Fenton Reaction for Enhancing Polishing Rate and Protonated Amine Functional Group Polymer for Inhibiting Corrosion in Ge 1 Sb 4 Te 5 Film Surface Chemical-Mechanical-Planarization

: A Fenton reaction and a corrosion inhibition strategy were designed for enhancing the polishing rate and achieving a corrosion-free Ge 1 Sb 4 Te 5 ﬁlm surface during chemical-mechanical planarization (CMP) of three-dimensional (3D) cross-point phase-change random-access memory (PCRAM) cells and 3D cross-point synaptic arrays. The Fenton reaction was conducted with 1,3-propylenediamine tetraacetic acid, ferric ammonium salt (PDTA–Fe) and H 2 O 2. The chemical oxidation degree of GeO 2 , Sb 2 O 3 , and TeO 2 evidently increased with the PDTA–Fe concentration in the CMP slurry, such that the polishing rate of the Ge 1 Sb 4 Te 5 ﬁlm surface linearly increased with the PDTA–Fe concentration. The addition of a corrosion inhibitor having protonated amine functional groups in the CMP slurry remarkably suppressed the corrosion degree of the Ge 1 Sb 4 Te 5 ﬁlm surface after CMP; i.e., the corrosion current of the Ge 1 Sb 4 Te 5 ﬁlm surface linearly decreased as the corrosion inhibitor concentration increased. Thus, the proposed Fenton reaction and corrosion inhibitor in the Ge 1 Sb 4 Te 5 ﬁlm surface CMP slurry could achieve an almost recess-free Ge 1 Sb 4 Te 5 ﬁlm surface of the conﬁned-PCRAM cells, having an aspect ratio of 60-nm-height to 4-nm-diameter after CMP.


Introduction
Recently, memory semiconductor device technologies have evolved to achieve fast switching, low-power consumption, and low memory bit-cost via the scaling-down of memory cells. Thus, two types of memory devices have been classified in terms of switching speed, power consumption, and cost per bit: dynamic random-access memory (DRAM) and three-dimensional (3D) NAND flash memory [1][2][3]. However, a 3D cross-point memory using phase-change random-access memory (PCRAM) has been proposed to generate a new storage memory in the memory hierarchy. Unlike DRAM, 3D cross-point PCRAMbased memory has demonstrated a nonvolatile memory characteristic with a fast operation speed of several hundred nanoseconds, and an excellent write-and-erase endurance cycle of >1 × 10 8 [4][5][6][7][8][9][10][11][12][13][14][15][16][17]. In addition, a cross-point synaptic array using PCRAM has been extensively researched for neuromorphic hardware performing parallel computing, because PCRAM can supply a large resistance ratio to achieve a sufficient synaptic weight range, low-power consumption due to fast switching, and good scalability for synaptic array [18][19][20][21]. In general, 3D cross-point PCRAM cells have been fabricated, and PCRAM-cells have a confined memory cell structure and a chalcogenide-based nonlinear-selector vertically stacked on the confined memory cell. Note that a confined memory cell structure with a vertical structure of the top electrode, PCRAM material filled in a confined hole, and bottom electrode are essential for minimizing the reset current. In addition, a chalcogenidebased nonlinear-selector stacked vertically on the confined memory cell is necessary for suppressing an integrated leakage current in a memory cell array.
