Issue



Processing considerations for CMP on thin-film head wafers


09/01/2004







As critical dimensions (CDs) in thin-film magnetic read/write heads have become smaller, chemical mechanical planarization (CMP) processes have become more critical to overall manufacturing-process performance. In this article, details of various CMP processes used in fabrication of recording heads are presented and discussed.

A hard disk drive contains the following major components: a thin-film magnetic recording read/write head (TFH), a rotating disk with thin-film magnetic media, a spindle motor to drive the disk, an electromagnetic voice-coil rotary actuator with a gimbal suspension to move the slider across the disk surface, and electronics. The TFH consists of an inductive electromagnetic coil writer, a giant magnetoresistive (GMR) reader, and a slider body with an air-bearing surface, which flies over the magnetic disk to perform the read and write functions.

The TFH transducers are produced using a thin-film wafer-processing technology. TFH wafer processing is similar to that used in the fabrication of semiconductor devices, involving deposition, photolithography, etch, electroplating, and CMP.

AlTiC is an advanced ceramic material (64% Al2O3-36%TiC) used for the TFH wafer substrate for its mechanical, electrical, and thermal characteristics. Figure 1 shows a cross-sectional view of a TFH from the following wafer-processing steps: AlTiC substrate; undercoat (Al2O3) deposition and polishing (UC CMP); first shield (NiFe) pattern, deposition/plating, fill, and CMP (S1 CMP); first gap (Al2O3) deposition, GMR film deposition, GMR track-width definition/hard bias/leads, GMR stripe-height definition; second gap (Al2O3) deposition; second shield (NiFe) pattern and plating, separating gap deposition, first write pole (NiFe) pattern, plating, fill, and CMP (P1 CMP); first write pole extension P1P (NiFe, CoNiFe, or CoFe) and first layer write copper coil pattern, plating, fill, and CMP (P1P CMP); write coil insulation and writer gap deposition, second write pole (CoNiFe, CoFe) pattern, plating, pattern of second pole width into first pole by ion milling, fill, and CMP (P2 CMP); second layer write coil and P3 (NiFe) yoke plating, copper connections, overcoat (Al2O3) deposition and polishing (OC CMP); contact pad gold plating; and final wafer test.

Each processed wafer (~20,000 heads) is then sliced by a diamond wheel slicing machine to individual bars (each bar contains 40–50 heads). Advanced lapping to the specified sensor target height with a closed-loop control finishes the head surfaces. After sensor lapping, the air-bearing surface (ABS) is formed by photolithography and ion milling, followed by a thin diamond-like carbon (DLC) film deposited by ion-beam processing to achieve good slider tribological performance. Finally, the bars are diced to separate the heads.


Figure 1. Cross-sectional view of a TFH design with UC CMP, S1 CMP, P1 CMP, P1P CMP, P2 CMP, and OC CMP.
Click here to enlarge image

In the inductive writer, the electromagnetic coil induces magnetic flux in the loops, and the magnetic field between two pole tips (write gap) writes information on a disk. The write gap determines the linear bit density. In the GMR reader, the magnetic field from the disk changes the resistance of the GMR sensor and indicates the transition information. The sensor gets the data back by seeing the vertical magnetic-field transition from the disk. The bottom shield S1 and the top shield S2 prevent the GMR sensor from responding to fields just before and immediately after the transition to improve the linear bit resolution. The distance between two shields is referred to as the read gap, which determines the linear density of reading. The GMR sensor effect is due to scattering between two magnetic layers — a free layer and a pinned layer. The free layer is a soft magnet and magnetization is free to rotate. The pinned layer is fixed by exchange coupling to antiferromagnet and magnetization, which keeps it stationary. The resistance of the GMR head changes depending upon the angle between the magnetization of two layers, due to the effect of magnetic field from the disk on the free layer. The antiferromagnetic exchange layer provides the pinning field to the pinned layer.

A typical GMR sensor stack has a total thickness of ~40 nm, comprising multilayers of thin films of antiferromagnetic layer (PtMn)/synthetic antiferromagnet pinned layer (CoFe/Ru/CoFe)/spacer (Cu)/magnetically free layer (CoFe/NiFe)/Ta cap. The permanent magnet (PM) (i.e., CoCrPt) hard material is used for longitudinal stabilization of the sensor. The left side in Fig. 1 is the surface of the head facing the disk as the head flies.

