CHAPTER 2 BACKGROUND AND THEORY
2.6 Wafer Bonding
To join two different materials, wafer bonding technique shows a benefit when compares with the ordinary growth technique because defects can be confined at the interface. Wafer bonding has two stages: first is wafer contacting and second is wafer binding. Surfaces of these two materials must be smooth while contact with each other. Weak forces form between them which appear in the following three types: (1) van der Waals force, (2) Capillary action, and (3) electrostatic force, as shown in Fig. 2.8.[8]
Although weak forces will form at the contact process, it still needs a high temperature annealing and a uniaxial pressure to construct high strength bonds between the two materials in binding process. By using a special treatment to the wafer surface, wafer binding can also be proceeded at low temperature. The weak bonds can finally be transferred to covalent bonds.
The exterior layers of all the solid state materials exist in non-equilibrium and reconstructed states. As we know, the property of this exterior layer is different from the inside which has higher energy. From the thermaldynamics point of view, the surface free energy is positive and the volume free energy is negative. In order to reduce the total energy, two bound surfaces tend to fuse together and eliminate the bonding interface. The driving force to the interface elimination is offered by the difference of the surface curvature. The following sentence expresses the relationship between the change of the free energy G and the radius of the curvature r.[9]
2
From this equation, we can easily see that the driving force is apparently inverse proportion to the radius of the surface curvature. When two wafers bound together, the micro-undulations lead to the diffusion of the atoms. As shown in Fig. 2.9, atoms at the sites with positive curvature (r > 0)
33
move to the negative ones (r < 0) using surface diffusion and lattice diffusion. When the binding completes, covalent bonds will form at the interface which are much stronger than the original contact force.
34
Figure 2.1 The schematic of a typical surface-emitting LED.
35
Figure 2.2 Intrinsic radiative transitions in semiconductors. (a) Band-to-band transitions; (b) free-exciton annihilation; (c) recombination of exciton localized at band-potential fluctuations.[1]
36
Figure 2.3 Band structure near a semiconductor p-n homojunction: (a) under zero bias; (b) under forward bias.[1]
37
Figure 2.4 Band alignment diagram of a quantum well structure.[1]
38
Figure 2.5 Behavior of light when traveling between two different mediums. Both reflection and refraction occur.
39
Figure 2.6 Illustration of the physical parameters of the concave mirror.
40
Figure 2.7 (a) Schematic illustration of the TLM patterns. (b) Plot of total contact as a function of L to obtain transfer length and contact resistance values.
41
Figure 2.8 Three situations when two wafers contact with each other.[8]
42
Figure 2.9 (a) Diffusion occurred because of the difference of surface curvature.
(b) Interface after annealing.
43
CHAPTER 3
EFFECT OF THE SILVER MIRROR LOCATION ON THE LUMINANCE INTENSITY OF DOUBLE-ROUGHENED GAN
LIGHT-EMITTING DIODES
3.1 Introduction
Recently, epitaxial growth techniques have significantly improved the luminance intensity of light-emitting diode (LED). GaN-based LEDs are attractive devices for use in a variety of applications including traffic signals, full-color displays, back lighting in liquid-crystal displays, and mini projectors.[1,2] However, the light extraction efficiency of GaN-based LEDs is limited by the large difference in refractive index between the GaN film and the surrounding air.
According to the Snell's law, light traveling from GaN to air travels only within a critical angle of 23°. The light reaching the surface beyond the critical angle will experience total internal reflection that will continue to be reflected within the LED until it is absorbed. Some methods have been used to reduce the percentage of total internal light reflection by roughening the top p-GaN surface.[3-8] It could also utilize wafer-bonding and laser lift-off technologies to transfer
the random texturing of undoped-GaN (u-GaN) surface onto Si substrates,[8-10] or bond various mirror systems between the GaN LED structures and the substrate.[9] In this study, GaN LEDs with double roughened (p-GaN and u-GaN) surfaces and a silver (Ag) mirror on the sapphire substrate were successfully fabricated using wafer-bonding and laser lift-off technologies. Effect of the Ag mirror location on the luminance intensity of LEDs was investigated.