To fabricate a confined PCRAM-cell structure, a PCRAM material such as Ge x Sb y Te z or Ge or N 2 doped Sb y Te z , etc., called a GST film, should be filled in the confined hole structure with a high aspect ratio of >10 [22]. Thus, sputtering followed by annealing, chemical vapor deposition (CVD), or atomic layer deposition (ALD) of a GST film in a confined-hole structure is utilized. Then, chemical-mechanical planarization (CMP) should be performed to eliminate the over-burdening of the GST film deposited on the confined hole structure and should be stopped as soon as a CMP stopping layer (i.e., Si 3 N 4 film surface) is exposed during the CMP process. It should be noted that CMP is conducted by a chemical reaction between the CMP slurry and the GST film surface as well as rubbing between nanoscale abrasives in the CMP slurry and the GST film surface. In particular, the GST films produced by CVD or ALD present a high chemical reactivity and extremely low mechanical hardness, so that the CMP of the GST film surface can easily result in corrosion of the GST film surface or surface-tensile-stress induced polishing-void after CMP. To avoid corrosion or polishing-void, the addition of an oxidizer (i.e., H 2 O 2 ) in the GST film surface CMP slurry is essential, which can passivate the GST film surface by forming a chemical oxidizer layer on the GST film surface. However, an oxidizer significantly reduces the polishing-rate of the GST film surface and cannot completely prevent the generation of a corrosion or polishing-induced-void on the GST film surface after a CMP [22,23]. In addition, the research has been reported for enhancing the GST film surface polishing-rate, presenting that the pH design of the GST film surface CMP slurry would be a key parameter to determine the GST-film surface polishing-rate [24,25].
In our study, to simultaneously enhance the polishing rate of the GST film surface and suppress surface corrosion during CMP, a novel GST film surface CMP slurry was designed with the introduction of the Fenton reaction and the addition of a columbic-interactive corrosion inhibitor. In general, the Fenton reaction has been used to accelerate the chemical oxidation degree of W or Cu during a CM via chemical decomposition of H 2 O 2 into dissolved O 2 and radicals [26][27][28]. However, for the first time, the Fenton reaction in the CMP slurry was introduced for the GST film surface CMP by designing a proper ferric-ionic catalyst working at a strong acid pH. Consequently, six different types of catalysts were evaluated: ferric-sulfate (i.e., Fe-sulfate), ferric oxalate (i.e., Fe-oxalate), ferric-nitride (i.e., Fe-nitride), PDTA-Fe, ferric-citrate (i.e., Fe-citrate), and ferric cyanide (i.e., Fe-cyanide) as shown in Table S1. Among them, PDTA-Fe as a ferric-ionic catalyst was selected, and the dependencies of GST-and Si 3 N 4 film surface polishing rate on the PDTA-Fe concentration in the CMP slurry were estimated. In addition, the dependencies of the static corrosion etch rate and surface roughness on the PDTA-Fe concentration were investigated, and their mechanism characterized by observing the chemical composition of the GST film surface depending on the PDTA-Fe concentration of the CMP slurry. Moreover, the effect of a corrosion inhibitor (i.e., polyethylenimine (PEI)) on the corrosion current reduction of the GST film surface and its mechanism was delineated by investigating the chemical composition of the GST film surface depending on the PEI concentration in the slurry. Note that polyethylenimine (i.e., PEI) is a polymer corrosion inhibitor having protonated amine functional group and can produce a nanoscale-thick polymer-passivation-layer on the GST film surface, so that the presence of corrosion on the GST film surface can be avoided. Finally, the dependencies of the GST-and Si 3 N 4 film surface polishing rate on the PEI concentration in the CMP slurry and the dependency of the polishing-induced recess on the PEI concentration in the slurry was estimated.

Chemical-Mechanical Planarization
The blanket or patterned Ge 1 Sb 4 Te 5 -film deposited 12-inch-wafers were cut into 4 × 4 cm squares. The CMP of the blanket or patterned Ge 1 Sb 4 Te 5 -film surface was conducted using a CMP polisher (POLI-300, G&P Tech. Inc., Pusan, Korea) attached to an industry standard CMP pad (IC 1000/Suba IV, Dow Chemical, Michigan, MI, USA). A pad break-in for warm polishing was carried out with a diamond conditioner for 10 min, and then two SiO 2 film dummy wafers were polished prior to the main polishing of the blanket or patterned Ge 1 Sb 4 Te 5 film surface. In situ pad conditioning was conducted after each polishing for various slurries. The applied head pressure was 6 psi, the rotation speed of the carrier holding the blanket or patterned Ge 1 Sb 4 Te 5 film samples was 70 rpm, and the rotation speed of the table attached to the CMP pad was 70 RPM/min. The flow rate of the CMP slurry was fixed at 100 mL/min, and the polishing time was set at 60 s. After 1 min of CMP, all blanket or patterned Ge 1 Sb 4 Te 5 film samples were buffed with DI water for 30 s to eliminate the remaining abrasives on the blanket or patterned Ge 1 Sb 4 Te 5 film surface.

