Electrodeposition of a Functional Solid State Memory Material: Germanium Antimony Telluride from a Non-Aqueous Plating Bath

The electrodeposition of germanium antimony telluride (GST) alloys from a single non-aqueous plating bath based on tetrabutylam- monium chlorometallate precursors is presented. The system provides a case-study for plating bath optimization in order to produce complex functional materials. GST deposits in the amorphous phase and the ﬁlm composition and morphology can be readily adjusted by tuning the three precursor concentrations and the electrodeposition potential. Adjustment of the precursor concentrations allows the preparation of deposits ranging from the binaries (GeSb, GeTe, Sb 2 Te 3 ) to ternaries with a wide range of compositions, including the standard Ge 2 Sb 2 Te 5 composition – more commonly known as GST-225 – which is widely used in the solid state memory industry. In this paper we present a detailed study discussing the complex interplay between the deposition of germanium, antimony and tellurium and how adjusting the concentrations of their chlorometallates allows control over the composition and also the morphology of the deposits. We also highlight the beneﬁts that arise from the wide separation in the deposition potentials for the three precursors, and in particular the ability to control the composition through modulation of the deposition potential.

The major challenge in electrodeposition is the achievement of adequate materials quality in order to ensure functionality. 1 For most functional materials their properties and performance are highly dependent on factors such as the composition and crystal structure, as well as the defect and impurity concentrations, which must be carefully controlled. The many degrees of freedom provided by electrochemical materials preparation routes are challenging to get right, but can also offer major advantages over other methods -such as physical and chemical vapor deposition (PVD, CVD) or sputteringsince the electrodeposition system allows simple, very fine adjustment of precursor concentrations, deposition rate, deposition driving force, temperature, etc.
In addition to the degrees of freedom available through electrodeposition, the process can be cheaper and potentially more environmentally benign. Whilst cost can be quite relevant in the coatings industries it is not always the most relevant factor in the semiconductor and electronics industry. However, while many material preparation techniques rely on line-of-sight, electrodeposition is inherently a bottom-up, space filling technique, and so is well suited to the filling of complex structures and the production of very small features in electrically defined locations. Examples of this are the Dual Damascene process for the preparation of copper vias 2 and the through-template electrodeposition methods developed for the fabrication of magnetic read-write heads for disk drives. 3 Both are very significant technological advances underpinning the production of microchips and the development of modern computers.
The quality of electrodeposited materials is often strongly affected by the plating bath. It is perhaps less obvious that it is also dependent on the inherent properties of the material itself. 4 Some compounds, such as BiTe alloys, naturally form crystalline materials during electrodeposition and the composition is relatively insensitive to the deposition potential. 5 CdTe is another example where stoichiometric crystalline material can be achieved over a wide potential range. 6 Other materials, such as CuZnSn films for CZTS/Se solar cells 7 or GeSbTe, 8 deposit as co-deposited crystalline or even amorphous materials. Especially for these materials, the composition is heavily influenced by the precursor ratio in the plating bath and the deposition potential. For materials where the deposit formation is not driven by the thermodynamic formation of a specific compound, the properties can be * Electrochemical Society Member. z E-mail: g.p.kissling@soton.ac.uk; p.n.bartlett@soton.ac.uk finely adjusted by tuning the plating bath, the deposition potential and other parameters, such as the plating bath temperature. Specialist equipment, such as the Hull cell, exists to study these effects on a technological level. 1 In this paper we present a study of the electrodeposition of materials in the ternary GeSbTe system from a non-aqueous plating bath. While GST based phase change memory materials are widely researched in the electronics community for their use in phase change random access memory (PCRAM), [9][10][11] only very limited work has been undertaken on the electrodeposition of these materials. This is due presumably to the significant challenges posed by the electrodeposition of germanium (due to germanium's very cathodic reduction potential) and the complexities of composition control in a ternary alloy. We are aware of only three papers discussing the electrodeposition of GST materials 8,12,13 and two of these were from work in our own group. 8,12 The other study was published by Stickney and co-workers, who prepared GST alloys using electrochemical atomic layer deposition from aqueous plating baths. 13 In addition, we are aware of two patents describing the electrodeposition of GST alloys either from an aqueous plating bath containing Ge, Sb and Te precursors 14 or again by electrochemical atomic layer deposition of sequential Ge, Sb and Te layers. 15 While GeSbTe deposits of a range of different compositions were obtained from both systems, the details provided are very sparse and no thorough characterization of the plating baths or the deposits was given in either patent.
