Wafer Bonding: Selecting The Right Process For Making Powerful, Vertical LEDs
LEDs with a vertical geometry are promising candidates for deployment in solid-state lighting products because they can handle the high drive currents needed to deliver a high luminous output. Manufacturing this form of LED requires a wafer-to-wafer bonding process,which involves many variables that need to be optimised for the specific device design, say Thomas Uhrmann, Eric Pabo,Viorel Dragoi and Thorsten Matthias from EV Group.
White LEDs are already impacting the general lighting market, and their penetration in this sector is widely expected to rise. The rate of adoption will be governed by three factors: luminous efficiency, cost per lumen installed, and lumens per socket.
One way to improve all three areas simultaneously is to increase LED efficiency. But even greater gains to the lumen output of the luminaire and its cost-perlumen are possible by combining gains in efficiency with a higher drive current for the device. Cranking this up, however, increases LED heating. And to cope with this, the system designer must carefully manage heat that flows from the device junction to the package, fixture and surrounding environment.
It is possible to increase the rate that heat flows out of the LED by using metal wafer bonding for transfer of the epistructure to another substrate. Take this step and the LED benefits on two fronts: it can rapidly conduct heat away through a metal bond with a low thermal resistance, and it can dissipate heat through a substrate with low thermal resistance.
This approach can not only enhance the electrical properties of the nitride-based white LED, but also its blue variant and its red, orange and yellow cousins that are made from the AlInGaP material family. At EV Group, which is based in St. Florian, Austria, we are supporting the manufacturing of LEDs produced with a metal bonding process. Our involvement includes the recent launch of the first tool dedicated to this fabrication step – the EVG 560HBL. This piece of equipment is designed to deliver very high yields thanks to optimisation of pressure and temperature distributions, and it sets a new benchmark for throughput of up to 176 bondsper- hour for 2-inch wafer equivalents.
A little history…
Wafer-to-wafer bonding is not new. It was developed20-30 years’ ago to address the need for wafer-levelcapping of MEMS devices. Pioneers of waferbondingused anodic bonding and glass frit bondingto attach one wafer to another. However, these twoapproaches are being superseded by metal bondingtechnologies, which offer a lower form factor.
The metal bonding approach is the only one that is applicable to high-brightness LEDs, due to the requirement for low thermal resistance. This is not the only benefit of this type of bond, however – it can also increase the luminous efficiency of the device. It was first used in AlInGaP-based LEDs that are grown on GaAs substrates. Spontaneous emission from these devices is assumed to be isotropic, with half of all the light generated propagating towards the substrate, where most of it is absorbed, leading to lowering of overall device efficiency. Inserting a distributed Bragg reflector beneath the light-generating region of the LED could prevent this light loss to the substrate, but in practice this only works effectively on one optimised direction of light emission, and it is better to turn to wafer bonding, where a reflective layer is included in the metal stack (see Figure 1).
Figure 1: A key features of the vertical LED is its highly reflective sub-structure that excels at conducting the heat away. This allows the device to operate at high currents, and consequently high luminous output
Manufacturing nitride LEDs with a metal bonding process presents some different challenges. Sapphire, the most widely used platform for making blue and white LEDs, has the desirable attribute of high transparency, but it is a poor heat conductor. Consequently, high-power LEDs employing a lateral design are poor at dissipating their heat and run hot, which degrades device performance. To combat this, some LED manufacturers have developed vertical LED designs, which involve substituting sapphire for another carrier with higher thermal conductivity (see Figure 1).
Switching to this design also simplifies the manufacturing process by eliminating an etching step required to form the n-contact in a lateral LED. In addition, the vertical architecture produces a vertical current path, leading to a lower forward bias and eliminating current crowding issues that are frequently seen for other LED designs. And there are other benefits too: the addition of the metal bonding layer ensures that all of the light exits from the top of the LED; and manufacturing may be simplified, because the vertical LED design uses the same process flow for different die sizes.
Every vertical LED process flow begins by depositing a stack of epitaxial layers on a substrate by MOCVD. Some engineers will then turn to the patterning of the LED dies, while others will begin with layer transfer by wafer bonding. The decision of what order to perform the various processing steps is primarily governed by the nature of downstream processing used to remove the growth substrate and individual design differences.
Manufacturers of AlInGaP LEDs tend to carry out full-area wafer bonding prior to patterning, because this process does not introduce strain and the substrate can be removed by grinding and chemical removal. Making nitride LEDs is more complicated, however. When sapphire is separated from the epitaxial film by the widely employed process of laser lift-off, strain is induced at a die or in multiple die regions. This strain stems from decomposition of the interfacial GaN layer, which is triggered by high power ultraviolet irradiation. Cracking of the active layers can occur, impacting yield, but this can be addressed by confining the strain region via previous structurisation of the die.
In addition to high thermal conductivity, the bondinterface in a vertical high-brightness LED musthave excellent electrical conductivity. Fortunately,high thermal conductivity and high electricalconductivity tend to go hand-in-hand, and arefound in germanium and metallic substrates. Both ofthese are popular, but silicon is emerging as acarrier material, featuring high heat dissipation andlow thermal expansion. Using silicon also enablesvertical LED producers to include a Zener diodedirectly into the carrier substrate, which serves forthe electrostatic protection of the sensitive GaN LEDs.