Measurement Equipment
The polishing rate of the Ge 1 Sb 4 Te 5 film surface was estimated by measuring the Ge 1 Sb 4 Te 5 film thickness before and after CMP using ellipsometry (V-VASE, J.A. Woollam Co., Inc., Lincoln, NE, USA). Scanning electron microscopy (SEM, S-4800, Hitachi, Tokyo, Japan) image of the Ge 1 Sb 4 Te 5 film surface after CMP and etching was observed at an acceleration voltage of 15 kV. The surface roughness of the Ge 1 Sb 4 Te 5 film surface after etching was estimated by atomic force microscopy (AFM, Park system, Suwon, Korea), and the corrosion potential and current were measured using a potentiostat (CH750, CH Instruments, Austin, TX, USA). The chemical composition of the Ge 1 Sb 4 Te 5 film surface after etching and CMP were characterized using X-ray photoelectron spectroscopy (K-Alpha+, Thermo Fisher Scientific, Waltham, MA, USA) at 12 keV and 6 mA with a KE source of Al Kα (1486.6 eV). Cross-sectional LRB-x-RRB-images of the confined memory cells before and after the CMP pattern film were observed using high-resolution transmission electron microscopy (HR-TEM, JEM-2010, JEOL, Tokyo, Japan) with an accelerating voltage of 200 kV.

Effect of Fenton Reaction on Enhancement in Ge 1 Sb 4 Te 5 Film Polishing Rate and Corrosion Suppression during CMP
As a PCRAM-cell material, an as-sputter ternary-chalcogenide film, such as Ge 1 Sb 4 Te 5 film, was laser-annealed at 20 mJ for 10 µs, which was well crystallized with a face-centered cubic structure of (200), (220), (222), and (111) dimensions, as shown in Figure S1a. The annealing did not enhance the film density (i.e.,~6 g/cm 3 ) or the mean film hardness (~3 GPa), as shown in Figure S1b,c. In general, a CMP of chalcogenide film surface (i.e., Ge 1 Sb 4 Te 5 film) is conducted by chemical oxidation on the chalcogenide film surface, followed by mechanical rubbing between abrasives in the CMP slurry and the chalcogenide film surface. Chemical oxidation on the chalcogenide film surface, such as GeO 2 , Sb 2 O 3 , and TeO 2 , requires a CMP slurry with a pH of 1-9, as shown in the Ge, Sb, and Te Pourbaix diagrams in Figure S2. To investigate the degree of chemical oxidation on the Ge 1 Sb 4 Te 5 film surface, the dependence of the chemical oxidation degree on the oxidant (i.e., H 2 O 2 ) concentration was examined by dipping a Ge 1 Sb 4 Te 5 film at 70 • C for 5 min, followed by surface chemical composition analysis by XPS. The 2p 3/2 peaks of Ge and GeO 2 were found at 1217.3 and 1220.2 eV, respectively, as shown in Figure 1a. The 2p 3/2 peak intensity of GeO 2 rapidly increased with a H 2 O 2 concentration of 1-wt% and then slightly increased with increasing H 2 O 2 concentration, while that of Ge decreased abruptly up to a H 2 O 2 concentration of 1-wt% and then decreased slightly with increasing H 2 O 2 concentration, as shown in the inset of Figure 1a. In addition, the 3d 3/2 and 3d 5/2 peak intensities of Sb were located at both 538.1 and 528.1 eV, while those of Sb 2 O 3 were presented at both 539.6 and 530 eV, respectively, as shown in Figure 1b. The 3d 3/2 and 3d 5/2 peak intensities of Sb 2 O 3 at 539.6 and 530 eV considerably increased with the H 2 O 2 concentration and then saturated at a H 2 O 2 concentration of 3-wt%, while the Sb peak intensities slowly decreased until there was a H 2 O 2 concentration of 3-wt%, as shown in the inset of Figure 1b. The peak intensity of Sb 3d 5/2 at 528.1 eV (i.e., 56,454 a.u.) was higher than that of