44
3.2 Experiments
3.2.1 Illustration and abbreviation of LEDs
Three kinds of LEDs were used in this study. As illustrated in Fig. 3.1, samples designated as “PR-LED” were LEDs with roughened p-GaN surface. Samples designated as “DRM-LED”
and “DRSM-LED” were LEDs with double roughened (p-GaN and u-GaN) surfaces and an Ag mirror system either between the GaN LED structure and sapphire substrate or on the backside of sapphire substrate.
3.2.2 Crystal growth and device process of PR-LED
The devices processes for PR-LED, DRM-LED and DRSM-LED were the same and were shown in Fig. 3.2. The InGaN-GaN films were grown by low-pressure metal organic chemical vapor deposition (MOCVD) on a sapphire substrate. The LED structures were consisted of a 5-nm-thick Si-doped n+-InGaN tunnel contact structure, a 0.4-µm-thick Mg-doped GaN layer, an InGaN-GaN multiple quantum well (MQW), a 2-µm-thick Si-doped GaN layer, a 2-µm-thick u-GaN layer film and a buffer layer on a sapphire substrate. The Si-doped n+-InGaN layer was used to form the ohmic contact between ITO and p-GaN.[11] The device mesa with a chip size of 300×300 µm2 was then defined by an inductively coupled plasma (ICP) to remove Mg-doped GaN layer and MQW until the Si-doped GaN layer was exposed. Then, the indium tin oxide (ITO) layer was deposited on the n+- InGaN layer to form a p-side contact layer and a current spreading layer. The Cr/Au layer was deposited onto the ITO layer to form the p-side and n-side electrodes.
The roughened p-type GaN surfaces of PR-LED were formed by lowering the epitaxy growth temperature of the p-type GaN layer.[7,8]
45
3.2.3 Fabrication of DRM- and DRSM-LEDs
The PR-LED was utilized to fabricate the DRM-LED and DRSM-LED. As shown in Fig.
3.3, the PR-LED wafer was bonded to a host substrate covered with an adhesive layer. After bonding, the sapphire substrate was then removed by irradiating it with KrF pulsed excimer laser at 248 nm, through the transparent sapphire substrate, which locally decomposed the GaN layer at the boundary between the GaN layer and sapphire substrate. After scanning the entire sample, the sapphire substrate was successfully separated. Ga residues were then removed by wet chemical etching using diluted HCl:H2O (1:1) solution for 60 s. Subsequently, the u-GaN epitaxial layer was treated with 45% KOH solution for 1 minute at 90°C to obtain a roughened u-GaN surface.[8,9] To fabricate the DRM-LED, the above double roughened GaN structures
were bonded to an Ag coated polished sapphire surface with an adhesive layer. The structures of DRM-LED and DRSM-LED were shown in Fig. 3.1. As for the fabrication of DRSM-LED, the double roughened structures were bonded to a double polished sapphire wafer. Then, an Ag mirror was coated on the backside of sapphire substrate. The bonding adhesive layer that consisted of a polycyclic aromatic hydrocarbon (C8H6) composed of a benzene ring fused to a cyclobutene ring. The optical transparency of the adhesive layer was exceeding than 90% across the visible spectrum. Wafers were annealed at 200°C for 60 min with a comprehensive load of 10 kg/cm2. The host substrate and glue layer were subsequently removed. For baseline comparison the performance of the PR-LED, DRM-LED and DRSM-LED were prepared from the same InGaN-GaN LEDs epitaxial structure, and sapphire was chosen as the substrate of DRM-LED and DRSM-LED. The samples described herein were only cut into chips without encapsulation.
46
3.3 Results and Discussions
To reflect the downward photons, an Al mirror was replaced with an Ag mirror, because the reflectivity of Ag is higher than that of Al at a wavelength of 470 nm. The poor adhesion of the Ag mirror with other materials is a well known issue. This problem can be solved by inserting an intermediate layer between them, like Ni and Ti which are identified as a good adhesive material.
