre-laminated with die attach film (DAF) on their back side after the wafer thinning and stress relief processes but before the wafer dicing process (Fig. 1) is not subject to the liquid adhesive-induced problems.3,4) Today, semiconductor dies of thicknesses less than 75 mm are mostly produced as thin-die-DAF laminates, and thermocompression bonding is regarded as the state-of-the-art technique to bond these thindie-DAF laminates onto the substrates. Unfortunately, high process temperatures of 100 –160 C and long process times of several seconds have imposed a great challenge to the thermocompression bonding by constraining the bonding process window and throughput.5) In this paper, a novel ultrasonic bonding technique is proposed to improve the process window and to increase the throughput of bonding thin-die-DAF laminates compared to the state-of-the-art thermocompression bonding technique. This new technique is essentially based on the proper transmittal of ultrasonic vibration energy generated from an
E-mail address:
[email protected]
Fig. 1. (Color online) Photograph of semiconductor thin-dies laminated with DAFs in form of a wafer.
ultrasonic transducer through the thin-dies to the underneath DAFs, thereby adhering the thin-die-DAF laminates onto the substrates. Similar ideas have been employed in wire bonding,6) flip-chip bonding,7) and anisotropic conductive film (ACF) bonding.8) The design and evaluation of a 40 kHz piezoceramic ultrasonic thin-die bonding transducer are described, together with its integrated automated equipment model. The effect of introducing ultrasonic energy on process parameters (e.g., process temperature and process time) and bondability (i.e., percentage of voids formed) is reported for both the proposed ultrasonic and state-of-the-art thermocompression bonding techniques. 2.
Ultrasonic Thin-Die Bonding Transducer
2.1 Structure and operating principle Figure 2(a) illustrates the solid model of the 40 kHz piezoceramic ultrasonic thin-die bonding transducer, while Fig. 2(b) shows the photograph of the fabricated transducer. The transducer is designed based on the bolt-clamped Langevin-type configuration for converting electrical driving signals supplied by an ultrasonic signal generator (not shown) into ultrasonic axial vibrations.9) It basically consists of four pieces of lead zirconate titanate (PZT) piezoceramic rings (CeramTec), each with an outer diameter of 25 mm, an inner diameter of 10 mm and a thickness of 3 mm, connected electrically in parallel (via four pieces of ringshaped copper alloy electrodes of thickness 0.2 mm each)
07GM19-1
# 2009 The Japan Society of Applied Physics
Jpn. J. Appl. Phys. 48 (2009) 07GM19
S. Y. Wong et al.
PZT rings (3 mm thickness, 25 mm diameter)
Back Slab (Titanium Alloy) 8.2 mm
Electrodes
56.2 mm (0.5 Wavelength) Flange
Bonding Horn (Aluminum Alloy) 35.2 mm
Bonding tip (18.8 mm diameter)
(a)
Table I. Material parameters used in the finite element harmonic analysis of the ultrasonic thin-die bonding transducer shown in Fig. 3. Material parameter
2.2 Electrical resonance characteristics The electrical resonance characteristics of the fabricated transducer in Fig. 2(b) were studied by (1) measuring its electrical impedance spectrum using an impedance analyzer (Agilent 4294A) and (2) computing the corresponding spectrum using a finite element analysis software (ANSYS Multiphysics 11.0),11) both in the vicinity of the designated working frequency of 40 kHz and under mechanically free
Steel —
800
—
—
523
—
—
—
tan e (%)
0.2
—
—
—
d33 (pm/V)
238
—
—
—
d31 (pm/V)
100
—
—
—
d15 (pm/V)
380
—
—
—
sE11 (pm2 /N) sE33 (pm2 /N)
9.90 14.01
— —
— —
— —
sE44 (pm2 /N)
36.28
—
—
—
sE66 (pm2 /N)
31.94
—
—
— —
sE12 (pm2 /N)
3:14
—
—
sE13 (pm2 /N)
4:47
—
—
—
2823
4400
7868
77
115
207
0.33 —
0.32 —
0.292 —
tan m (%)
and mechanically in series,10) and clamped between an aluminum alloy bonding horn of 35.2 mm long and a titanium alloy back slab of 8.2 mm thick under a hightension prestress bolt. The transducer has a length of 56.2 mm, corresponding to half a longitudinal wavelength at the designated resonance frequency of 40 kHz. The halfwavelength design makes the transducer to be short enough for minimizing the undesirable tilting effect during the manipulation of the transducer to make contact with the thindie surface. A mounting flange is affixed to the vibration node of the transducer, locating at 32.2 mm from the tip of the bonding horn. The diameter of the tip of the bonding horn is set to 18.8 mm to match with the lateral dimensions of the thin-dies and to acquire sufficient amplification of the axial vibrations.
