GaSb deoxidation without altering the surface topography.
Fully bonded wafer pairs with high bonding strengths.
Optical transparency of the boundary layers.
Low interface resistivities <5 mΩ cm2 by optimization of the process parameters.
Direct wafer bonds of the material system n-GaSb/n-Si have been achieved by means of a low-temperature direct wafer bonding process, enabling an optical transparency of the bonds along with a high electrical conductivity of the boundary layer. In the used technique, the surfaces are activated by sputter-etching with an argon fast-atom-beam (FAB) and bonded in ultra-high vacuum. The bonds were annealed at temperatures between 300 and 400 °C, followed by an optical, mechanical and electrical characterization of the interface. Additionally, the influence of the sputtering on the surface topography of the GaSb was explicitly investigated. Fully bonded wafer pairs with high bonding strengths were found, as no blade could be inserted into the bonds without destroying the samples. The interfacial resistivities of the bonded wafers were significantly reduced by optimizing the process parameters, by which Ohmic interfacial resistivities of less than 5 mΩ cm2 were reached reproducibly. These promising results make the monolithic integration of GaSb on Si attractive for various applications.
Direct wafer bonding;
Argon-beam surface activation;
The monolithic integration of III–V semiconductors on silicon offers a wide range of innovative applications, like in the fields of power electronics , photonicsand photovoltaics and . III–V layers on silicon allow combining favorable material characteristics of compound semiconductors with the low-cost, the mechanical stability and the advantages in the processing of silicon. Hence, direct wafer bonding between GaAs/Si and and InP/Si and was investigated extensively in the past years. Likewise, GaSb offers a wide choice of promising and unique characteristics such as small effective masses, high electron and hole mobilities and a band gap suitable for long-wavelength optical devices . In this paper, we report about the development of a direct wafer bonding process for the formation of transparent and electrically conductive GaSb/Si heterojunctions with promise for the integration of antimonide layers on silicon. The low-temperature bonding process, which was used, is carried out mainly analog to the approach of Suga et al., which was first published in 1992 for the formation of Al/Al and Al/Si3N4 wafer bonds .
The activation of the semiconductor surfaces is achieved by removing of the oxides and contaminations by sputtering with an argon beam in a vacuum environment of the reactor chamber (<3 × 10−6 Pa). At the same time, the crystal lattice is destroyed in the first few nanometers creating an amorphous layer . The polished wafers are pressed together, bringing the activated surfaces in close contact. This enables dangling bonds on the surfaces to form covalent bonds, permanently joining the semiconductors .
An Ayumi SAB-100 wafer bonder was used to bond 4 in. monocrystalline Si wafers to 2 in. monocrystalline GaSb wafers. The bonder contains two saddle field FAB sources, where the Ar atoms are first ionized, then accelerated and finally neutralized. In this setup, the energy of the argon atoms is approximately 0.4–0.6 times the product of the elementary charge and the acceleration voltage. The 300-μm thick Si wafers with an (1 0 0) orientation received a thermal phosphorous diffusion, leading to an n-type doping of 1 × 1020 cm−3 within the first 50 nm. A 500-nm thick GaSb epitaxial layer with an n-type (Te) doping of 1 × 1018 cm−3 was grown onto the GaSb wafers, which have an orientation of (1 0 0) 4° off towards (1 1 1) A. This doping concentration is in the range of the highest active doping concentration achievable for n-type GaSb A high surface carrier concentration is beneficial to overcome potential barriers at the interface, thus achieving a high conductivity over the heterojunction.
An essential condition for a successful process is, that the wafer surfaces have a RMS roughness <1 nm . This was accomplished by means of chemical-mechanical polishing, resulting in a RMS roughness of 0.2 ± 0.03 nm for the Si substrates and of about 0.5 ± 0.05 nm for the GaSb substrates. No further surface cleaning treatment was applied prior to the bonding process. Furthermore, it is elementary that the roughness is not significantly increased by the FAB treatment.
Therefore the influence of the Ar bombardment on the GaSb surface was investigated using different durations of the sputter treatment (between 2 and 60 min) and argon acceleration voltages (500, 620 and 740 V). The RMS roughness of the GaSb surfaces was measured before and after the sputter-etching at 5 random locations of the sample by means of atomic force microscopy (1 μm × 1 μm scan fields with a resolution of 512 × 512 pixels).
Several bonds were processed, at which some essential process parameters were varied. This includes the temperature during the sputter-etching (20 and 120 °C), which represents also the bonding temperature at which the wafers are pressed together. Furthermore, the acceleration voltage of the Ar ions (500, 620 and 740 V) and the duration of the sputtering on the GaSb wafers (4, 8 and 12 min) were varied. The Si substrates were always sputtered for 8 min with an Ar acceleration voltage of 620 V. The anode current, which correlates to the sputter dose , was kept constant at a value of 50 mA. The wafers were pressed together directly after the activation for 5 min with a force of 10 kN, to overcome remaining variations of the wafer thicknesses. Lastly, the temperature of the post-annealing treatment was varied (1 min at 300, 350 and 400 °C).