In the non-aqueous plating bath described here, GST is deposited as an amorphous film 8,12 and the composition and morphology of the deposit are heavily influenced by the bath composition. The effects of the relative and absolute precursor concentrations and the deposition potential on the deposits are investigated and discussed. loys, since they represent the most common substrate used for phase change materials in the electronics industry. 24,25 The initial voltammetric characterization was performed on glassy carbon due to the higher resistivity of the TiN films (see supporting information). 23,26 TiN substrates consisting of a 2 × 1 cm 2 silicon chip with a 500 nm SiO 2 backing layer, a 200 nm TiN conducting layer and a 200 nm SiO 2 masking layer defining the electrode (0.126 cm 2 ) and contact areas were prepared using sputtering techniques. 23 The masking layer was patterned using a lift-off process. The electrodeposition procedure consisted of an initial 5 s 0.5 V equilibration step followed by a single potentiostatic electrodeposition step. The length of the electrodeposition step was either defined by a set cutoff charge relating to a chosen nominal deposit thickness or by the deposition time. The deposition potential was varied to study its effect on the deposits. It must be noted that determining the film thickness using the charge is only approximate since the faradaic efficiency of the electrodeposition process is not known exactly.
Characterization.-A Philips XL30 ESEM scanning electron microscope (SEM) was used to image the deposits. An acceleration voltage of 10 keV and a working distance of approximately 10 mm were used for all measurements. The elemental composition was determined by energy dispersive X-ray spectroscopy (EDX) using a Thermo Scientific NORAN System 7 X-ray Microanalysis System. The composition was determined in spectrum mode on an approximately 2800 μm 2 area. The calibration of the instrument was confirmed by comparison to a standard Ge 2 Sb 2 Te 5 sample prepared by PVD.
X-ray diffraction of the annealed deposits was performed on a Rigaku SmartLab diffractometer (Cu-K α ) with a 1 • grazing incidence angle and a DTex250 1D detector. Data were fitted using the GSAS package 27,28 in Le Bail mode, with structure models from ICSD. 29 All samples were annealed in a JipElec Rapid Thermal Annealer system.

Results and Discussion
The effects of a range of factors on the obtained deposits are discussed in this article. The results and discussion are structured into the following sections: i) voltammetric characterization of the precursor species and ii) the alloy plating baths; composition of asdeposited thin films as a function of iii) the plating bath composition, iv) the plating bath concentration, v) the deposition potential and vi) crystallographic characterization of annealed deposits as a function of the thin film composition.
Sample labeling.-In this paper individual deposits are labeled using bold capital Latin letters and relevant additional information to provide an overview over the 61 different deposits discussed. A summary of all the samples including information about the deposition potential (E dep ), the nominal deposit thickness (d) and the plating bath composition are provided in the Supplementary Information in Table S1.
Voltammetric characterization of the precursor species.-In order to understand the alloy plating baths, it is essential to characterize the voltammetry of the individual precursors. Figure 1 presents cyclic voltammograms recorded in plating baths containing the germanium (a), antimony (b) or tellurium (c) chlorometallates, respectively.
Several papers exist describing Ge deposition from non-aqueous plating baths. In these plating baths hydrogen evolution, which generally competes with Ge deposition in aqueous systems, can be avoided. Ge has been deposited from deep eutectics, 30,31 organic and supercritical solvents. 32,33 In our research we use a non-aqueous plating bath to facilitate the incorporation of Ge into the material.
The germanium voltammetry ( Figure 1a) is characterized by two major cathodic peaks at −0.7 V (c2) and −1.6 V (c3) which can be associated with the reduction of Ge IV → Ge II and the reduction of Ge II → Ge 0 , respectively. 32 On the reverse scan two plateaus are observed, suggesting diffusion controlled regions for both reduction reactions. A nucleation loop is not observed for this species. The voltammetry is irreversible and no stripping peak was observed in the investigated potential range. During the cyclic voltammetry a layer of germanium was deposited on the electrode surface.