Magnetic read/write head design and fabrication technology are following the same trends as semiconductors. For the last several years, annual performance enhancement of read/write heads for magnetic data storage (areal density gigabits/in.2 = linear density bits/in. × track density tracks/in.) has doubled each year. With shrinking device dimensions and new magnetic materials, CDs in the read/write heads have actually become smaller than those in semiconductors.

CMP challenges

CMP is used to remove unwanted topographical features to achieve a smooth and planar surface, as well as to reach film target thickness. Thickness control from CMP is needed for device performance (reader-writer separation distance, overwrite from top pole, etc.). Surface smoothing is needed for better control of building subsequent ultrathin dielectric, electric, and magnetic layers The reader gap determines the data-storage bit density (areal density = bit density × track density).

Planarity is needed for better control of photolithography for achieving subsequent layers with CD structure, such as sensor width and top pole width, which determine the data-storage track density (areal density). The overall photoresist thickness from spinning , CD from stepper exposure, and depth-of-focus (DOF) are more controllable on a planarized surface, as a result of less total height variation within the stepper field. This is more important for short-wavelength optical lithography tools, which give better CDs but less DOF. Planarity also ensures proper step coverage in TFH wafer fabrication. Therefore, CMP enables building subsequent layers on flat and smooth surfaces, with target thickness and much better control, reliability, and yield.

The wafer substrate for TFH fabrication is AlTiC (Al2O3-TiC). Sputter-deposited Al2O3 is used for dielectric/insulating materials to match to the substrate material properties, especially thermal expansion. NiFe, CoNiFe, and CoFe are used as the magnetic shield and magnetic conducting pole materials. Copper is used for the electromagnetic coil in the writer. Copper and NiFe are applied for the reader and writer electrical conductor pads. High aspect-ratio copper damascene coil process/copper damascene CMP has been developed for advanced writers to reduce the coil pitch and yoke for a high-frequency writer to reach high data rates.

CMP input parameters to be considered include polishing down-force, table and carrier speed, backside pressure (BSP), removal rate and polishing time, slurry, pad, conditioning, and cleaning. The response outputs are planarity, uniformity, surface roughness, film target thickness, overpolishing and underpolishing, and defects and contamination.

CMP processes in TFH wafer fabrication

UC CMP is used for smoothing the Al2O3 undercoat surface on the substrate. Sputter-depositing ~3µm of Al2O3 on a wafer substrate followed by CMP provides electrical isolation from the substrate. Since the bottom shield S1 (NiFe) is plated on the undercoat, any surface defects and roughness in the undercoat will affect the magnetic properties of the bottom shield; therefore, the Al2O3 undercoat has to be smooth and defect-free. Undercoat Al2O3 CMP is similar to semiconductor oxide CMP with SiO2 slurry at pH 10 (KOH). Subnanometer surface roughness (Ra) of the Al2O3 surface can be easily achieved. A GMR sensor stack is then built on the bottom shield.

The process is as follows: seed deposition, pattern, plating, strip, pattern, field etching, fill, and CMP. The bottom shield is the largest magnetic component of the head. The magnetic domain and interaction with the reader (material, stress, anisotropy, shape, thickness, annealing, etc.) must be understood and well controlled. Since the subsequent Al2O3 gap layer and sensor stack layers are very thin, any defects and roughness in the bottom shield will be reflected on gap and sensor. Thus, the bottom shield has to be polished to a smooth surface for stack build and has to be planar for reader track-definition photolithography control.

Bottom shield CMP (Al2O3-NiFe) is a two-step process: the first step, oxide CMP, quickly removes Al2O3, and then the second step, metal CMP, polishes the NiFe surface. The first step to remove Al2O3 is similar to typical semiconductor oxide CMP (SiO2 slurry, pH 10, KOH or NH4OH); however, NiFe protrusion is a challenge due to the high selectivity of Al2O3 to NiFe. The second step of metal CMP (Al2O3 slurry, pH 3–4, BTA, H2O2) smoothes the NiFe surface and reduces NiFe protrusion due to high metal selectivity. Proper balance between chemical activity and mechanical action is very important to achieve the optimum final surface finish and planarity in NiFe CMP. For example, the same slurry with a low polishing pressure resulted in NiFe surface roughness of Ra = 4nm (with deep corrosion pits), but Ra was improved to 0.4nm (5×5µm AFM scan) by simply increasing polishing pressure (the mechanical action). Also, with increasing down-force, the NiFe to Al2O3 changed from recession to protrusion. After bottom shield S1 CMP, the Ra in NiFe is <0.2nm under our optimum process (the Ra of incoming plated NiFe is ~3nm). The topography (step height variation) is >3µm after 3µm Al2O3 sputter deposition in the bottom shield. After CMP, the step height is <0.05µm (NiFe protrusion to Al2O3).

