In this experiment, three types of polishing pad (pads A, B and C) with different designs were selected. CMP was performed separately with the three different polishing pads, for which the solution dilution ratio and polishing time were kept at 1:10 and 45 min, respectively. The root mean square (RMS) roughness and MRR of the GaSb wafers are shown in Figure 1. The MRR values for pads A, B and C are 0.236, 0.216 and 0.286 μm/min, respectively. The mean RMS roughness values of pads A, B and C are 0.635, 0.796 and 1.043 nm, respectively.

It is well known that the polishing pad has a significant effect on the surface quality of wafers processed by CMP (Li et al., 1995). The surface layers of the three polishing pads are made of the same type of polyurethane with no grooving, but have different microporous structures.

The SEM images of the surfaces and cross-sections of the three polishing pads are shown in Figure 2. Pads A and B consist of double layers, whereas pad C has a triple-layer structure. The pore size, pore density, velvet length, thickness, compressibility and hardness of the polishing pads are listed in Table 2. Comparison of different pads under the same polishing conditions shows that the MRR of the GaSb wafers is a function of the pad surface characteristics. During polishing, the pad is under shear stress and the mechanical properties typically drop significantly with increasing pore size. The pores on the pad surface aid in transport of the polishing solution to all parts of the wafer, thereby ensuring chemical erosion and abrasive transport. When the size of the pores is larger, the solution transport capacity is also stronger. Furthermore, a low modulus of elasticity can be inferred from low compressibility of the polishing pad, leading to a reduction in the MRR. Therefore, pad C with a pore size of 45–83 µm and compressibility of 18% was found to have a higher MRR.

Conversely, the harder the polishing pad, the more difficult it is for the GaSb wafers to run in with the polishing pad. Thus, non-uniform surface removal caused by non-uniform friction results in lower surface quality and the RMS roughness values for multi-point tests are more widely distributed (Lee et al., 2013). Although the MRR value for pad B is low, owing to the high hardness, the roughness is greater than for pad A and the RMS roughness values measured have a narrower distribution. Therefore, pad A is judged to be a suitable pad for polishing of GaSb wafers.

The subsequent experiments of the polishing process were carried out on pad A with a polishing time of 45 min. The RMS roughness of the GaSb wafers varies with the dilution ratio of the polishing solution, as depicted in Figure 3. When the dilution ratio is 1:3, the mean RMS roughness value is 0.855 nm (based on a large scan area of 63 × 48 μm) and the MRR value is 0.597 μm/min. When the dilution ratio is 1:20, the mean RMS roughness is 0.4244 nm (based on a large scan area of 63 × 48 μm) and the MRR is 0.216 μm/min. It has been reported that the RMS roughness of wafers is reduced from 0.722–0.966 to 0.345–0.465 nm and can be considered a suitable range (Allen et al., 2011; Allen et al., 2012b).Figure 4 demonstrates the surface morphology (63 × 48 μm) of GaSb wafers polished using different dilution ratios for the polishing solution. It is evident that the morphology of the GaSb wafer surface is improved by increasing the dilution ratio of the polishing solution. Different degrees of deep scratches are apparent on wafers subjected to polishing with dilution ratios of 1:3, 1:7 and 1:10. For a dilution ratio of 1:20, the degree of deep scratches reduced, but small amounts of shallow scratches could still be observed via AFM, as shown in Figure 5.

The polishing solution mainly contains abrasive particles and oxidant. SDIC was used as the oxidant in this experiment. The oxidant can be hydrolyzed in aqueous solution via the following hydrolytic reaction: C 3 C l 2 N 3 Na O 3   +   2 H 2 O → ( C 3 H 2 N 2 N O 3 ) −   +   2 HC l O   +   N a + (2)

Under the condition of the aqueous solution containing ClO⁻, the oxidation reaction of GaSb can be represented by the following processes (Bermudez, 2013): CI O − → C I − + ( O ) (3) 2 GaSb + 6 ( O ) → G a 2 O 3 + S b 2 O 3 (4) S b 2 O 3 + 2 ( O ) → S b 2 O 5 (5) 2 GaSb + S b 2 O 3 → G a 2 O 3 + Sb (6)

For a given applied stress and polishing pad, the MRR is primarily a function of the nature of the polishing solution. With increasing dilution ratio of the polishing solution, the surface oxidation rate and abrasive concentration decrease, which lead to a reduction in the chemical and mechanical action. At a dilution ratio of 1:20, the mechanical and chemical action becomes more appropriate and results in gradual optimization of the RMS roughness of the GaSb wafer and decreasing MRR. The number of abrasive particles deposited on the surface of the polishing pad is another factor that controls the removal rate. Excess abrasives lead to a large MRR for the GaSb wafer surface, and settling of excess abrasives on the surface of the polishing pad may scratch the wafer surface (Xie et al., 1996). With increasing dilution ratio of the polishing solution, the MRR decreases from 0.597 to 0.216 μm/min and scratch levels on the GaSb wafer surface are weakened and reduce in number.