After the first cycle the current densities of the reduction peaks are reduced. Since germanium is a semiconductor with a significant bandgap of 0.66 eV 34 it is conceivable that the deposit introduces a non-conducting layer on the electrode surface, hindering the electrochemical processes. When electrodepositing germanium at a constant potential the current generally falls to zero over time. This suggests that the maximum deposit thickness is self-limiting; probably related to the conductivity of the deposited germanium layer.
A small preceding peak at −0.2 V (c1) was also observed in the first scan. Similar preceding peaks were observed by Endres and co-workers when studying the electrodeposition of germanium from GeCl 4 in an ionic liquid using gold electrodes 30 and by our group studying the electrochemistry of [GeCl 6 ] 2complexes in CH 2 Cl 2 using a Pt disc working electrode. 32 Since c1 is only present in the first scan, but occurs on a range of different substrates, including glassy carbon, Pt and Au it may be related to non-specific interactions between Ge IV ions and the electrode surface.
The voltammetry of the antimony ( Figure 1b) and tellurium precursors ( Figure 1c) were discussed previously. 23 The voltammetry of [NBu n 4 ][SbCl 4 ] is characterized by a reduction peak at −0.83 V (c1), followed by the onset of a second peak at approximately −1.8 V (c2) and a stripping peak (a1) at 0.73 V (Figure 1b). A nucleation loop associated with the deposition of antimony is clearly visible. The deposition onset potential for antimony is at −0.4 V on glassy carbon and the nucleation overpotential is −0.23 V. The first reduction peak (c1) can be attributed to the reduction of Sb III to Sb 0 , as confirmed by the ability to electrodeposit antimony at these potentials using this system. 23 The origin of the second reduction peak (c2) is unclear. When cycling an as-deposited Sb film in the pure supporting electrolyte (0.1 × 10 −3 mol dm −3 [NBu n 4 ]Cl in CH 2 Cl 2 ) both reduction peaks are visible in the first scan but disappear in subsequent scans, probably due to removal of trapped reagents in the deposits. No significant changes are visible when inspecting the deposits visually before and after cycling in the supporting electrolyte excluding deposit stripping as a cause for the peak disappearance. It is thus assumed that c2 is indeed related to the antimony precursor species.
The reverse current in the potential region of c2 is diffusion limited and indistinguishable from the reverse current in the potential region of c1, suggesting that the same electrochemical process occurs across the whole potential range. Previous works on the voltammetry of antimony species in ionic liquids, 35 molten salts 36 and our own CH 2 Cl 2 work 23 have not shown the second peak, as it is outside the potential range investigated in those studies. In this paper we employed a wider potential range since the deposition of GST is performed in that region. It must be noted that the electrodeposition of Sb (at least as part of a compound) is possible from potentials more cathodic than −1.5 V. A main stripping peak (a1) at 0.7 V and a small preceding shoulder at around 0.35 V are observed. The preceding peak can be attributed to c2, as it is not present when the scan direction is reversed before c2. 23 The total stripping charge is approximately 3.6 times smaller than the deposition charge. The lower stripping charge may be due to the low adherence of Sb to the electrode surface. Upon continued scanning, the magnitude of the stripping peak decreases. The reduction peaks remain relatively constant.
The onset of the tellurium reduction was at approximately −0.05 V ( Figure 1c). However, the first reduction peak (c1) is only established at −1.25 V. A nucleation loop with a nucleation overpotential of approximately −0.35 V was observed for [NBu n 4 ] 2 [TeCl 6 ]. The deposition of tellurium onto glassy carbon requires a significant overpotential. Subsequent further tellurium deposition is easier, as can be seen by the earlier onset of the reduction current. The second and third scans also exhibit much larger currents at the more anodic potentials and show a further reduction peak (c2) at −1.7 V. These peaks are likely related to the further reduction of Te 0 to Te -II , evident in the reverse scan direction by two distinct diffusion controlled regions. The voltammetric behavior of [NBu n 4 ] 2 [TeCl 6 ] is very complex due to the formation of Te -II . 37 A small stripping peak (a1), suggesting poor dissolution of the Te deposit, was observed at 0.77 V. Other groups have discussed the electrochemical properties of Te IV species, but the electrochemical behavior of tellurium is very sensitive to the plating bath and the observed voltammograms differ widely. [37][38][39][40] Voltammetric characterization of the alloy plating bath.-It is possible to combine the Ge IV , Sb III and Te IV chlorometallate compounds into a single chemically stable plating bath and deposit functional alloy materials. 8,12 This allows a fine control over the composition by adjustments to the plating bath. Therefore, in order to control the composition of the deposits the complex electrochemical behavior of the plating bath must be investigated. By screening a range of precursor ratios, the features observed in the voltammograms of the resulting plating baths were assigned to the three elements. Figure 2 illustrates the changes observed when incrementally increasing the concentration of one precursor species in the plating bath. By comparison to Figure 1 it is apparent that the observed peak potentials are shifted compared to the peak potentials identified for the individual precursors indicating interactions between the precursor species during electrochemical deposition.