The lower-pole CMP (P1 CMP), lower-pole extension CMP (P1P CMP), and top-pole CMP (P2 CMP) can be similar to bottom shield S1 CMP. The similar two-step process can be used if the pole material is NiFe. The new one-step slurry has achieved a much lower step height (low NiFe protrusion to Al2O3). This slurry is specifically effective for P1P CMP, which involves polishing NiFe, Al2O3, copper, and hard-bake photoresist at the same time. Figure 2 gives this CMP slurry system and the removal rates of NiFe, Al2O3, and copper on the oxidant concentration.


Figure 2. CMP slurry system for TFH wafer polishing and removal rate of NiFe, Cu, and Al2O3 on the APS concentration.
Click here to enlarge image

Ammonium persulfate (APS, (NH4)2S2O8) is key to achieving low NiFe protrusion. (S2O82-) from APS is an oxidizer for NiFe removal rate control, and (2NH4+) from APS is a complexing agent for copper removal rate control. The alumina was removed much faster than the NiFe without APS. When the oxidant concentration was increased, the alumina removal rate was unchanged, but the NiFe removal rate reached a peak and then declined at higher concentration, and copper removal rate was proportional to the (NH4+) ammonia concentration. By varying abrasive and oxidant concentrations, it was possible to adjust the polishing rate of NiFe and alumina to obtain a planar surface of NiFe to Al2O3 in a single-step P1 and P2 CMP, as well as a planar surface of NiFe, copper coil, and hard-bake photoresist to Al2O3 in a single-step P1P CMP.

The thickness control in the multiple layers of CMP is also critical and has to be well controlled for device performances (i.e., reader-writer separation distance from P1 to P1P, overwrite from P2). The topography on the wafer before CMP in lower pole and lower-pole extension is ~5µm (step height), and the step height after CMP is <0.02µm. The thickness variation sigma after CMP are WIW 60nm and WTW 40nm for a target thickness of 2µm (EE 5mm).

High moment material CoNiFe and CoFe have been replacing NiFe as pole material (P1P and P2) to increase the head writing ability because CoNiFe and CoFe are more chemically active. The chemical attack during CMP is stronger in CoNiFe and CoFe than in NiFe. One way to solve this is to increase the pH for the second-step slurry (metal polishing) to near neutral (Al2O3 slurry, pH 6–7, H2O2) to prevent corrosion if CoNiFe is used for the pole material. The other way is by properly adjusting the mechanical action, such as polishing pressure and speed, to balance the chemical action for low-pH slurry (Al2O3 slurry, pH 3–4, H2O2) to reach good surface roughness and planarity. The H2O2 concentration is critical to reach a planar surface and avoid CoFe corrosion. The alumina is removed much faster than the CoFe without H2O2. When the H2O2 concentration was increased, the alumina removal rate was unchanged, but the CoFe removal rate rose to a peak and then declined, and finally stabilized at a higher concentration. By varying the H2O2concentration, we can adjust the polishing rate of CoFe and alumina to obtain a corrosion-free planar CoFe surface (to Al2O3) in a single-step CoFe P1P and P2 CMP.

The CMP process to build copper damascene coils for an advanced writer is similar to a semiconductor copper damascene process. Reactive ion etching is used to etch the hard-bake photoresist coil trench with SiO2 hard mask, and the Ta barrier and Cu are deposited and plated, and finally Cu damascene CMP is performed. The slurry (SiO2 + APS + BTA) we developed can remove Cu, Ta, SiO2, hard-bake photoresist, NiFe, and Al2O3 at the same time, which is required but challenging for coil damascene CMP. We also can use a standard Cu CMP two-step process, in which the first step polishes to remove bulk copper using barrier as a selective stop, and the second step polishes to gently remove barrier. Typically, Al2O3 slurry (pH 6, H2O2, BTA) is used for the first step, and SiO2 slurry (pH 10) for the second step.