A post annealing process could be necessary to enhance the adhesion, but it could also degrade the reflectivity of the mirror. To identify the point, two samples were prepared using Ti as an adhesion layer (thickness = 10 nm) on sapphire substrate, followed by a 200 nm Ag film deposited as a mirror. One sample was then treated at 300℃ for 24 hr in Ar ambient, while the other with no treatment. Samples with only Ag and Al mirror were also prepared for comparison.
As shown in Fig. 3.4, sample with Ti/Ag as deposited showed high reflectivity (>90%) among the visible spectrum, nearly the same with Ag as deposited. When Ti/Ag was after annealing, the reflectivity of it still maintained high and showed 92% at around 470 nm. Compared with Al mirror, Ti/Ag after annealing not only performed a better ability of reflecting the downward photons, but gives a comparable reliability when the sample is under processing. So it was chosen as the mirror of the DRM- and DRSM-LEDs.
Figure 3.5 shows the sideview scanning electron microscopy (SEM) micrograph of the u-GaN layer with the treatment of 45% KOH solution for 1 min at 90 °C. The SEM image was
taken at an angle of 45° from the perpendicular direction of the etched surface. The sample was measured using atomic force microscopy (AFM) to identify the degree of texturing. The root-mean-square (rms) roughness of the u-GaN layer was 125 nm.
Figure 3.6 shows the current-voltage (I–V) characteristics of the LEDs. It was found that the forward voltages of the DRM-LED and DRSM-LED were about 3.65 V at 20 mA, which was
47
similar to that of PR-LED (3.5 V), indicating that transfer method did not change much of the LED performance.
Figure 3.7 depicts the effects of injection current on the luminous intensity of the LEDs.
During the testing, LEDs were put onto a graphite plate with a collection angle of approximately 27°. The light intensities of DRM-LED and DRSM-LED were much greater than that of PR-LED.
The light intensity of DRSM-LED was 162.4 mcd, which was 2.10 times higher than that of the PR-LED at an injection current of 20 mA. These results are similar to the conclusions drawn by Peng et al.[9] during their studies on the enhancement of the GaN LED light intensity by roughening the u-GaN surface and applying a mirror coating to the backside of sapphire substrate.
The luminance intensity of roughened-mirror-LED (RM-LED) was greatly enhanced because the roughened u-GaN surfaces not only provided the photons multiple opportunities to escape the LED surface, but also redirected the photons which were originally emitted out of the escape cone, back into the escape cone. By adding a mirror to the backside of the sapphire, the light intensity was further enhanced by redirecting the downward-traveling light back to the surface of the LED. They found that the light intensity of RM-LED was 2 times higher than that of the conventional LED (C-LED). In contrast to our study, the roughened u-GaN surface and Ag mirror were applied to the PR-LED instead of C-LED. Compared with PR-LED, the light intensity of DRSM-LED was also increased by a factor of 2.
On the other hand, the light intensity of DRM-LED was 235.8 mcd, which was 3.05 times higher than that of the PR-LED, and 1.45 times higher than that of the DRSM-LED. Clearly, this further enhancement of light intensity was caused by the change of the Ag mirror location. There are two possible explanations for the effect of the mirror location on the light intensity: (1) the bonded interface and (2) the total internal reflection in sapphire. As mentioned earlier, redirecting the downward-traveling light back to the top surface enhanced the LED light intensities. When
48
the redirecting light passes through the bonded interface, the Fresnel losses resulting from various materials might have a negative effect on the luminance intensity. Besides, during the bonding process, we might create interfacial defects at the bonded interface. These defects also have a negative effect on the optical properties. As shown in Fig. 3.8, compared with the DRM-LED, the photon path inside the DRSM-LED structure has to pass through an extra bonded interface (adhesive layer (n~1.5) /sapphire (n~1.7)) 2 times. This may explain why the luminance intensity of DRSM-LED was less than that of DRM-LED.
The other factor that may affect the light intensity is the internal reflection in sapphire.