In Figure 2 three sets of voltammograms are contrasted. Each column represents data recorded in plating baths containing constant concentrations of two elements (e.g. Each element causes some obvious changes to the voltammetry, in addition to some more subtle effects. Germanium addition (Figures  2a-2f) causes a significant suppression of the stripping currents and quenches the current density measured in the reverse scans; especially for the second and third scans the current density can drop to nearly zero. Germanium addition also appears to cause the most cathodic peak (c4), which is not observed in the voltammogram for Sb : Te only (Figure 2a). In addition, the presence of germanium in the plating bath also leads to the formation of the c1 peak, which is the germanium pre-peak discussed previously (see c1 peak in Figure 1a).
Antimony addition significantly increases the current density of the first and third cathodic peaks (c2 and c4) (Figures 2g-2l) while c3 remains unchanged. A slight increase in the stripping current was observed for high concentrations of antimony.
Tellurium addition (Figures 2m-2r) shifts the reduction onset potential by +0.17 V, indicating that tellurium significantly facilitates the material deposition onto glassy carbon. Similar to the addition of germanium, an increasing tellurium concentration leads to an increase in the c3 peak.
As the behavior of peak c3 demonstrates, the increase in the various peak currents with increasing precursor concentrations is complex and the peak intensities are dependent on more than one plating bath species. This is partly caused by changes in the background due to the peak overlaps. However, the effects are too clear to be explained by background currents alone. Plots of the peak currents as a function of the incrementally-added plating bath species can be found in the supporting information ( Figure S1).
Most peak positions remain relatively stable when incrementally adding one of the precursor species to the plating bath (changes in the peak positions are illustrated in Figure S2). The one major exception to this is a significant anodic shift in c2 (by +0.17 V) and c3 (by +0.27 V) upon adding the first increment of tellurium (compare Figures 2m and  2n). Peak c2 remains relatively constant after the first addition and the current density remains quite constant. Peak c3 shifts back toward the initial potential as the current density increases once more tellurium was added to the plating bath. Other peak shifts were also observed, such as a cathodic shift in c4 when adding more antimony to the plating bath. However, the peak is not very clear and precise determination of the peak position was not possible.

Thin film properties as a function of the plating bath
composition.-GST films were prepared on TiN substrates from the plating baths discussed in the previous section (see Figure 2). The thickness of the deposits was controlled by the deposition charge (Q). Thin films (Q thin = −0.4 C cm −2 ) and thick films (Q thick = −2 C cm −2 ) were produced. The apparent film thicknesses -assuming 100% faradaic efficiency -were 200 nm and 1 μm, respectively.
As was determined by XRD, GST deposits as an amorphous material from the CH 2 Cl 2 based plating baths. 8 In addition to the deposit composition, the choice of plating bath composition also affects the morphology of the deposits.
Deposits were characterized by SEM and EDX analyses to understand the effect of the three precursors on the composition and the morphology. Figure 3 shows SEM images of the thin deposits on TiN substrates. Some significant changes are apparent when the plating bath composition is changed. The series depicting the addition of germanium to the plating bath (Figures 3a-3f) shows clearly that the germanium precursor also acts as a leveler. Whilst the low-germanium concentration plating baths yield generally smooth deposits with overlaid irregular structures, a higher concentration of germanium prevents the formation of this overlayer. The tellurium series shows the opposite effect (Figures 3m-3r). While the pure GeSb deposit is very smooth the addition of tellurium introduces the overlaying irregular deposits. Spot EDX measurements and XRD investigations 8 of un-annealed GST deposits suggest that these irregular structures are composed of crystalline tellurium. Addition of antimony has the smallest effect on the morphology (Figures 3g-3l), highlighting that it is mostly the interactions between germanium and tellurium that dictate the deposit appearance.