It is noteworthy that the polishing time also significantly affects the surface quality of GaSb wafers. As shown in Figure 6A, the wafer RMS roughness decreases sharply to 0.41 nm (63 × 48 μm) as the polishing time increases from 0 to 40 min, and then remains nearly constant. Similarly, the MRR value decreases obviously from 1.2 to 0.159 μm/min and is then maintained at 0.16 μm/min from 40 to 150 min. Figure 6B shows the wafer PV value measured by laser interferometry, which decreases from 6.77 to 5.06 µm as the polishing time increases from 0 to 50 min. Subsequently, with increasing polishing time the wafer PV deteriorates obviously and increases up to 8.43 µm with a polishing time of 150 min. It can be concluded that 50 min is the optimum polishing time for planarization of GaSb wafers. Figure 7 demonstrates surface WLI images (63 × 48 μm) of a polished GaSb wafer surface for different polishing times. The grinding grain on the GaSb wafer surface gradually flattens out and disappears. For a polishing time of 50 min, the surface of the wafers shows low RMS roughness and no obvious deep scratches. The wafer surface then remains stable over time.

Figure 6C shows that the full-width half-maximum (FWHM) value for the Bragg diffraction peak of GaSb exhibits a decreasing trend, which is consistent with the RMS roughness and MRR. The FWHM value decreases from 28” to 14.3” with a polishing time of 40 min and is then maintained at this value. FWHM values can be used to characterize damage on the GaSb surface. In the early stage of polishing, initial crystalline damage to the GaSb surface is serious and surfaces are more susceptible to oxidation. Before a polishing time of 40 min, surface crystalline damage on the GaSb is easy to remove via polishing, resulting in high RMS roughness and fast MRR. After a polishing time of 40 min, surface damage on the GaSb is completely removed under certain polishing solution and pad conditions, and the RMS roughness and MRR of the GaSb wafers both reach a stable level. Figure 8 demonstrates surface WLI images (63 × 48 μm) of GaSb wafers polished using different polishing times. The GaSb wafers surface of the grinding grain disappears and flatten out gradually. For a polishing time of 50 min, the wafer surface shows low RMS roughness and no obvious deep scratches. The surface of the wafer then remains stable over time. This indicates that 50 min is the optimum polishing time for planarization of GaSb wafers.

Since the SDIC oxidant used in the polishing solution reaches a hydrolytic equilibrium, the HClO concentration in the aqueous solution is dynamic. The chemical reaction is as follows. With the consumption of ClO⁻ in the polishing process, the balance constantly shifts to the right and H⁺ is slowly released. The pH environment of the polishing solution then changes, leading to a non-ideal polishing effect. In this regard, NaHCO₃ can serve as a pH buffer to stabilize the pH environment of the polishing solution by reacting with H⁺ and to a certain extent optimizes the stability of the chemical oxidation action. HClO → H + + Cl O − (7) NaHC O 3 → N a + + HC O 3 − (8) H + + HC O 3 − → H 2 O + C O 2 (9)

The polishing process was conducted with a polishing solution (its dilution ratio is 1:20), a 2 wt% NaHCO₃ additive and a polishing time of 50 min. Figure 7A presents the AFM image of the GaSb wafers, showing a well-controlled surface without scratches. The RMS roughness of the polished wafers can be controlled to less than 0.13 nm, indicating a better quality than that of state-of-the-art GaSb wafers available on the market (Allen et al., 2011; Allen et al., 2012b). To further evaluate the crystalline quality of the GaSb wafers, a mapping image was taken using the XRT (Wang and Matyi, 1992), which has a high resolution for crystalline defects and strain resulting from the CMP process, as shown in Figure 7B. The polished wafers show a uniform surface without obvious damage, defects or strain, indicating their high surface quality.

To further verify the quality of the polished GaSb wafers, an “epi-ready” GaSb substrate was prepared for buffer layer growth via MBE. The sample was degassed at 300 C for 30 min in a pretreatment chamber and then deoxidized at 540 C for 10 min in the reactor before growth. A buffer layer with a thickness of 1 μm was grown at 520 C. The GaSb growth rate was established as 0.5 monolayer/s using reflective high-energy electron diffraction. The Ga and Sb fluxes were kept at 3.0 × 10^–7 and 2.4 × 10^–6 Torr, respectively, during growth of the buffer layer, which were directly measured as the beam equivalent pressure using an ion gauge under the substrate. Figures 9A,B show an AFM image and the HRXRD rocking curve of the sample, respectively. In Figure 9A, an atomically smooth surface with RMS roughness of 0.159 nm can be observed over a 10 × 10 μm scan area, revealing a well-controlled two-dimensional growth mode. In Figure 9B, the FWHM of the Bragg diffraction peak of GaSb is 14.2”, indicating very high crystalline quality of the GaSb wafers. Hence, the optimized CMP process for GaSb can meet the requirements for MBE.