The bond interfaces were macroscopically investigated for the presence of voids by means of scanning acoustic microscopy (SAM) using a frequency of 100 MHz for the measurement. The optical transparency of the bonds was examined via spectrometric transmittance measurements using a PbS detector, which is sensitive for light up to 2.5 μm. The bonding strengths were investigated with the Maszara crack-opening method and the electrical carrier transport over the GaSb/Si interface was investigated by dark IV-measurements. For this purpose, the bonded wafers were metallized on both sides and diced into 3 mm × 3 mm pieces, which were then post-annealed, as described above. A Ti/Pd/Au/Ag contact was chosen for n-GaSb and a Ti/Pd/Ag contact for n-Si. The actual resistance and IV-characteristic of the bond interface was determined, by subtracting the Ohmic metal contact and substrate resistances from the measured resistance of the metallized sample.
3. Results and discussion
3.1. GaSb surface roughnesses after deoxidation
Fig. 1 shows the resulting RMS roughness of the GaSb wafer surfaces after deoxidation by sputter-etching with varying durations of the Ar beam exposure and acceleration voltages.
Summary of the RMS roughness of the GaSb surfaces after sputtering with varying durations of the Ar-beam exposure and acceleration voltages.
It was found that within the first 10 min sputtering, the RMS roughness is not increased due to the sputtering with argon atoms accelerated at a voltage of 620 V. After 20, 40 and 60 min of the sputter treatment, the roughness increases to 0.7, 1.3 and 2.1 nm, respectively. As GaSb exhibits different binding energies for Ga and Sb in its crystal structure (Ga: 18.6 eV, Sb: 31.7 eV ), this behavior can be explained by selective sputtering of the GaSb, which becomes noticeable at longer sputtering durations after the oxide is already removed. Malherbe et al. reported on a comparable selective sputtering behavior of InP . In this case, the phosphide is preferentially sputtered and the mainly unbound indium on the surface forms agglomerations in an atomic scale.
With the use of an increased acceleration voltage of 740 V, the RMS roughness is already increased after 10 min of FAB treatment to a value of 0.9 nm, as there is extended damage induced by the higher energy of the argon atoms. An even lower acceleration voltage of 500 V on the other hand, leads to no measurable change of the roughness as it was the case for 620 V.
These findings are also represented by exemplary AFM images of the GaSb surface, which are shown in Fig. 2.
AFM images of the GaSb surface topography after: no sputtering treatment (a), 10 min sputtering with an acceleration voltage of 620 V (b), 10 min sputtering with an acceleration voltage of 740 V (c) and 60 min sputtering with an acceleration voltage of 620 V (d).
According to the changes of the RMS roughnesses, there is no significant altering of the GaSb surface topography after 10 min sputtering with an acceleration voltage of 620 V (Fig. 2(b)) compared to a polished GaSb surface without any sputter-etching (Fig. 2(a)). On the other hand, after 10 min sputtering with 740 V (Fig. 2(c)) or 60 min sputtering with 620 V (Fig. 2(d)) the altering of the wafer surface becomes obvious.
In the case of silicon, it was shown by Essig et al. and Howlader et al. that the RMS roughness is not altered by the Ar atom bombardment using comparable atom energies and doses  and .
3.2. Characterization of the GaSb/Si wafer bonds
The GaSb and Si wafers could be bonded successfully with all process parameters, except when an increased acceleration voltage of 740 V was used for the activation of the GaSb surface. In this case, the wafers could not be bonded, which can be explained by the roughening of the GaSb surface after the sputter-etching at an increased acceleration voltage (see Fig. 2(c)).
In general, the wafer bonds reveal only small circular areas which were not bonded due to the presence of particles. Otherwise, the bonds were complete up to the rounded bevel edge of the GaSb wafer. In Fig. 3, an exemplary SAM image of a GaSb/Si bond is shown. Here, both wafer surfaces were activated for 8 min with an Ar acceleration voltage of 620 V.
SAM image of an exemplary GaSb/Si bond. The darker region indicates the bonded area, as the light spots indicate the not bonded areas, where a higher reflection of the acoustic wave at the boundary surface is prevalent.
It was not possible to determine the bond strengths by the Maszara crack-opening method as the blade could not be inserted in the bond interfaces without destroying the GaSb substrate. This was the case for all successful bonded wafer pairs and speaks for a strong and mainly covalent bond at the interface, resulting in high bond energies, which are comparable to the energies in the GaSb bulk crystal. This effect was also reported by Kopperschmidt et al. for hydrophobic GaAs/Si bonds, which were annealed at 850 °C .