Recently, abrasive-free reactive liquid (pH 3, H2O2, malic acid, BTA) has been developed for Cu CMP. CMP results from this liquid are very promising. With traditional abrasive-containing slurry (Al2O3 slurry, pH 6, H2O2, BTA), in a 0.5µm/0.5µm high-density line/space (L/S) area, the dishing can be near zero, but erosion is around 80nm, while in the large features (80×100µm contact pad), the dishing is 150nm. Erosion and dishing from traditional slurry is caused by rolling (three-body abrasion) abrasives. These occurrences can be minimized but cannot be eliminated with traditional abrasive slurry. For abrasive-free reactive liquid slurry in a 0.5µm/0.5µm high-density L/S area, the erosion could be near zero with dishing of 10nm, while in the large feature regime (80×100µm), dishing is 20nm.

After the second step to remove the TaN barrier layer, outstanding planarity (dishing and erosion) is achieved. The reduction in erosion in high-density areas and small dishing in large features are the direct result of the abrasive-free reactive liquid that eliminates three-body abrasion. Mechanical action to remove the reaction layers is only from the pad rather than from the abrasives as in traditional three-body abrasion-type polishing. Fixed abrasive (two-body abrasion and self-stopping) also may be a good technology to improve CMP planarity, resulting in less copper dishing and oxide erosion.

After Al2O3 overcoat deposition, a CMP process (OC CMP) is performed to remove Al2O3, open up the Cu studs, and bring overcoat thickness to the final specification. Oxide CMP (SiO2, pH 10) is used to remove Al2O3 and then Cu CMP (Al2O3, pH 6, H2O2) is used to polish a Cu stud: ~14µm Al2O3 is removed. After OC CMP, gold bond pads are plated; then the wafer is tested and shipped out.

CMP parameter optimization and integration

A CMP process generally goes through three processing stages — surface topography planarization, bulk removal to the oxide-metal interface, and a breakthrough to reach final target thickness. Adequate Al2O3 fill thickness on NiFe is needed to remove surface topography and to achieve a planar surface prior to reaching the oxide-metal interface breakthrough. The better the planarization efficiency, the less fill thickness needed. The Al2O3 fill thickness should be at least ~0.8µm (based on planarization efficiency) higher than NiFe with APS SiO2 slurry. Plated NiFe thickness should be ~0.8µm (based on plating, fill, and CMP WIW and WTW uniformity) higher than post-CMP target thickness to make sure all the Al2O3 on NiFe features is completely removed.

CMP table speed, down-force, carrier speed, and BSP mainly influence the material removal rate (MRR). The MRR significantly increases with an increase of table speed and down-force, but only slightly increases with the increasing of carrier speed and BSP. The MRRs from a hard pad and a soft pad are very close; however, the planarity length from a hard pad is much better than from a soft pad. The MRR with oxide CMP slurry (SiO2 slurry at pH 10) in processing is ~0.6µm/min for an Al2O3 film wafer and ~0.5µm/min for a SiO2 film wafer. The post-CMP roughness is Ra = 0.4–0.5nm on Al2O3 film and 0.3–0.4nm on SiO2 film (AFM 10×10µm).

The factors that influence post-CMP nonuniformity are much more complicated than that of MRR. Post-CMP within-wafer nonuniformity (WIWNU) depends on many factors, such as incoming-wafer film uniformity, down-force, wafer curvature, BSP, wafer-to-retaining-ring protrusion, retaining ring pressure, pad, conditioning, table and carrier speed, slurry distribution, etc. The oscillation of the carrier in a rotary polisher also helps to improve polishing uniformity. Polishing nonuniformity from machine and carrier setup cannot be solved through process changes. The spindle run-out, table run-out, and table sweep should be controlled. The carrier rebuild (gimbal, insert film, ring pressure, BSP, and down-force) should be carefully conducted.


Figure 3. A profile scanning across a wafer a) before and b) after CMP.
Click here to enlarge image

Bow (convex) is the typical global geometry of wafer deformation due to the wafer substrate bow and film stress. The compressive stress from deposition processing causes convex bending. Figures 3a and 3b show the profile scanning across a wafer before and after CMP. More than 6–8µm total Al2O3 film was deposited on this wafer before this step, then 5µm more was deposited. After CMP, 5µm step height and 2µm field film were removed.