Titanium alloy
S11
Y (GPa)
Fig. 2. (Color online) (a) Solid model of the 40 kHz piezoceramic ultrasonic thin-die bonding transducer and (b) photograph of the fabricated transducer.
Aluminum alloy
S33
(kg/m2 )
(b)
PZT piezoceramic
7529 — — 0.1
condition. Table I summarizes the material parameters used in the finite element harmonic analysis. In order to take into account the dissipation of vibration energy in the fabricated transducer, an effective damping ratio of 0.012 and an effective friction coefficient of 0.1 were introduced into the analysis to describe the material losses of the transducer components and the contact losses of the high-tension prestress bolt and its associated threaded hole situated in the aluminum alloy bonding horn (Fig. 2), respectively.11) The effective electromechanical coupling coefficient (keff ) of the transducer was determined using12) sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 fr ; ð1Þ keff ¼ 1 fa where fr and fa are the resonance and anti-resonance frequencies as observed from the electrical obtained from12) Qm ¼
fr ; f
ð2Þ
where f is the 3 dB bandwidth, also observed from the electrical impedance spectrum. Figure 3(a) shows the measured and computed electrical impedance spectra of the transducer in the frequency range of 34 – 48 kHz and under mechanically free condition. It is clear that the measured resonance and anti-resonance frequencies of the transducer are 39.6 and 41.8 kHz, which are slightly lower than the computed values of 40 and 42.1 kHz using finite element analysis. The effective electromechanical coupling coefficient and mechanical quality factor determined from measurement are 0.32 and 461, respectively. These values are comparable to those obtained from finite element analysis of 0.31 and 487, respectively. Figure 3(b) displays the mode shape associated with the computed working (resonance) frequency of the transducer at 40 kHz. The fundamental (symmetric) axial mode with a 0.5 longitudinal wavelength is evident. From a physical standpoint, this axial mode produces a piston-like motion of the transducer in the axial direction that, in turn,
07GM19-2
# 2009 The Japan Society of Applied Physics
Jpn. J. Appl. Phys. 48 (2009) 07GM19
S. Y. Wong et al.
1M
1.4 1.2
Normalized Axial Displacement
Electrical Impedance (Ω)
100k
10k
1k
100
10
40 kHz
39.6 kHz
1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 -1.2 -5
1 34
35
36
37
38
39
40
41
42
43
44
45
46
47
0
5
10
15
20
25
30
35
40
45
50
55
60
Length of Transducer (mm)
48
Frequency (kHz)
Fig. 4. (Color online) Measured (solid symbol) and computed (open symbol) axial displacement distributions along the length of the ultrasonic thin-die bonding transducer under mechanically free condition.
(a)
2.0 1.8
Vibration Amplitude (μm)
1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2
(b)
0.0 37
Fig. 3. (Color online) Measured (solid symbol) and computed (open symbol) electrical resonance characteristics of the ultrasonic thin-die bonding transducer under mechanically free condition.