Thicker deposits with a nominal thickness of 1 μm were also produced and analyzed. The electron microscopy of these deposits is shown in the supplementary information ( Figure S3). The trends remain the same, with addition of the germanium species having a levelling effect on the deposits and addition of tellurium causing the formation of irregular overlayers. Significant cracking was observed in some of these thicker deposits. The overlayers also become more pronounced. As for the thinner films, changes in the antimony concentrations only have minimal effects on the morphology.
The composition of the deposits was investigated with EDX spectroscopy. Triangular composition plots and composition scatter plots for the thinner deposits are presented in Figure 4. Further data for the EDX analysis of thicker deposits is provided in the supporting information ( Figure S4). The trends observed in Figure S4 are very similar to the data presented in Figure 4.
As with the morphology, the effect of the concentrations of the three precursors on the deposit composition varies. The most linear and simple trend can be observed for the addition of the tellurium precursors. For this reason, we will discuss this first.
When increasing the concentration of [NBu n 4 ] 2 [TeCl 6 ] in the plating bath, the resulting deposits tended to a near-pure tellurium composition, as can be seen in Figures 4c, 4f and Figures S4 c, f. In the case of thin deposits (Figures 4c, 4f), the series started out from an equimolar GeSb deposit. For the thicker deposits ( Figure S4 c, f) the series originated at a GeSb 4 deposition. When the tellurium content in the deposits is increased the ratio of Ge : Sb remains stable. The crystal structure of GeSbTe has been reported as a metastable NaCl type crystal structure where tellurium occupies one sublattice and Ge and Sb share the second sublattice, which often also contains a significant amount of vacancies. 41 This may suggest that Ge and Sb are incorporated similarly into the lattice formed by the tellurium.
The deposits obtained at varying Ge precursor concentrations represented in Figures 4a and 4d  ] were present in 1 × 10 −3 mol dm −3 and 2 × 10 −3 mol dm −3 concentrations in the plating bath. If the two elements were plated at similar rates this should have led to a SbTe 2 composition. The lack of deposited antimony in the pure Sb-Te plating bath highlights that the electrodeposition of GeSbTe compounds is not straightforward. In addition, it appears that the added germanium substitutes antimony in the deposit: initially the tellurium content decreased rapidly when more [NBu n 4 ][GeCl 5 ] was added to the plating bath. However, for germanium precursor concentrations of more than 1 × 10 −3 mol dm −3 the tellurium content remained relatively constant at around 45%, while the antimony content in the deposits decreased significantly from a maximum of approximately 20% to nearly zero. A 1 × 10 −3 mol dm −3 [NBu n 4 ][GeCl 5 ] concentration leads to a deposit composition of approximately Ge 2 Sb 2 Te 5 . This is the preferred composition for solid state memory alloys and it appears that the alloy also favors approximately 20% vacancies in the Ge : Sb sublattice. 41 This may explain why the added germanium mostly replaces the antimony in the deposit, as the two elements compete for the same sites in the deposits.
An increased Sb concentration in the plating bath (Figures 4b,  4e) causes a relatively linear addition of antimony into the deposits. This is particularly apparent in the data representation provided in (e). The original GeTe deposits were very tellurium rich, with tellurium contents of above 80%. Interestingly, the addition of antimony appears to displace the tellurium, while the germanium concentration remains more constant. The composition shows a linear behavior when represented in a composition triangle ( Figure  4b). When extrapolating the data to the GeSb line a composition of approximately Ge 3 Sb 7 would be expected.