The bow shape of the global wafer geometry requires a conformable pad for polishing uniformity. Harder pads can incur less recession) within a chip but poor polishing uniformity within a wafer. Softer pads produce better polishing uniformity across the wafer, but poor planarity (more recession). A careful balance of softer and harder pads must be maintained. The common solution is to use a composite stack pad (a harder pad backed by a softer pad — the harder rigid top pad for planarity and softer back pad for uniformity) to achieve less recession and better uniformity.

Higher polishing down-force (and lower polishing speed) helps to improve polishing uniformity, but lower polishing down-force and higher polishing speed can reduce dishing and erosion. Table and carrier speed must be tuned in the process recipe to reach relative polishing-speed uniformity across the whole wafer. The theoretical calculation in the rotary polisher showed that the relative speed difference between the wafer edge and the center is zero when the table speed and the carrier speed are equal, and the relative polishing speed of the edge is always higher than that of the wafer center if the table speed and the carrier speed are not matched. A DOE showed the significant effect of the interaction between table speed and carrier speed. The ratio of carrier speed:table speed has more effect than carrier speed or table speed individually on uniformity.

The effect from BSP on post-CMP uniformity is much more significant than polishing down-force and table and carrier speed. The uniformity can be dominated by BSP control. Based on the incoming wafer and process maps, the BSP in the process recipe can be adjusted to bend the wafer by positive, vacuum, or radical zone BSP, and optimized to obtain polishing uniformity. BSP can also compensate for film center-to-edge thickness or incoming film thickness. BSP can push the back of the wafer and accelerate the center polishing rate for center-thick edge-thin film or center-slow edge-fast processes. It also can vacuum the back of the wafer and decrease the center polishing rate for the center-fast edge-slow process.

Patterned density also plays an important role in WIWNU. Because of the differences of polishing-pressure distribution from patterned density, the oxide or metal in large features is polished slower than in small features (resulting in residue in large features or erosion in small features), and the oxide or metal in wider trenches is polished faster than in narrow trenches (resulting in dishing in wider trenches). Dummy features in the dicing street are an effective method to improve within-wafer pressure distribution and achieve polishing uniformity.

In situ endpoint detection can minimize overpolish and thus enable better control of planarity and improve wafer-to-wafer nonuniformity (WTWNU). Motor current from spindle and table has good performance for NiFe/Al2O3 CMP. Eddy current-optical endpoint is one of the best performance areas for Cu CMP. It also provides real-time feedback for uniformity control (edge-fast or center-fast), a very good function for process development.

Advanced process control (APC) run-to-run closed-loop control helps improve WIWNU. The predicted polishing rate and optimized BSP are estimated based on historical run-to-run thickness and center-to-edge uniformity (CTE) data. Polishing times and BSP settings for every wafer are calculated based on the predicted polishing rate and optimized BSP, as well as various feedforward data. APC with in situ NOVA metrology can speed up feedback control. With APC, our CMP rework (WTWNU) reduced 80% and WIWNU improved 40%.

Summary

The wafer fabrication process for TFH is similar to that used to build semiconductor ICs, including photolithography, deposition, electroplating, etching, and CMP. The areal density (gigabits/in.2) of hard disk drives has increased by about 100%/year for the last several years. These areal density increases have been achieved by actively shrinking the device dimensions, as well as developing new magnetic materials and designs. As dimensions decrease, CMP assumes a more critical role in wafer processing.

References

  1. M. Jiang, L. Stearns, "Application of Chemical Mechanical Polishing (CMP) in Thin Film Magnetic Head Wafer Fabrication," CMP short course, VMIC, 2002.
  2. M. Jiang, S. Hao, R. Komanduri, "On the Advanced Lapping in the Precision Finishing of Thin Film Magnetic Recording Head for Rigid Disc," Applied Physics A: Materials Science and Processing, 76, Dec. 2002.
  3. P.B. Phipps, F.H. Dill, F.O. Eschbach, E. Lee, F. Martin, "The Use of CMP in Making Read/Write Heads," Proc. Electrochemical Society, Vol. 29, 2000.

Ming Jiang is a technical leader at Hitachi, 5600 Cottle Rd., San Jose, CA 95193; ph 408/717-6220, e-mail [email protected].