Sb 3d 3/2 at 538.1 eV (i.e., 35261 a.u.). Furthermore, the 3d 3/2 and 3d 5/2 peaks of TeO 2 were discovered at 586.4 and 576 eV, while those of Te were detected at 583.4 and 573 eV, respectively, as shown in Figure 1c. Both 3d 3/2 and 3d 5/2 peak intensities of TeO 2 increased slightly with H 2 O 2 concentration and then saturated at a H 2 O 2 concentration of 2-wt%, while those of Te weakly decreased with increasing H 2 O 2 concentration and then saturated at a H 2 O 2 concentration of 2-wt%. Comparing Figure  1a-c, a higher XPS peak intensity was observed for Sb 2 O 3 (421,875 a.u.), TeO 2 (160,629 a.u.), and GeO 2 (158,978 a.u.), indicating that the chemical oxidation degree of the Ge 1 Sb 4 Te 5 film surface was highly sequenced by Sb, Te, and Ge atoms.
Generally, a Ge x Se y Te z film surface CMP slurry uses an oxidizer to enhance the polishing rate, inducing corrosion pits on the Ge x Se y Te z film surface after CMP [22,23]. To enhance the Ge x Se y Te z film surface polishing rate and to avoid corrosion pits simultaneously, the Fenton reaction between the ferric-ionic catalyst and oxidizer was introduced for a Ge x Se y Te z film surface CMP slurry. Examining the slurry stability (i.e., no sedimentation of colloidal abrasives in a CMP slurry) among Fenton reaction catalysts such as ferricsulfate, ferric oxalate, ferric-nitride, PDTA-Fe, ferric-citrate, and ferric cyanide, PDTA-Fe showed good slurry stability and a relatively high Ge 1 Se 4 Te 5 film surface polishing rate, as shown in Figures 2 and S3. Without the Fenton reaction in a CMP slurry (i.e., only using oxidizer H 2 O 2 of 1.0-wt%), the Ge 1 Se 4 Te 5 film surface polishing-rate was~27 nm/min. However, the Ge 1 Se 4 Te 5 film surface polishing rate increased considerably from 27 to 105 nm/min when the PDTA-Fe concentration increased from 0 to 0.05-wt%, as shown in Figure 2, indicating that the Fenton reaction significantly enhanced the polishing rate of the Ge 1 Se 4 Te 5 film surface. In addition, the Si 3 N 4 film surface polishing-rate slightly and exponentially increased from 1.3 to 6.0 nm/min, which should be reduced because the Si 3 N 4 film surface should be utilized as a polishing stopping layer.
The dependency of the static corrosion rate (i.e., static etch rate) on the PDTA-Fe concentration was investigated after dipping Ge 1 Se 4 Te 5 films into slurries composed of 1.0-wt%, H 2 O 2 (i.e., oxidizer) of 1.0-wt%, PDFA-Fe (i.e., Fenton reaction catalyst) of 0, 0.01, 0.02, 0.03, 0.04, and 0.05-wt%, and HNO 3 (titrant) for 5 min at 70 • C, as shown in Figure 3. Note that the pad temperature of the Ge 1 Sb 4 Te 5 film CMP was~70 • C. As soon as PDTA-Fe 0.01-wt% was added to the slurry, the static corrosion rate rapidly decreased from 60.0 to 7.5 nm/min. Then, it slightly decreased from 8.8 to 3 nm when the PDTA-Fe concentration increased from 0.01 to 0.05-wt%. The SEM images of the Ge 1 Sb 4 Te 5 film surface after dipping into the slurries revealed the presence of corrosion-induced pits. Many of the corrosion pits of 10-50 µm in size were found without the Fenton reaction (i.e., only using H 2 O 2 ), as shown in the SEM image in Figure 3. The number and size of corrosion pits decreased when the PDTA-Fe concentration increased from 0 to 0.02-wt%, as shown in (i)-(iii) SEM images of Figure 3. When further increasing the PDTA-Fe concentration from 0.03 to 0.05-wt%, corrosion pits could not be found, as shown in (iv)-(vi) SEM images of Figure 3.