In general, it appears that tellurium is the most dominant of the plating bath species. Increase in the tellurium precursor species leads to a relatively linear increase of tellurium in the deposits and the antimony to germanium ratio remains relatively stable. When adding germanium to the deposits the composition tends toward a GeTe film, and the antimony is displaced. These significant compositional changes, in conjunction with the knowledge that the deposits are mostly amorphous (apart from some crystalline tellurium present in the amorphous deposit; see 8 ), suggests that the deposits are formed as a solid solution 4 with no clear driving force due to the formation of the alloy. Annealing of the deposits then leads to the formation of crystalline phases found in the GeSbTe, SbTe, GeTe system. It should also be noted that the thinner films tend to be more germanium rich than the thicker deposits.  Figure 5 summarizes the voltammetry, the deposit morphology and the composition as a function of the net plating bath concentra-tion. As expected, the morphology of the deposits becomes grainier when the plating bath concentration was increased, due to the faster deposit formation rate. The most concentrated plating baths produced deposits with a homogeneous surface appearance consisting of large globular features covering the whole electrode. These features were smaller for the deposits obtained from less concentrated plating baths and the irregular overlayer discussed previously is developed (cf. Figure 3). Rather less expected was that the rate of tellurium deposition ] while the third precursor concentration was varied from 0 to approximately double the standard concentration. The composition was determined by EDX and was obtained while scanning across 2,800 μm 2 areas. The composition shown is an average obtained from at least three different areas evenly spaced across the whole sample. The composition is representative of the average deposit composition. The labels in the triangle plots (a -c) and on the top axis of the scatter plots (d -f) correspond to the sample IDs. The pink star corresponds to the reference GST-225 sample. All films were deposited on TiN substrates (4 mm diameter exposed area) at a potential of −1.75 V vs. Ag/AgCl (0.1 mol dm −3 [NBu n 4 ]Cl in CH 2 Cl 2 ). A Pt gauze was used as the counter electrode. Table S2 provides the numerical data of the EDX quantification.  Table S4 provides the numerical data of the EDX quantification.  Figure 1 highlighted that Ge, Sb and Te have very different deposition potentials, with a potential difference of up to 1 V. This provides another degree of freedom for the tuning of the deposit composition by adjusting the deposition potential. The effect of the deposition potential on the composition of GST was thus investigated. This was done for seven different tellurium concentrations. Figure 6 shows the results for one tellurium concentration and is representative of the general behavior. It shows the effect of the deposition potential on the composition of the deposited material. By changing the deposition potential from −2 V to −1.5 V the concentration of Ge in the deposits can be significantly reduced and SbTe deposits are obtained. This opens up the possibility to fine-tune the composition of the deposits at a final stage after the plating bath was optimized.

Effect of deposition potential on the composition.-
Crystallographic characterization as a function of thin film composition.-The as-deposited GST films are amorphous, as was discussed previously. 8 Considering the non-uniform appearance of  Figure 7 presents XRD data for deposits obtained from plating baths with different tellurium concentrations at −1.75 V. Sample W-1.75 V was discussed in the previous section (see Figure 6) and plots for other plating bath concentrations are provided in the supplementary information ( Figure S5). There are significant changes when going from the deposit obtained with the smallest tellurium concentration to the deposit obtained with the largest tellurium concentration. While there is very little to no crystalline tellurium present in T-1.75 V and U-1.75 V a clear signal for tellurium becomes visible in X-1.75 V and Y-1.75 V. A number of structure types form in the Ge-Sb-Te system. In the ternary compositions there are trigonal phases with a range of stacking sequences and hence c-axis lengths (Table S6), and there are disordered cubic rocksalt-type phases (Table S7). The patterns did not match the cubic phases well, but the reflections did indicate trigonal phases, in addition to peaks from the TiN substrate and Te. The trigonal phases form with a wide range of compositions and lattice parameters, but refinement of the pattern of T-1.75 V (Ge 0.40 Sb 0.45 Te 0.15 ) and U-1.75 V (Ge 0.29 Sb 0.38 Te 0.33 ) (see  ]. The XRD pattern of an annealed GST standard prepared by PVD is also included (PVD standard). Figures 7, 8, S6 and S7, respectively) resulted in lattice parameters close to those of Sb-rich GST phases (Figure 8). Structurally this phase adopts the same stacking sequence as elemental Sb (Table S8), and one Sb pattern in ICSD also had lattice parameters very close to these. U-1.75 V (Ge 0.29 Sb 0.38 Te 0.33 ) also contained a second phase with a longer c-axis (see Figures 8 and S7) and this became the main Ge-Sb-Te phase as the Te concentration was increased in W-1.75 V (Ge 0.19 Sb 0.25 Te 0.55 ) (Figures 8, 9 and S8), Y-1.75 V (Ge 0.13 Sb 0.24 Te 0.63 ) (Figures 8 and S9) and X-1.75 V (Ge 0.08 Sb 0.14 Te 0.78 ) (Figures 8 and  S10). Peaks due to Te also became increasingly prominent and are dominant in the most Te-rich composition. The lattice parameters for the Ge-Sb-Te phase varied smoothly as expected for the varying composition, but comparison with literature phases (Figure 8) shows that Sb-Te phases along with some GST compositions are the main comparisons available for this structure type. Unfortunately there is a large range of reported values for Sb 2 Te 3 and it is not possible to infer the composition accurately by comparing the lattice parameters with literature compositions. However, it is clear that the composition of the crystalline components after annealing is changing and that in most cases more than one crystalline phase is present. Refined patterns are available in the supplementary information ( Figures S6-S12). Significant amounts of SbTe were also observed in a PVD prepared GST sample annealed under the same conditions (Figure 7, PVD standard), suggesting that the GST prepared by electrodeposition is comparable to conventionally prepared GST.
