1. Introduction
Group III-V nitride based compounds have a wide range of potential applications including UV light emitting diodes,laser diodes,high power microwave devices and switches,and high power and high frequency microelectronic devices[1]. Recently,there has been considerable interest in the growth of these nitrides along non-polar and semi-polar directions. Materials grown along these directions eliminate the problem of the polarization induced reduction in the internal quantum efficiency and photoluminescence red shift due to quantum confined stark effect (QCSE)[2, 3, 4, 5]. AlN,in particular,has attracted a lot of attention for III-V nitride epitaxy application due to its unique properties including high thermal conductivity,low thermal expansion compared to GaN,excellent mechanical properties,electrical insulation and widest band gap (6.2 eV at room temperature)[6].
AlN with semi polar (11¯22) orientation is commonly grown hetero-epitaxially on m-plane sapphire by the MOCVD method. In this growth process a high density of micro structural defects and a rough surface are generated due to the difference in thermal expansion,lattice and stacking mismatch between sapphire and AlN[7]. The poor surface morphology and defective microstructure of the AlN surface degrade the fabricated device performance. It is essential,therefore,to produce a high optical quality and defect free AlN surface to realize the full potential of this material in micro- and optoelectronics applications. In order to achieve a high quality epitaxial device layer,surface planarization is very important for the fabrication of an atomically flat,smooth and defect free AlN surface. Chemical mechanical planarization (CMP) is an established process to planarize semiconductor wafers for electronics application. CMP involves both physical and chemical forces to polish and planarize wafer surfaces and is able to remove both surface and subsurface damages,which is likely to occur during polishing[8].
Numerous reports are available on CMP of different planes of AlN including polar[1, 9, 10, 11] and non-polar[12] surfaces.
CMP of the Al face and N face of AlN surfaces was reported by Schowalter et al. using KOH based slurry containing colloidal silica suspension[13]. It is reported that for the Al faced AlN surfaces,the MRR increased from ∼ 1 μm/h to over 10 μm/h as the angle between the surface normal and the c axis increased from zero to over 20°. In another work by the same research group,the MRR was reported ∼ 0.1 μm and 60 μm for the Al- and N-face of the AlN surface,respectively,for 1 h of polishing[14]. Bobea et al. reported CMP of AlN with proprietary alkaline slurry of sub-micron sized abrasives and have confirmed the presence of surface damage by RSM techniques[15]. Till now,most of the reported AlN CMP processes report the use of abrasives nanoparticles in the CMP polishing slurry. Although use of abrasive nanoparticles results in higher MRR in the CMP process,it creates polishing related defects and scratches on the surface.
In this work a novel single step,abrasive-free CMP (AFCMP) process,using proprietary slurry chemistry and processing parameters,has been employed to polish semi-polar (11¯22) AlN epilayers grown on 2 inch diameter sapphire substrate. The advantage of this AFCMP process is that since abrasive particles are not employed in the CMP slurry,abrasive related defect generation during polishing can be eliminated. The effect of the three main CMP process parameters,such as polishing pressure,platen velocity,and slurry pH,on the material removal rate (MRR) and surface finish of semi-polar (11¯22) AlN surface have been reported using abrasive free KMnO4 slurry and the results have been compared with those obtained on semi-polar (11¯22) AlN and GaN surfaces. An atomically flat and smooth surface with RMS surface roughness ∼1.2 nm has been obtained on the AFCMP processed semi-polar (11¯22) AlN surface under optimized CMP process conditions.
2. Experimental
Semi-polar (11¯22) AlN wafers of 2 inch diameter grown on m-plane (1-100) sapphire substrate by the MOVPE method were used in this study. All the AFCMP experiments have been carried out on a Buehler Ecomet 250 variable speed Grinder Polisher with an Automet 250 power head. A Cabot D100 pad was used for all the AFCMP experiments. The AlN wafer was mounted on a 2” stainless steel carrier using Crystal Bond 509 clear epoxy resin. The wafer was characterized for crystal structure and surface quality before the AFCMP experiments. Solutions of KMnO4 with a fixed concentration of 0.4 M with varying pH without any abrasive particles have been used as CMP slurries in this investigation. The slurry was continuously delivered on the polishing pad by a peristaltic pump (Ravel,Model No. RH-P100VS-100) at a flow rate of 10 mL/min. The typical time of each polishing experiment was 15 min,with an intermittent conditioning of the pad by a diamond pad conditioner (Model no.S341038N-2A2,Abrasive Technology Asia) after each 5 min. In this work numbers of AFCMP experiments on semi-polar (11¯22) AlN surfaces have been carried out to study the effect of slurry pH,platen velocity and polishing pressure on MRR. AFCMP experiments were carried out at varying polishing pressures (24 to 38 kPa),platen speeds (60-100 rpm),and slurry pH (1-3). Each experimental data point was generated from 3 experiments. The MRR for each experiment was estimated from a difference in weight of the AlN wafer,measured by a high precision electronic balance (Sartorius Model No. CPA225D),before and after AFCMP experiments. Post CMP cleaning,by repeated washing with alcohol,acetone and deionised water,was carried out to clean the wafer surface after each AFCMP experiment. The surface quality of as-grown and AFCMP processed semi-polar (11¯22) AlN wafers were characterized using the Bruker Optical Surface profiler (Model: Contour GT-K0).