Fenton Reaction Mechanism for Enhancing Polishing Rate and Suppressing Corrosion on the Ge 1 Sb 4 Te 5 Film Surface during CMP
The chemical passivation (i.e., oxidation) on the Ge 1 Sb 4 Te 5 film surface via the Fenton reaction was characterized by chemical composition analysis via XPS of the Ge 1 Sb 4 Te 5 film surface dipping into the slurries at 70 • C for 1 min. Without the Fenton reaction, the relative XPS peak intensities were at 1220-eV (i.e., GeO 2 ) and 1218-eV binding energy (i.e., Ge) were 40,232.6 and 23,321.9 a.u., as shown in Figure 5a. When the catalyst PDTA-Fe concentration increased from 0.01 to 0.05-wt%, the relative XPS peak intensity of GeO 2 increased exponentially from 54,128.1 and 158,677 a.u., while for Ge, it slightly decreased from 17,040.3 to 6467.8 a.u., as shown in the inset of Figure 5a. This result implies that the dissolved O 2 and radicals (i.e., HO 2 • and OH • ) via the Fenton reaction between the Fe 3+ catalyst and H 2 O 2 can oxidize the Ge 1 Sb 4 Te 5 film surface by producing GeO 2 , and the chemical oxidation degree of the Ge 1 Sb 4 Te 5 film surface linearly increased with the catalyst (i.e., PDTA-Fe) concentration. In addition, without the Fenton reaction, the relative XPS peak intensities at 539.5-eV binding energy (i.e., Sb 2 O 3 of Sb3d 3/2) and 530-eV binding energy (i.e., Sb 2 O 3 of Sb3d 5/2) were 108,718.2 and 307,888.6 a.u., as shown in Figure 5b. Both peaks linearly increased from 141,982 to 294,615.2 and 307,888.6 to 623,299.1 a.u., respectively, when the PDTA-Fe concentration increased from 0.01 to 0.05-wt%. Otherwise, the XPS peak intensities at 538-eV (i.e., Sb of Sb3d 3/2) and 528.2-eV (i.e., Sb of Sb3d 5/2) binding energy were independent of the catalyst PDTA-Fe concentration. This result also means that the Fenton reaction could oxidize easily the Ge 1 Sb 4 Te 5 film surface by producing a chemically oxidized layer (i.e., Sb 2 O 3 ). Furthermore, without the Fenton reaction, the relative XPS peak intensities at 586-eV (i.e., TeO 2 of Te3d 3/2) and 576-eV (i.e., TeO 2 of Te3d 5/2) binding energies were 190430.6 and 274607 a.u., respectively, as shown in Figure 5c. Although, the catalyst PDTA-Fe concentration increased from 0.01 to 0.05-wt%, both peak intensities very slightly increased with the PDTA-Fe concentration. Moreover, the relative XPS peak intensities at 584-eV (i.e., Te of Te3d 3/2) and 573-eV (i.e., Te of Te3d 5/2) binding energies were almost independent of the catalyst PDTA-Fe concentration. This result indicates that the Ge and Sb atoms in the Ge 1 Sb 4 Te 5 film surface are well chemically oxidized rather than the Te atoms with dissolved O 2 and radicals (i.e., HO 2 • and OH • ). Thus, the formation of a chemically oxidized layer on the Ge 1 Sb 4 Te 5 film surface is mainly induced by Sb 2 O 3 and TeO 2 rather than GeO 2 , and the chemical oxidation degree of Sb 2 O 3 and TeO 2 was evidently enhanced with the catalyst PDTA-Fe concentration under the Fenton reaction. As a result, the polishing-rate of the Ge 1 Sb 4 Te 5 film surface linearly increased with the catalyst (i.e., PDTA-Fe) concentration, as shown in Figure 2.