When the mirror of DRSM-LED redirecting the downward-traveling light, according to Snell’s law, light traveling from sapphire (n~1.7) to adhesive layer (n~1.5) will only cross within a critical angle of 61.9°. The light reaching the adhesive layer beyond the critical angle will undergo internal reflection and continue to be reflected within the sapphire. As for the DRM-LED, since the Ag mirror located at the adhesive layer/sapphire interface, no light will penetrate through the Ag mirror. As a result, the light intensity of DRM-LED was higher than that of the DRSM-LED.
3.4 Summary
Three kinds of LEDs were used to investigate the effect of the mirror location on the performance of LEDs. PR-LED was LED with roughened p-GaN surface. The PR-LED was utilized to fabricate the DRM-LED and DRSM-LED by wafer-bonding, laser lift-off and surface-roughening technologies. It was found that the light intensity of DRSM-LED was 2 times higher than that of PR-LED. This enhancement is not surprising, since the Ag mirror redirected the downward-traveling light back to the top surface. By changing the mirror location from the
49
backside of the sapphire to the u-GaN/sapphire interface, the LED light intensity was further enhanced. This is because, compared with DRM-LED, the photon path inside the DRSM-LED structure has to pass through an extra bonded interface (adhesive layer/sapphire) 2 times. Besides, when the mirror of DRSM-LED redirecting the downward-traveling light, light traveling from sapphire to the adhesive layer will only cross within a critical angle of 61.9°. The light reaching the adhesive layer beyond the critical angle will undergo total internal reflection.
50
Figure 3.1 Schematic diagrams of (a) PR-LED, (b) DRSM-LED, and (c) DRM-LED.
51
Figure 3.2 Fabrication process of the PR-LED.
52
Figure 3.3 Experiment flowchart.
53
400 450 500 550 600 650
70 75 80 85 90 95 100
Reflectivity (%)
Wavelength (nm)
Ag as deposited Ti/Ag as deposited
Ti/Ag 300oC 24h annealing in Ar Al as deposited
Figure 3.4 Reflectivity of different mirrors.
54
Figure 3.5 SEM image of the u-GaN surface etched by 90oC KOH for 1 min.
55
0 50 100 150 200
0 1 2 3 4 5 6 7 8
Voltage (V)
Current (mA)
DRM-LED DRSM-LED PR-LED
Figure 3.6 I-V characteristic of the LEDs.
56
0 50 100 150 200
0 200 400 600 800 1000 1200
Luminance Intensity (mcd)
Current (mA)
DRM-LED DRSM-LED PR-LED
Figure 3.7 The effects of injection current on the luminous intensity of the LEDs.
57
Figure 3.8 Possible photon paths inside the structures of the (a) DRSM-LED and (b) DRM-LED.
58
CHAPTER 4
IMPROVED PERFORMANCE OF INGAN-GAN LIGHT-EMITTING DIODE BY A PERIODIC N-BOWL MIRROR ARRAY
4.1 Introduction
High-brightness GaN-based light-emitting diodes (LEDs) for blue and ultraviolet light sources have attracted intense research for their versatile applications in mobile phones, full-color displays, mini projector and lightings in the past few years.[1] Internal quantum efficiency and light extraction efficiency are two principal parameters for improving the efficiency of LEDs;
both of which can be improved through the improvement of crystal quality and the modification of the LED structure. The internal quantum efficiency of GaN LEDs can exceed 70% now, which is much higher than the 10%-25% efficiency of conventional light sources.[2] However, the light extraction is limited by their inability to emit all of the light that is generated from the active layer.[3] According to Snell’s law, the critical angle for light traveling from GaN (n=2.5) to air (n=1.0) is 23°. Only the light within the critical angle will cross the air, while the other experience total internal reflection until it is absorbed. Many methods have been developed to overcome this problem. These methods were not beyond the scope of changing the shaping and surface morphology of the LED, which included random texturing of the LED’s surface,[4]
sidewall obliquing by dry etching,[5, 6] pattern sapphire application,[7, 8] and etc.. In this study, a periodic n-bowl mirror array was introduced into LEDs to improve the LED performance and adjust the view angle.