3. Results and discussion
3.1 The influence of slurry pH on MRR
Figure 1 shows the effect of slurry pH on the MRR of semi-polar AlN surface using the AFCMP process. The pH value of the slurries was adjusted to 1-3 by using HNO3 or NaOH as pH regulator agents. From this figure it is observed that the MRR of AlN is very low and nearly unchanged upto pH 2 and then it increases drastically to a high MRR with an increase in slurry pH to 3. The maximum removal rate was found to be ∼ 562 nm/h at pH 3. An increase in MRR by 60 % has been observed on increasing the slurry pH from 1 to 3 for KMnO4 containing AFCMP slurries. From this result it is inferred that the chemical reaction between the (11¯22) AlN surface and the slurry is promoted by less acidic medium leading to passivation of the surface with subsequent removal of the surface modified layer by the mechanical action of the polishing. The possibility of the formation of Al(OH)3 on the AlN surface cannot be ruled out in the aqueous medium of the CMP experimental condition,which may be accelerated at an elevated temperature arising from the frictional heat of polishing. It is also anticipated that the conversion of AlN into Al(OH)3 should be favoured in a less acidic medium (pH-3) and therefore,an increasing MRR with increasing pH of the CMP slurry is expected. Any change in slurry pH should affect the kinetics of the surface reaction causing a change in MRR of the process[16].
At pH-3,both dissolution and oxidation are accelerated,leading to faster corrosion and hence a higher material removal from the AlN surface. In chemical mechanical polishing of an aluminum alloy film it has been observed that the material removal decreases beyond a certain pH (3-4)[17]. The reduction in MRR at a higher pH (pH > 3) has been attributed to the dominance of the surface oxidation rate over its dissolution rate. It is well known in CMP that when the surface oxidation/passivation rate is larger than the oxide/surface passivated layer dissolution rate in the polishing chemical the material removal becomes insensitive to the oxidizer concentration and decreases for slurry pH where surface oxidation/passivation is favoured. Once a passivated layer is formed the surface will not be exposed to the oxidizer and hence the removal rate decreases. If the diffusion of oxidizing species through the oxide layer becomes the rate limiting step of the material removal in the CMP process then a reduction in MRR at higher pH for AlN CMP is anticipated. The oxidant diffusion rate leading to passivation should be slow enough to sustain a sensible removal rate,and fast enough to prohibit severe scratching on the surfaces. From the knowledge of pH dependence on MRR in CMP and the reported data on aluminum film mentioned in Reference [17],the pHs of the slurries have been maintained up to 3 in the current investigation.
Figure 2 shows the effect of slurry pH on the surface quality of (11¯22) AlN surfaces. The surface morphology and topography of the polished wafer are much improved compared to the unpolished wafer. The RMS roughness of the polished surface has decreased by about 80%-85% from its original value. However,the surface roughness is very weakly dependent on slurry pH. It is almost similar for slurries with pH 1 and 2,but a marginal improvement has been observed for slurry with pH 3.
Figures 3(a) and 3(a′) show the 2D and 3D Optical surface profiler (OSP) images,respectively,of the (11¯22) AlN surfaces before polishing. Figures 3(b) and 3(b′), 3(c) and 3(c′) and 3(d) and 3(d′) show the OSP images (b,c,d are 2D images and b′,c′,and d′ are 3D images) of the AFCMP processed semi-polar (11¯22) AlN surfaces for slurry pHs 1,2 and 3,respectively,under the fixed experimental condition of a downward pressure of 38 kPa,platen velocity of 90 rpm,carrier velocity of 30 rpm,and KMnO4 concentration of 0.4 M. From these figures it can be seen that the surface quality is improved with increasing slurry pH. The RMS surface roughness decreased from 44.5 nm on as-grown semi-polar AlN surface to 3 nm (for slurry pH 3) on the AFCMP processed surface over a scan area of 0.70 × 0.96 mm2. The decrease in surface roughness or the improvement of the surface quality with increasing slurry pH could be attributed to enhanced oxidation of the rough surface with subsequent removal of the oxide layer either by polishing or dissolution in the polishing chemical.