Effect of Corrosion Inhibitor with Protonated Amine Groups on Corrosion Suppression and Corrosion Inhibition Mechanism
Although the Fenton reaction in a Ge 1 Sb 4 Te 5 film surface CMP slurry can significantly suppresses CMP-induced corrosion on the Ge 1 Sb 4 Te 5 film surface, as shown in Figure 3, an actual Ge 1 Sb 4 Te 5 film surface CMP process using a Fenton reaction-based CMP slurry can also result in CMP-induced corrosion, producing a polishing-induced void or a recess at a nanoscale confined-structure PCRAM-cell [22,23]. Thus, to avoid corrosion at the nanoscale confined-structure PCRAM-cell after Ge 1 Sb 4 Te 5 film surface CMP, a corrosion inhibitor was mixed with the Ge 1 Sb 4 Te 5 film surface CMP slurry using the Fenton reaction. A corrosion inhibitor with a protonated amine functional group was selected, that is, polyethylenimine (PEI) with a molecular weight of 25,000 g/mol. The dependence of the potentiodynamic polarization curve for the corrosion potential (E corr ) vs. corrosion current density (I corr ) on the corrosion inhibitor (i.e., PEI) concentration was observed for the Ge 1 Sb 4 Te 5 film surface being dipped into the Ge 1 Sb 4 Te 5 film surface CMP slurry using the Fenton reaction and a corrosion inhibitor (i.e., PEI), as shown in Figure 6. From the potentiodynamic polarization curves in Figure 6a, the dependency of E corr and I corr on the corrosion inhibitor concentration was calculated, as shown in Figure 6b. When the corrosion inhibitor (i.e., PEI) concentration increased from 0 to 0.09 wt%, E corr increased almost linearly from 0.28 to 0.3 V while I corr decreased exponentially from −6.15 to −6.93 A/cm 2 . This result indicates that the addition of a corrosion inhibitor (i.e., PEI) into the Ge 1 Sb 4 Te 5 film surface CMP slurry clearly suppressed corrosion, and the corrosion inhibition degree was greatly enhanced with increasing PEI concentration. To understand the mechanism by which PEI suppresses corrosion on the Ge 1 Sb 4 Te 5 film surface, the chemical composition of the Ge 1 Sb 4 Te 5 film surface after CMP using a Fenton reaction-based CMP slurry containing a corrosion inhibitor was analyzed by XPS as a function of the PEI concentration. NH 3 + and NH 2 + peak intensities adsorbed on the Ge 1 Sb 4 Te 5 film surface were found at 401.3 and 399.1 eV, respectively, as shown in Figure 7a. Note that both peaks represent the adsorption of a corrosion inhibitor (i.e., PEI) on the Ge 1 Sb 4 Te 5 film surface after CMP. Both peak intensities rapidly enhanced as soon as the PEI of 0.01-wt% was added in the Ge 1 Sb 4 Te 5 film surface CMP slurry and then very slightly increased with the PEI concentration when PEI concentration increased from 0.01 to 0.07-wt%, as shown in Figure 7b. With a further increase in the PEI concentration above 0.07-wt%, both peak intensities remarkably increased with increasing PEI concentration. In addition, the relative NH 3 + peak intensity is approximately two times higher than the relative NH 2 + peak intensity. The chemical composition analysis results evidently prove that a corrosion inhibitor (i.e., PEI) is well adsorbed on the Ge 1 Sb 4 Te 5 film surface, since the positively charged protonated amine groups (NH 3 + and NH 2 + ) of PEI are statically attractive to the negatively charged Ge 1 Sb 4 Te 5 film surface at pH 2. Note that the surface zeta potential of the Ge 1 Sb 4 Te 5 film surface at pH 2 is −1.29 mV, as shown in Figure S4. In addition, a higher concentration of the catalyst with protonated amine groups (i.e., NH 3 + and NH 2 + ) in a Ge 1 Sb 4 Te 5 film surface CMP slurry led to a higher adsorption of the catalyst on the Ge 