causes the tip of the bonding horn to move back and forth axially for facilitating the largest axial displacement. As this piston-like motion is essentially normal to the surface of the thin-die-DAF laminates to be bonded, axial mode is the most desirable resonance mode in ultrasonic thin-die bonding.9) 2.3 Vibrational characteristics The vibrational characteristics of the ultrasonic thin-die bonding transducer were evaluated under mechanically free condition by a laser Doppler vibrometer (Graphtec AT3700/ AT0042). An ultrasonic generator (Ultra Sonic Seal HVS540) operating in a constant voltage, phase-locked-loop mode was used to drive the transducer. Figure 4 plots the axial displacement amplitudes measured at the tip of the bonding horn and at the tip of the back slab when the transducer was driven at the resonance frequency of 39.6 kHz. The axial displacement distribution along the transducer computed using finite element harmonic analysis is also enclosed in the same figure for comparison. It is confirmed that the transducer vibrates in the fundamental axial mode with the vibration anti-nodes located at the two ends and the vibration node positioned at the mounting flange. The axial displacement amplitudes measured at the two ends of the transducer agree well with those of the computed values.
38
39
40
41
42
43
44
Frequency (kHz)
Fig. 5. (Color online) Free vibrational amplitude spectrum of the ultrasonic thin-die bonding transducer measured at the tip of the bonding horn using a driving voltage of 10 V and under mechanically free condition.
Figure 5 shows the vibrational amplitude spectrum of the transducer measured at the tip of the bonding horn under a driving voltage of 10 V. The transducer gives an enhanced vibration amplitude of 1.56 mm at the resonance frequency of 39.6 kHz. This value is about 20 times larger than the nonresonance values of about 0.08 mm. Figure 6(a) provides the relationship between the (free) vibration amplitude and the driving voltage at the transducer’s resonance of 39.6 kHz, while Fig. 6(b) presents the relationship between the driving power and the driving voltage. From Fig. 6(a), the transducer has an excellent linearity between its vibration amplitude output and driving voltage input over a broad range of driving voltage in excess of 80 V. In other words, by driving the transducer at 80 V, it can produce a significantly large vibration amplitude of 12.5 mm, corresponding to a driving power of about 420 W [Fig. 6(b)]. 3.
Automated Equipment Model
Figure 7(a) shows the in-house automated equipment model for the ultrasonic thin-die bonding. The equipment model is basically an integration of the 40 kHz ultrasonic thin-die bonding transducer [Fig. 7(b)] with a mechatronic test bed formed by an ultrasonic signal generation system, a thermal
07GM19-3
# 2009 The Japan Society of Applied Physics
Jpn. J. Appl. Phys. 48 (2009) 07GM19
S. Y. Wong et al.
16
Vibration Amplitude (μm)
14 12 10 8 6 4 2 0 0
10
20
30
40
50
60
70
80
90
100
Driving Voltage (V)
(a) 600
(a)
Driving Power (W)
500
400
300
200
100
0 0
10
20
30
40
50
60
70
80
90
100
Driving Voltage (V)
(b) Fig. 6. (a) Relationship between the (free) vibration amplitude and the driving voltage at the transducer’s resonance of 39.6 kHz and (b) relationship between the driving power and the driving voltage.
management system, a pressure control system, a threeaxis linear motion system, a device mounting system, and a vision system. The ultrasonic signal generation system has an automatic frequency tracking function which ensures the transducer is working at the resonance state. Besides, it can automatically detect the change in mechanical load experienced by the transducer and keep the vibration amplitude to be constant during bonding. The thermal management system provides a wide range of process temperatures from 25 to 250 C. The pressure control system has the operating force range of 4 – 400 N. The three-axis linear motion system offers X-, Y-, and Z-motions, each with the maximum travel distance of 110 mm and an accuracy of 5 mm. The device mounting system provides a mounting bracket for the ultrasonic thin-die bonding transducer by mounting it at the predetermined nodal location via the mounting flange [Fig. 7(b)]. The vision system assures the placement accuracy of the thin-dieDAF laminates on the substrates and the instant inspection of the bonded specimens. Table II lists some important specifications of the in-house automated equipment model shown in Fig. 7.
(b) Fig. 7. (Color online) The in-house automated equipment for the ultrasonic thin-die bonding. The zoom-in view of the 40 kHz thin-die bonding transducer. Table II. Some important specifications of the in-house automated equipment model shown in Fig. 7. Process temperature ( C)
25 – 250
Process force (N)
4 – 400
Process time (s)
0.3 –10
Maximum travel along the X-, Y-, and Z-axes (mm)
110
Ultrasonic frequency (kHz)
40
Ultrasonic power (W)
20 – 500
4.