For the verification of the optical transparency of the GaSb/Si bond interfaces, the spectral transmittance of bonded GaSb/Si wafer pairs was compared with the transmittance trough sole GaSb and Si wafers. It showed that there is no significant absorbance of light intensity in the bond interface. This corresponds to the expectations, as the direct bonding process should not induce any absorbing layers at the boundary interface  and . Yet, it has to be noted that GaSb is absorbing until a wavelength of about 1750 nm, so just light with a higher wavelength could be measured.
In Fig. 4, IV-curves of the metallized samples are shown for a bonding (and deoxidation) temperature of 120 °C and different post-annealing steps at 300, 350 or 400 °C for 1 min each. One further IV-curve is shown for a bonding temperature of 20 °C with an annealing of 1 min at 350 °C.
IV-characteristics of n-GaSb/n-Si wafer bonds, which were processed at different bonding and annealing temperatures.
Table 1 shows the interfacial resistivity after the bonding for different process parameters. The calculation of these values is based on the dark IV-curves. The resistivity for the bond processed at 20 °C was calculated in the mostly linear region of ±0.1 V (compare with Fig. 4). As standard parameters for the bonding process, a bonding temperature of 120 °C, an acceleration voltage of 620 V, an Ar beam exposure duration of 8 min were chosen. All variations from these parameters are listed in Table 1. The stated error takes into account uncertainties for the measurement of the metal contact, substrate and bond interface resistance.
Electrical interface resistances of GaSb/Si bonds, depending on the process parameters.
Bond resistances (mΩ cm2) after 1 min post-annealing at
Despite the high bond strength, non-linear Schottky diode like IV-characteristics were found for all samples bonded at 20 °C room temperature. This can be correlated with the formation of a potential barrier exceeding ∼3kBT prevalent at the boundary interface . According to Bengtsson et al., such a barrier in the conduction band may be formed by acceptor-like defects  and explained by the amorphous layer created during the FAB activation process. Carriers are trapped in the defect states, hindering them to contribute to the electrical conductivity. Differences in the electron affinities of GaSb and Si do not explain the diode like IV behavior as both material have similar electron affinities (4.06 eV for GaSb and 4.05 eV for Si ). It can be assumed, that with a higher bonding temperature, the defect density is reduced and in fact, we have observed an Ohmic IV-characteristics for the GaSb/Si bonds at 120 °C (compare Fig. 4) with interface resistivities of <5 mΩ cm2, as it is listed in Table 1. This low resistivity speaks for a high amount of covalent bonds at the interface .
Besides the influence of the temperature during bonding and deoxidation, also the temperature of the annealing step after bonding has a strong effect on the conductivity. In Fig. 4, it is shown, that the bond resistance decreases, when the samples are annealed for 1 min at 350 °C instead of 300 °C. It has been already observed for direct bonds between GaAs and Si, that the amorphous layer at the interface is partly recrystallized during annealing, resulting in lower resistances . Surprisingly, all samples annealed at 400 °C show a higher bond resistance (compare Table 1). A negative effect of such a high temperature annealing could be induced thermal damage, resulting from the difference in the thermal expansion coefficient (7.8 × 10−6 K−1 for GaSb and 2.6 × 10−6 K−1 for Si ). When an annealing temperature of 400 °C is used, the negative effect of a higher annealing temperature seems to outbalance its positive influence.
It was found, that with the use of an acceleration voltage of 500 V the bond resistivity increases compared to the use of voltage of 620 V. It is possible, that the oxide on the GaSb surface was not completely removed within 8 min sputter treatment with an acceleration voltage of 500 V. Zhou et al. showed that remaining oxides at the bond interface hinders the carrier transport, thus increasing the surface resistivity . The bonding at an elevated Ar acceleration voltage of 740 V was not successful, due to the roughening of the surface described above.
4. Summary and conclusion
In conclusion, we have demonstrated direct GaSb/Si wafer bonding after argon-beam surface activation. It was shown that surface roughening of the polished GaSb wafers can be avoided with a sputter-etching treatment of up to 10 min. Fully bonded GaSb/Si wafer pairs with high bond energies were found, which are comparable to the bulk energies in the GaSb crystal. With the optimized process parameters of a bonding temperature of 120 °C, an Ar acceleration voltage of 620 V and a 1 min annealing step at 350 °C, it was possible to achieve a reproducible Ohmic carrier transport over the bond interface with resistivities <5 mΩ cm2 independently of the GaSb sputtering duration of 4, 8 and 12 min. Together with the high bond energies, these low resistivities speak for a high amount of covalent bonds at the interface. With these characteristics of the bond, the presented wafer bonding process is suitable for various advanced applications in high speed electronics or long-wavelength optical devices like for multi-junction solar cells.
The authors would like to thank M. Grave, A. Dilger, I. Semke, K. Mayer and R. Koch for help in the processing of the samples in this publication. The help of H. Nahme from the Fraunhofer Institute for High-Speed Dynamics is acknowledged for SAM measurements. F. Predan gratefully acknowledges the funding of his Ph.D. by the German Federal Environmental Foundation (contract 20014/344). This work was partly funded by the German BMWi through the project HekMod4 (contract 0325750).