Thermal expansion coefficients must also be considered when selecting a substrate material to bond to the epiwafer. All common wafer-to-wafer bonding processes require elevated temperatures and the bond process is fine tuned to accommodate thermal expansion mismatch between the substrates.
The bonding process must be based on metallic bonding layers because the technique used for wafer-to-wafer bonding must meet requirements for high thermal and high electrical conductivity. This limits the bonding process choices to solder (which includes eutectic and transient liquid phase) and thermo-compresion bonding. Both approaches are discussed in detail in the side panel “Attaching the wafers together".
Figure 2: A scanning electron microscopy cross-sectional image highlights the high quality of the join between an InP and GaAs wafer after Au:Sn eutectic wafer bonding
To decide which process is most appropriate, makers of vertical LEDs must consider the characteristics of both the growth and carrier substrates, and account for differences, such as those in thermal expansion and surface properties. For AlInGaP LEDs, which usually have a highly flat surface thanks to their lattice match to the GaAs substrate, thermo-compression bonding between two gold surfaces is frequently used.
Figure 3: The EVG560HBL, which is available in both fully automated and semi-automated operation, supports metal, adhesive and fusion bonds of various substrate types. EVG claims that it is well suited to high-volume LED manufacture, thanks to its combination of cassette-to-cassette operation, multisubstrate bonding capability and its modular design that features up to four swap-in process modules. According to the toolmaker, the combination of fieldproven wafer bonding technology and a unique approach to low-temperature metal wafer bonding results in unprecedented throughput and yield
For InGaN LEDs, however, surface roughness and defect density are considerably larger and eutectic or transient liquid phase bonding is preferred. This tends to result in a high yielding bonding process (see the example of Au:Sn bonding in Figure 2).
Further decisions to be made by any vertical LED manufacturer include the thickness of the bonding metals and how they are deposited, and also the selection of adhesion and barrier layers. These layers – typically made from platinum, aluminium and gold, or stacks combining these metals – are often needed to ensure sufficient adhesion and prevent migration of bonding metals into either of the substrates.
Yet another choice facing any vertical LED manufacturer is the choice of tool that will use for the bonding process, which will include alignment of the wafers prior to bonding. We believe that our EVG 560 HBL is worthy of consideration, thanks in part to its combination of high throughput and versatility – it supports metal, adhesive and fusion bonds of various substrate types. It also delivers high yields, so it is capable of helping to drive far greater adoption of LEDs in general lighting, by helping to make these devices deliver more light more efficiently.
© 2011 Angel Business Communications. Permission required.
Further reading E.F. Pabo, V. Dragoi, “Wafer Bonding Process Selection", MEMS Industry Group
Attaching the wafers together
The two approaches to bond one wafer to another during the manufacture of high brightness LEDs are solder bonding (also known as eutectic or diffusion bonding) and thermocompression bonding.
Solder bonding is a general term for a metal bond formed by liquid metal, which could be a pure metal, but is typically a binary alloy and in some cases a ternary one. A eutectic wafer bonding alloy is formed at the bonding interface in a process which goes through a liquid phase: for this reason, eutectic bonding is less sensitive to surface flatness irregularities, scratches, and particle contamination, compared to the direct wafer bonding methods.
A successful eutectic bonding process requires bonding equipment that combines good temperature control with temperature uniformity across the entire wafer. The temperature ramp for heating and cooling processes are important. Selection of the details of this process should depend on the particular substrate materials employed to avoid thermal shock for dissimilar materials, and should also be governed by device requirements. For example, process engineers must consider whether the device will be impaired by heating or cooling cycles.
The liquid melt formed during the bond process allows the embedding of interfacial particles in the melt without creating defects. Good wetting is achieved even on very rough surfaces, which are typical for InGaN-based LEDs. This contributes to enhanced device yield and performance.
For some high-brightness LED manufacturing process flows, the material should be kept below the bonding temperatures for the most usual eutectic alloys (300°C - 400°C). In such situations an alternative process can be used – diffusion soldering or transient liquid phase (TLP) bonding, which results in an inter-metallic compound bonding layer. This technique uses one thin metal layer – typically 1-10 μm thick – which inter-diffuses with its bonding partner during a thermal process to yield an intermetallic compound layer with a re-melting temperature higher than the bonding temperature. Cu-Sn and Au-Sn are the most popular TLP systems. Like eutectic wafer bonding, diffusion soldering bonding is an attractive option for high-brightness LED manufacturing, because it can planarize over surface defects or particles resulting from prior processes due to surface wetting by the molten metal.
The alternative process, thermo-compression bonding, involves adhesion of two surfaces to one another through diffusion of the metal molecules, such as gold, copper and aluminium, across the bond interface. The diffusion rate is a function of: the metal; the diffusion barriers on the surface, such as oxides; the pressure; the temperature; and the surface roughness. Cranking up the pressure increases the fundamental diffusion rate and also enhances diffusion. The latter results from deformation of the two surfaces in contact, which leads to disruption of any intervening surface films and enables increased metal-to-metal contact. As this diffusion continues, grain growth occurs across the bond interface. Heating the metal increases diffusion and slightly softens the metals, increasing deformation at any given pressure. Excellent bonding yield results when a high force capability combines with pressure uniformity across the bonding area.