Ultrasonic Thin-Die Bonding Studies
4.1 Specimens Figure 8 illustrates the photograph and schematic diagram of a standard test specimen used in the present study. The test specimen has a thin-die of 8 mm long, 6 mm wide, and 50 mm thick laminated with a 10 mm thick DAF and the thindie-DAF laminate is to be bonded on an 1 mm thick glass substrate. The glass substrate is a glass slide of dimensions 76 mm in length, 25 mm in width, and 1 mm in thickness.
07GM19-4
# 2009 The Japan Society of Applied Physics
Jpn. J. Appl. Phys. 48 (2009) 07GM19
S. Y. Wong et al. Static compressive force
Thin-die (50 μm) Glass substrate
Die Attach Film
DAF (10 μm)
Vacuum-based, temperature-controlled worktable
50 μm 10 μm
Thin-die
Bonding tip
1 mm
Glass Substrate
Fig. 8. (Color online) Photograph (top) and schematic diagram (bottom) of a standard test specimen used in the present study. Table III. Characteristics of DAF for thermocompression bonding. Thickness (mm)
10
Process temperature ( C)
100 –160
Process time (s)
3
Glass transition temperature ( C)
180
The reasons for selecting glass as the substrate are: (1) It is transparent and the appearance of bonds can be inspected visually immediate after each bonding; (2) the surface of the glass substrate is so flat that it can eliminate undesirable void formation due to substrate unevenness; and (3) this chip-on-glass (COG) configuration represents one of the most important configurations in modern semiconductor packages.13) Table III summarizes the characteristics of DAF for thermocompression bonding. It is seen that the process window for the thermocompression bonding of the DAF is 100 –160 C process temperature and 3 s process time, while the glass transition temperature of DAF is 180 C. 4.2 Bonding process Figure 9 shows the operation of an ultrasonic thin-die bonding process. The glass substrate is first held by a vacuum-based, temperature-controlled worktable, and a thin-die-DAF laminate is picked and placed onto the glass substrate. Second, the specimen is pressed by the transducer with a pressure, and the ultrasonic energy is applied by the transducer through the thin-die to its underneath DAF in order to adhere the thin-die-DAF and the glass substrate within a process time. Comparing to the state-of-the art thermocompression bonding, this ultrasonic bonding technique essentially requires much lower process temperature and shorter process time to be described in the next section. 4.3 Results and discussion Based on the equipment model and the test specimens, a series of experiments were performed to investigate the effect of introducing ultrasonic vibration energy on the reduction of process temperature, process time, and void formation in the proposed ultrasonic thin-die bonding. The
Fig. 9. (Color online) Schematic diagram showing the operation of an ultrasonic thin-die bonding process.
experiments were repeated without the addition of ultrasonic energy and these corresponded to the state-of-the-art thermocompression bonding. For each bonded specimen, the percentage of voids formed and its associated morphology were evaluated using a digital optical microscope (Leica DM4000M) equipped with an image analysis software (Leica Q550MW). The ultrasonically bonded test specimens were compared with the thermocompressively bonded test specimens in terms of ultrasonic power, process temperature, process time, and percentage of voids formed. Table IV shows some typical results of the proposed ultrasonic thin-die bonding and the state-of-the-art thermocompression thin-die bonding. The white areas shown in the optical micrographs indicate the formed voids. The ultrasonic power of the transducer was fixed at 150 W throughout the whole process studies. Based on the lowest process temperature of 100 C characterized by DAF for thermocompression bonding in Table III, a lower temperature of 80 C was selected as the process temperature in the first stage of process studies. The result clearly shows that the use of ultrasonic power of 150 W is capable of establishing the quantitatively good bondability with no more than 5% voids at a lower process temperature of 80 C and a shorter process time of 1.5 s (Specimen A) as compared with the thermocompression bonding of 160 C process temperature and 2 s process time (Specimen B). Thus, the introduced ultrasonic vibration energy in the ultrasonic bonding process can effectively reduce the process temperature and time as required by the thermocompression process. As process temperature is regarded as the most influential process parameter in the thin-die bonding process, roomtemperature bonding was attempted in the second stage of the process studies.14) The result of this experiment can realize the technologically important and physically interesting room-temperature thin-die bonding (Specimen C) with void formation less than 10%. Interestingly, the voids associated with this room-temperature ultrasonic bonding are more than 50% reduced from the thermocompression bonding at 120 C with 2 s (Specimen D). 5.