1 Sb 4 Te 5 film surface. To observe how the adsorption of the catalyst affects the CMP performance, both Ge 1 Sb 4 Te 5 -and Si 3 N 4 film polishing rates were estimated as a function of the corrosion inhibitor (i.e., PEI) concentration. The Ge 1 Sb 4 Te 5 film surface polishing rate slightly and linearly decreased from 90.2 to 67.2 nm/min when the corrosion inhibitor concentration increased from 0 to 0.09, as shown in Figure 8. In addition, the Si 3 N 4 film surface polishing-rate decreased significantly from 4.8 to 0.97 nm/min when the corrosion inhibitor concentration increased from 0 to 0.03 and then saturated at~0.9 nm/min for a further increase in the corrosion inhibitor concentration. The effect of the corrosion inhibitor on the suppression of CMP-induced corrosion was clearly observed by observing the SEM image of the Ge 1 Sb 4 Te 5 film surface after a CMP without a corrosion inhibitor, the relative corrosion-induced surface roughness (i.e., Rq of pixel intensity in SEM images) of the Ge 1 Sb 4 Te 5 film surface after a CMP was 14.8, as shown in (i) of Figures 8 and S5. As soon as a corrosion inhibitor (i.e., PEI) was added into the CMP slurry, the relative corrosioninduced surface roughness rapidly decreased from 3.24 to 2.25 when the corrosion inhibitor concentration changed from to 0.01 to 0.09-wt%, as shown in the SEM images of (ii)-(vi) in Figure 8. In addition, we confirmed the Rq of the polished Ge 1 Sb 4 Te 5 film surface was~0.362 nm, which is a typical surface performance of CMP, as shown in Figure S6. These results indicate that the addition of a corrosion inhibitor (i.e., PEI) in the CMP slurry evidently suppresses the degree of corrosion-induced surface roughness.  In general, a PRAM cell is fabricated with a confined hole structure with a high aspect ratio (i.e., 4-nm in diameter and 60.7-nm height), as shown in Figure 9 [22,23]. The polishing rate of the Ge 1 Sb 4 Te 5 film surface without a confined-hole structure PRAMcell array pattern in Figure 8 would be lower than that of the Ge 1 Sb 4 Te 5 film surface with a confined-hole structure PRAM-cell array pattern, because the polishing-rate of the Ge 1 Sb 4 Te 5 film surface in a confined-hole structure PRAM-cell array pattern could be enhanced because of the confined memory cell structure-induced stress [22,23]. Note that a higher stress leads to a higher polishing rate. Thus, the dependence of the recess degree on the corrosion inhibitor concentration was observed by cross (x)-sectional TEM images of the confined PRAM cells after CMP. In this confined memory-cell structure, a CMP should be automatically stopped as soon as a stopping layer (i.e., Si 3 N 4 layer) is exposed during CMP. Without using a corrosion inhibitor, the Ge 1 Sb 4 Te 5 layer in the confined-memory cell was completely etched off, owing to the severe corrosion of the Ge 1 Sb 4 Te 5 film surface, as shown in the empty confined-hole structure memory-cell (i.e., white image) in Figure 9a. As expected, the recess degree at the confined-hole structure memory-cells decreased significantly and linearly from 4.9 to 0.8 nm, when the corrosion inhibitor (i.e., PEI) concentration increased from 0.01 to 0.09 wt%, as shown in Figure 9b-f. This result indicates that the addition of a corrosion inhibitor (i.e., PEI) to a Ge 1 Sb 4 Te 5 film surface CMP slurry could evidently suppress the generation of corrosion-induced recess after a Ge 1 Sb 4 Te 5 film CMP; that is, a higher degree of suppression of corrosion via the corrosion inhibitor led to less recess on the surface of the confined-hole structure memory-cell after a CMP stopped on the stopping layer (i.e., Si 3 N 4 ).