3.2 The influence of platen velocity on MRR
Figure 4 shows the effect of platen velocities on the MRR of a semi-polar (11¯22) AlN surface using AFCMP slurry. The experiments were conducted under different platen velocities at other fixed experimental parameters. From Figure 4 it is observed that the MRR increases with increasing platen velocity and becomes almost double at 100 rpm compared to MRR obtained at 60 rpm. The increase in MRR with increasing platen velocity is consistent with the classic Preston's equation[18]. According to this equation,the MRR is proportional to the product of polishing pressure and relative velocity. When the downward pressure and other experimental parameters remain constant,the MRR is simply a function of platen velocity and thus should increase with platen velocity. A maximum MRR of 455 nm/h has been observed for the AlN surface at a platen velocity of 100 rpm and under other fixed experimental conditions of carrier velocity of 30 rpm,downward pressure of 38 kPa,pH of 2 and 0.4 M KMnO4 concentration.
As the platen velocity increases,the relative velocity between the AlN wafer surface and the polishing pad increases,improving the polishing efficiency of the process. Increasing the platen velocity seems to promote fresh slurry delivery on the pad and effective removal of the by-product from the pad,which enhances surface passivation followed by dissolution of the passivated oxide from the surface[19]. More cyclic mechanical action on the chemically soft layer could improve MRR at higher platen velocities. For a high removal rate,the rate of oxidation or passivation should match with the rate of material removal. In the present case the mechanical action could be in balance with the chemical action at a platen speed of 100 rpm,leading to the maximum MRR in the AFCMP process.

3.3 The influence of polishing pressure on MRR
Figure 6(a) shows the effect of polishing pressure on the MRR of a semi-polar (11¯22) AlN surface using AFCMP. From this figure it is observed that the MRR linearly increases with increasing polishing pressure. The increase in the MRR with increasing downward pressure is seen to follow Preston's equation[18]. According to this equation,when other chemical and mechanical parameters remain constant,a change in the polishing pressure is expected to change the mechanical action of polishing,causing a change in the material removal rate (MRR). Increasing the polishing pressure increases the contact area between the pad and the wafer surface,leading to increased frictional forces between them. The increased frictional force causes more wafer surface to get abraded with the polishing pad resulting in an increase in the MRR. A maximum MRR of 223 nm/h has been obtained for a semi-polar (11¯22) AlN surface at a downward pressure of 38 kPa,platen velocity 90 rpm,slurry pH of 2 and KMnO4 conc. 0.4 M. A much higher MRR (1850 nm/h) has been observed for (11¯22) GaN surfaces under similar experimental conditions. The work was reported separately in our previous paper. The root mean square (RMS) surface roughness of ∼0.8 nm,over a large scanning area of 0.70 × 0.96 mm2,has been achieved on AFCMP processed semi-polar (11¯22) GaN surfaces using optimized slurry chemistry and processing parameters[20]. These results suggest that semi-polar (11¯22) GaN surface and AlN surface could have different in-plane responses towards applied pressure or any mechanical force along the (11¯22) direction,leading to a very different MRR under similar experimental conditions. Also,the difference in the MRR could be due to a difference in the hardness of semi-polar (11¯22) AlN and AlGaN planes. To investigate the effect of Ga in AlN crystal structure,AFCMP experiments were conducted on semi-polar (11¯22) AlGaN surfaces. Figure 6(b) shows the effect of polishing pressure on the MRR of an AlGaN surface using the AFCMP process. From this figure it is observed that the MRR increases marginally up to 52 kPa and beyond that it shoots up abruptly with a further increase in pressure. The maximum MRR was found to be ∼727 nm/h at 62 kPa. It can be noted that the MRR obtained on semi-polar (11¯22) GaN,AlN and AlGaN surfaces is in the order,MRRGaN > MRRAlN > MRRAlGaN (Figure 7). The observed trend in MRR suggests the hardness (H) of the planes will be in the reverse order,i.e. HGaN < HAlN < HAlGaN. It is also interesting to note that even though the semi-polar (11¯22) GaN plane is much softer than that of the (11¯22) AlN plane,the (11¯22) plane of their alloy AlGaN is much harder than the (11¯22) AlN plane.

