Figure 2-1 shows a successful sample of an AlGaInP LED wafer bonded on a Cu substrate at 500°C for 30 min. After removing the GaAs substrate, no peeling or cracks were observed on the mirror-like surface despite the large difference between the thermal expansion coefficients of GaAs (6.86×10-6/°C) and Cu (16.8×10-6/°C). The fused epilayer then was mesa-etched into isolated devices, deposited with contact dots, and then alloyed for ohmic contact metalizations.
Figure 2-2 shows the cross section image of this bonded LED structure. Neither cavity nor unbonded area was found on the bonded interface. The Aauger depth profiles of AlGaInP LED/ ITO/ Cu substrate are shown in Fig. 2-3. It revealed no Cu element diffused into the AlGaInP active layer of LED structure. For the purpose of comparison, an ITO-free AlGaInP LED substrate was also bonded to Cu substrate at same condition. The Cu element was found to diffuse into the AlGaInP LED structure and destroy the LED structure. The results indicated that ITO layer employed here not only acted as a current spreading layer but also a barrier layer.
When ITO film was used as the bonding intermediate layer, it was also found that the bonding temperature would affect the bonding yield of the Cu-substrate LED. The sample did
27
not bond at temperatures below 400°C. At temperatures greater than 400°C, the bonded area increased when temperature increased as a result of both the reaction and diffusion rates increasing with temperature. However, when bonding temperature reached 600°C, the Cu element could penetrate through the ITO layer and destroy the LED structure.
Figure 2-4 presents the current versus voltage (I-V) of the Cu-substrate LED bonded at 500°C for 30 min. The device exhibited normal p-n diode behavior with a forward voltage of 1.96 V at 20 mA, which was similar to that of GaAs-substrate LED. This indicated that the wafer-bonding process did not degrade the performance of LED. It is worthy to note that the series resistance of Cu-substrate LED (1~2 Ω) is lower than that of GaAs-substrate LED (5-7 Ω) [9]. A small series resistance is the most important factor for decreasing the Joule heating effect, thus it indicates an increase in the quantum efficiency of the LED.
The effects of heat sink observed in high power LED are very important with regard to the removal of heat from LED devices. The performance of the LED would otherwise degrade when the operation temperature increased. Figure 2-5 shows the peak spectral wavelength as a function of the DC drive current. During the test conditions, these two LED samples, which had been cut into chips without encapsulating, were put on a metal chunk.
The data shows that the emission peak wavelengths shift toward longer wavelengths with increasing DC drive current, which is caused by the joule heating. Unlike the 4nm red shift exhibited at 170 mA in GaAs-substrate LED, the Cu-substrate LED devices exhibits a more favorable 2 nm red shift at 170 mA. The red shift of the emission wavelength occurred as the temperature increased. From the shift in emission wavelength and a wavelength shift with temperature of 0.096 nm/°C for 620 nm AlGaInP LED [10], it could be estimated that the thermal resistance of the Cu-substrate LED was approximately 60 °C/W smaller than that of GaAs-substrate LED. Thus the temperature of GaAs-substrate LED was about 20°C higher than that of the Cu-substrate LED at the forward current of 170 mA.
28
Figure 2-6 shows the effects of injection current on the luminous intensity of the conventional GaAs-substrate LED and the Cu-substrate-bonded LED. When the injection current was less than 100 mA, the luminous intensities of two samples were the same, meaning that the wafer-bonding process did not degrade the performance of Cu-substrate LED. When the injection current reached 100 mA, the GaAs-substrate LED were saturated at the luminous intensity of 400 mcd. However, the intensity of the Cu-substrate LED did not saturated at 100 mA. The maximum luminous intensity of Cu-substrate LED could reach as high as about 1230 mcd at 800 mA, which was three times higher than the saturation intensity of the GaAs-substrate LED at 100 mA. It was obvious that the luminous intensity of Cu-substrate LED was much higher than that of GaAs-substrate LED.
These results were due to a much smaller series resistance of the Cu-substrate LED as compared with GaAs-substrate LED, thereby reducing the total amount of joule heating.
Since the reduction of the joule heating would increase the quantum efficiency of the LED.
Therefore, Cu-substrate LED had a smaller red shift, and could be driven at a higher current and therefore obtain a higher luminous intensity. This difference might also be a result of the disparity between their respective thermal conductivities; the thermal conductivity of Cu (401 Wm-1K-1) is nine times higher than that of GaAs (46 Wm-1K-1). Therefore, by using these wafer bonding techniques to bond LED to Cu substrates, the joule-heating problem in conventional LED can be significantly reduced.
The other important issue of the Cu-substrate LED is the reliability. Copper is an impurity long known to cause strong degradation in LED [11]. The efficiency of Cu-contaminated LED would significantly decreased after life test. In this study, the life test of Cu-substrate LED was performed at forward current of 20 mA in room temperature, corresponding to 32 A/cm2. As shown in Fig. 2-7, the degradation of the luminescence- output intensity was less than 5 % after 500 hours of life test.
29
2-3 SiC-substrate LED
2-3-1 Experiments
The AlGaInP MQW LED structures were grown on GaAs substrates through low-pressure metalorganic chemical vapor deposition (MOCVD). An Au/AuBe/Au layer was deposited on the AlGaInP LED to act as an ohmic contact and mirror layer. On the other hand, the bonding metals of Ti/Au/In were deposited on the host SiC substrate that played the role of a heat sink substrate. The thermal conductivity of single crystal SiC substrate was 370 Wm-1K-1. The purpose of choosing the indium as the bonding metal between the LED structure and SiC was to achieve good bonding yield of bonded-sample during low temperature (< 300°C). These two wafers were cleaned in acetone and isopropyl alcohol with ultrasonic agitation, followed by rinsing in de-ionized (DI) water. Following the surface treatment, the samples were dried by nitrogen gas and placed in contact with each other immediately using a fixture that ensured uniform pressure on each of the wafers. Then, the fixture was annealed at temperatures of 250°C for 60 min. After the wafers were removed and were allowed to cool to room temperature, the absorbing GaAs substrate and the InGaP etching stop layer were removed by chemical etching in solutions of 1NH4OH: 10H2O2 and 1HCl: 10H2O, respectively. The fused epilayer was mesa-etched into isolated devices with 300 × 300 µm2 area before the ITO layer was deposited on the n-GaAs ohmic contact layer.
The Cr/Au with a diameter of 100 µm and Ti/Au were deposited onto the ITO layer and the backside of SiC substrate, respectively. Following the wafer was alloyed for ohmic contact metalizations. Figure 2-8 shows the schematic diagram of the flowchart bonded LED. In comparison, the LED samples with the GaAs and SiC substrates were prepared from the