Conclusions
To enhance the polishing rate of a Ge 1 Sb 4 Te 5 film surface and to minimize recess in a confined-hole structure PRAM-cell array pattern, the Fenton reaction was introduced, and a corrosion inhibitor (i.e., PEI) with protonated amine groups (i.e., NH 2 + and NH 3 + ) was added to a Ge 1 Sb 4 Te 5 film surface CMP slurry. The Fenton reaction between the ferric-ionic catalyst (i.e., PDTA-Fe) and H 2 O 2 finally decomposed H 2 O 2 into dissolved O 2 , OH • , and HO 2 • in the slurry, as shown in Figure 10a. Then, the dissolved O 2 , OH • , HO 2 • , and corrosion inhibitor (i.e., PEI) diffused and adsorbed on the Ge 1 Sb 4 Te 5 film surface as shown in Figure 10b. The adsorbed O 2 , OH • , and HO 2 • on the Ge 1 Sb 4 Te 5 film surface chemically oxidized the Ge 1 Sb 4 Te 5 film surface by forming GeO 2 , Sb 2 O 3 , and TeO 2 on the Ge 1 Sb 4 Te 5 film surface, as shown in Figure 10c. In particular, the higher sequence of chemical oxidation on the Ge 1 Sb 4 Te 5 film surface was followed by the forming of GeO 2 , Sb 2 O 3 , and TeO 2. As soon as the chemically oxidized layer composed of GeO 2 , Sb 2 O 3 , and TeO 2 was produced on the Ge 1 Sb 4 Te 5 film surface, corrosion inhibitors with protonated amine groups (i.e., NH 2 + and NH 3 + ) were attached on the Ge 1 Sb 4 Te 5 film surface by the attractive force between protonated amine groups of corrosion inhibitors and the negatively charged chemically oxidized Ge 1 Sb 4 Te 5 film surface, thus causing a passivation of the organic layer of the film surface and suppressing corrosion, as shown in Figure 10c. Afterward, colloidal silica abrasives polished off the chemically oxidized Ge 1 Sb 4 Te 5 film surface (having a passivated organic layer) via rubbing between the colloidal silica abrasives and the chemically oxidized Ge 1 Sb 4 Te 5 film surface; i.e., the CMP of the Ge 1 Sb 4 Te 5 film surface, as shown in Figure 10d. After several iterations of Figure 10a-d, the CMP automatically stopped as soon as the Si 3 N 4 film surface was exposed, since corrosion inhibitors with protonated amine groups (i.e., NH 2 + and NH 3 + ) were preferentially attached to the highly negatively charged chemically oxidized Si 3 N 4 film surface, as shown in Figure 10e. It should be noted that the polishing rate selectivity between the chemically oxidized Ge 1 Sb 4 Te 5 film surface and the Si 3 N 4 film surface was 79:1, so that the patterned Ge 1 Sb 4 Te 5 film surface after CMP could attain a minimum recess (i.e., 0.8 nm). Therefore, the introduction of the Fenton reaction in a Ge 1 Sb 4 Te 5 film surface CMP slurry could significantly increase the polishing rate of the Ge 1 Sb 4 Te 5 film surface via efficient chemical oxidation (i.e., formation of GeO 2 , Sb 2 O 3 , and TeO 2 ) on the Ge 1 Sb 4 Te 5 film surface. Moreover, the addition of corrosion inhibitors with protonated amine groups (i.e., NH 2 + and NH 3 + ) effectively coated the Ge 1 Sb 4 Te 5 film surface by forming an organic passivation layer on the Ge 1 Sb 4 Te 5 film surface to suppress corrosion. In particular, the protonated amine groups (i.e., NH 2 + and NH 3 + ) were preferentially highly negatively charged on the Si 3 N 4 film surface rather than the negatively charged Ge 1 Sb 4 Te 5 film surface, so that the CMP was stopped and could achieve minimum recess of the Ge 1 Sb 4 Te 5 film surface. Because the Fenton reaction between the ferric-ionic catalyst and H 2 O 2 is a strong exothermic reaction, after H 2 O 2 was mixed with a CMP slurry, the remaining H 2 O 2 decreased remarkably with time. Thus, research on the addition of a scavenger to suppress the Fenton reaction is essential for achieving stability of the CMP slurry.