Figures 8(a) and 8(a′) show the 2D and 3D Optical surface profiler images,respectively,of as-grown semi-polar (11¯22) AlGaN plane and the RMS roughness was found to be ∼ 8.1 nm over a scan area of 0.70 × 0.96 mm2. Figures 9(a,a′),(b,b′) ,(c,c′) and (d,d′) show the (2D,3D) OSP images of semi-polar (11¯22) AlGaN surfaces after AFCMP at polishing pressures of 31,38,52 and 62 kPa,respectively,under the fixed experimental conditions of a slurry pH of 2,platen velocity of 90 rpm,carrier velocity of 30 rpm,and KMnO4 concentration of 0.4 M. From these figures it can be observed that a minimum RMS surface roughness of 0.7 nm was obtained on the semi-polar (11¯22) AlGaN surface at 38 kPa,whereas that on the (11¯22) AlN surface was ∼6 nm,as in Figure 3(c),under similar polishing conditions and measured over the same scan areas. These surface roughness data suggest that even though the hardness of the (11¯22) AlGaN plane is more than that of the (11¯22) AlN plane,its chemical reactivity in the typical polishing condition could be more than that of AlN leading to lower surface roughness at low polishing pressure (38 kPa). Evolution of surface pits and scratches at higher polishing pressures (52 and 62 kPa) leading to degradation of the surface roughness,as observed in Figures 9(c,c′) and 9(d,d′),supports the low pressure surface roughness data. Apart from these factors (slurry pH,platen velocity,polishing pressure),which have been studied in this investigation,there are several other processing variables (such as oxidizer type and concentration,carrier speed,abrasive type and their concentration,surfactant type and concentration,electrolyte type and concentration),which effect the CMP process as well. As the CMP process combines the mechanical and the chemical removal mechanisms in a synergistic effect,the balance between the chemical (such as slurry pH,slurry concentration,oxidizer,slurry flow rate) and mechanical parameters (such as downward pressure,platen velocity,carrier velocity,pad characteristics) is essential to obtain optimized CMP results. A better set of CMP parameters for desired and optimal CMP output are generally obtained by performing a design of an experiment (DOE) approach,say by the Taguchi method,from various experimental data points. Optimization of the CMP process variables for single crystal GaN polishing using DOE by the Taguchi method has been investigated recently and will be communicated separately. The parameters studied in this investigation have shown a significant effect on CMP output such as MRR and surface roughness,thus none of them can really be ignored.
4. Conclusion
In this work,AFCMP of semi-polar (11¯22) AlN surface has been demonstrated using abrasive free slurry containing KMnO4 as oxidizer. The influences of slurry pH,polishing pressure and platen velocity on MRR and the surface quality of semi-polar (11¯22) AlN have been investigated. The effect of polishing pressure on AFCMP of the (11¯22) AlN surface has been compared with that of the (11¯22) AlGaN surface. The maximum MRR has been found to be 562 nm/h for the semi-polar (11¯22) AlN surface under optimized conditions of 38 kPa,90 rpm platen velocity,30 rpm carrier velocity,slurry pH 3 and 0.4 M oxidizer concentration. The best RMS surface roughness of AFCMP processed semi-polar AlN wafer was found to be ∼1.2 nm over a large scanning area of 0.70 × 0.96 mm2 under polishing conditions of a slurry pH of 2,downward pressure of 38 kPa,platen velocity of 100 rpm,carrier velocity of 30 rpm,and 0.4 M KMnO4 solution. The same polishing parameters have resulted in an RMS surface roughness of ∼0.7 nm on the (11¯22) AlGaN surface over the same scan area. The (11¯22) AlGaN surface has been found to be harder mechanically but more reactive chemically compared to the (11¯22) AlN surface. A comparison of the MRR of semipolar AlN,GaN and AlGaN surfaces has been shown. These AFCMP processed defect free and atomically smooth (11¯22) AlN surfaces could be used for further epitaxial growth of AlN.
5. Acknowledgement
The authors thank Nirupam Hatui,A. Azizur Rahman and Arnab Bhattacharya,Tata Institute of Fundamental Research (TIFR),Mumbai for providing us with the MOVPE grown AlN and AlGaN epilayers for the AFCMP experiments and for discussions. Authors greatly acknowledge the financial support from the Department of Science and Technology (DST),Government of India (No,SR/S2/Cmp-0009/2011),and partial support from the Board of Research in Nuclear Sciences (BRNS),Department of Atomic Energy (DAE),Government of India (No.-34/14/43/2014-BRNS) with ATC,BRNS is gratefully acknowledged.