Conclusions
A novel ultrasonic thin-die bonding technique has been successfully developed to improve the process window and to increase the throughput limited intrinsically in the stateof-the-art thermocompression thin-die bonding technique.
07GM19-5
# 2009 The Japan Society of Applied Physics
Jpn. J. Appl. Phys. 48 (2009) 07GM19
S. Y. Wong et al.
Table IV. Comparison of ultrasonic and thermocompression process study results for the test specimens.
Specimen
Bonding technique
Ultrasonic power (W)
Process temperature ( C)
Process time (s)
Percentage of voids formed (%)
A
Ultrasonic bonding
150
80
1.5
5
B
Thermocompression bonding
0
160
2
5
C
Ultrasonic bonding
150
25a)
2
9
D
Thermocompression bonding
0
120
2
20
Optical micrograph
a) Room temperature
An ultrasonic transducer has been specifically designed and evaluated for the ultrasonic thin-die bonding process. The developed transducer has been integrated with a mechatronic test bed to form an automated equipment model. A series of process studies has been performed to investigate the effect of introducing ultrasonic vibration energy on the reduction of process temperature, process time and void formation in the ultrasonic thin-die bonding. Room-temperature ultrasonic bonding of thin-die-DAF laminates on glass substrates has been successfully demonstrated using 2 s process time with less than 10% voids. To obtain comparable results, it has been found that process temperatures in excess of 120 C are generally required for the thermocompression bonding. Acknowledgment This work was supported by the Innovation and Technology Fund of the Hong Kong Special Administrative Region (HKSAR) Government under Grant No. GHP/003/06. Support from ASM Assembly Automation Ltd. is also acknowledged.
1) International Technology Roadmap for Semiconductors 2008 Update. 2) A. C. M. Chong and Y. M. Cheung: Proc. 5th Int. Conf. Electronic Packaging Technology, 2003, p. 44. 3) Y. M. Cheung and A. C. M. Chong: Proc. 5th Int. Conf. Electronic Packaging Technology, 2003, p. 309. 4) Materials for Advanced Packaging, ed. D. Lu and C. P. Wong (Springer, Heidelberg, 2009). 5) A. C. M. Chong and Y. M. Cheung: Proc. 7th IEEE CPMT Int. Conf. High Density Microsystem Design and Packaging and Component Failure Analysis, 2005, p. 36. 6) Y. M. Cheung, S. W. Or, and S. Ching: Proc. 24th IEEE/CPMT IEMT Symp. SEMICON , Southwest, 1999, p. 196. 7) Q. Tan, W. Zhang, B. Schaible, L. J. Bond, T. H. Ju, and Y. C. Lee: Proc. 47th ECTC, 1997, p. 1128. 8) K. W. Lee, H. J. Kim, M. J. Yim, and K. W. Paik: Proc. 56th ECTC, 2006, p. 918. 9) S. W. Or, H. L. W. Chan, and P. C. K. Liu: Sens. Actuator A 133 (2007) 195. 10) J. R. Frederick: Ultrasonic Engineering (Wiley, New York, 1965). 11) http://www.ansys.com 12) ANSI/IEEE Std. 176-1987 (1987). 13) I. Watanabe, T. Fujinawa, M. Arifuku, M. Fujii, and G. Yasushi: Proc. 9th Int. Symp. Advanced Packaging Materials, 2004, p. 11. 14) M. F. Rosle, I. Abdullah, S. Abdullah, M. A. A. Hamid, A. R. Daud, and A. Jalar: Am. J. Eng. Appl. Sci. 2 (2009) 17.
07GM19-6
# 2009 The Japan Society of Applied Physics