59
4.2 Experiments
4.2.1 Illustration and Abbreviation of LEDs
Two kinds of LEDs were investigated in this study. Their specifications and structures are schematically illustrated in Fig. 4.1(a) and (b). Samples designated as “PR-LED” were LEDs with roughened p-GaN surface, while samples designated as “NBM-LED” were LEDs composed of roughened p-GaN surface, n-bowl structure, and mirror on n-bowl surface. Three types of NBM-LEDs were investigated. They were denoted as 25-4 (25 μm diameter and 4 μm spacing, as shown in Fig. 4.1(c)), 4-3 and 3-3 NBM-LEDs, as listed in Table 4.1. The cross section scanning electron microscope (SEM) image of 4-3 n-bowl structure was shown in Fig. 4.1(d).
4.2.2 Crystal Growth and Device Process of PR-LED
The basic processes of these LED were the same, and were shown in Fig. 4.2. The InGaN-GaN films were grown by low-pressure metalorganic chemical vapor deposition (MOCVD) on a sapphire substrate. The LED structures were consisted of a 5-nm-thick Si-doped n+-InGaN tunnel contact structure, a 0.4-μm-thick p-type Mg-doped GaN layer, an InGaN-GaN multiple quantum well (MQW), a 2-μm-thick Si-doped GaN layer, a 2-μm-thick undoped-GaN (u-GaN) layer film and a buffer layer on a sapphire substrate. The Si-doped n+-InGaN layer was used to form the ohmic contact between ITO and p-GaN.[9] The device mesa with a chip size of 300×300 μm2 was then defined by an inductively coupled plasma (ICP) to remove Mg-doped GaN layer and MQW until the Si-doped GaN layer was exposed. After annealed at 600 °C for 10 min, the indium tin oxide (ITO) layer was deposited to form a p-side contact layer and a current spreading layer. The Cr/Au layer was deposited onto the ITO layer to form the p-side and n-side
60
electrodes. The roughened p-type GaN surfaces of PR-LED were formed by lowering the epitaxy growth temperature of the p-type GaN layer.[10, 11]
4.2.3 Fabrication of NBM-LEDs
Fig. 4.3 shows the fabrication processes of the NBM-LEDs. To fabricate the NBM-LED, PR-LED wafer was bonded to a host substrate using a glue layer.[9] The bonding process was performed at 200 °C for 1 hour. After bonding, the sapphire substrate was lifted off by irradiating it with KrF pulsed excimer laser (248 nm) to expose the u-GaN surface. To fabricate the n-bowl structure, photoresist reflow and dry etching method were used.[12] A photoresist layer was spun onto the u-GaN surface and patterned by standard photolithography into cylindrical disks. When baked at the temperature higher than 200 °C, they turned to hemispherical shapes because of the surface tension. The performed shapes were then transferred to u-GaN using ICP process. Fig.
4.4 shows the SEM image of the 4-3 n-bowl structure. A mirror system (Ni/Ag/Ni) was then introduced on the etched surface to reflect the downward photons.[13] Sample was subsequently bonded to Si substrate with an adhesive layer.[9] The host substrate and glue layer were subsequently removed. For baseline comparison the performance of the PR-LED, sapphire was chosen as the substrate of NBM-LED. With the same sapphire substrate, we could investigate the influence of n-bowl structure on the performance of the LED chips. The samples described herein were only cut into chips without encapsulation.
4.3 Results and Discussions
Fig. 4.5 shows the current-voltage (I–V) characteristics of LEDs. It was found that the
61
forward voltages of NBM-LEDs were in the range of 3.4-3.6 V (at 20 mA), which were similar to that of PR-LED, indicating that transfer method did not change much of LED performance. The
forward voltages of NBM-LEDs were in the range of 3.4-3.6 V (at 20 mA), which were similar to that of PR-LED, indicating that transfer method did not change much of LED performance. The