In Figure 9 XRD data for deposits W-  6 ] at different deposition potentials -are shown. The compositions of these deposits were shown in Figure 6. As is the case for deposits obtained from plating baths with different amounts of tellurium in the bath the crystal structure of the deposits changes with the deposition potential. Refined patterns are shown in Figures S8, S11 and S12. Deposition at −1.5 V (W-1.5 V) resulted in a Te-rich composition with significant elemental Te and the longer c-axis phase resembling Sb 2 Te 3 that was also observed in the more Te-rich compositions described above. Lattice parameters for these samples are provided in Figure 10. At −1.75 V (W-1.75 V) the intensity of the reflections due to the Sb 2 Te 3 -type phase became much stronger and both lattice parameters shortened. This lattice parameter change continued at −2 V (W-2 V), with no elemental Te observed and the reappearance of the Sb-type phase that was also seen in the Te-poor phases in the variable concentration studies.
The structures observed after the annealing process are strongly dependent on the composition of the deposit. Te-rich deposits generally show the presence of elemental Te, and two Ge-Sb-Te phases are observed in proportions that vary consistently with the amount of Te present. However, it has not been possible to pin down precise compositions of these phases as the lattice parameters of the alloys in the literature does not vary as systematically with composition as might be hoped. However, Sb and Te individually deposit as crystalline phases under similar conditions, 23 so the deposition of amorphous material does suggest alloying at the deposition stage. The segregation into a mixture of phases during annealing is consistent with the behavior of sputtered Ge 2 Sb 2 Te 5 .

Conclusions
A non-aqueous plating bath for the electrodeposition of ternary GeSbTe solid state phase change random access memory materials is presented. GeSbTe alloys plate amorphously and it is possible to accurately control the average atomic ratio of the deposit by adjusting the precursor ratio, the precursor concentrations, the deposition potential and the thickness of the deposit. In addition to their effect on the composition of the deposits the precursors also affect the morphology: an increase in the tellurium precursor concentration causes the formation of crystalline tellurium clusters on the amorphous GeSbTe deposit, an increase in the germanium precursor concentration acts as a leveler on the morphology of the deposits. The as-deposited GeSbTe alloys are mostly amorphous apart from the presence of crystalline tellurium clusters found in some cases. Upon annealing, the GeSbTe phases are formed.
Based on the results presented here, a strategy to prepare GeSbTe deposits with a desired average atomic ratio was developed: initially the correct Ge:Sb ratio is evaluated by varying the [ ] concentration is adjusted to define the tellurium content in the deposit in order to obtain the desired composition. This approach makes use of the fact that the ratio of Ge to Sb in the films is unchanged with changes in the Te content, as shown above in Figure 4.
The electrodeposition of GeSbTe allows the preparation of a wide range of material compositions from a single plating bath. This provides a simple approach to screening different compositions for their phase change properties and to depositing GeSbTe into high-aspect ratio nanostructures.
Future work will focus on improving the homogeneity and smoothness of the prepared deposits and on determining the origin of the inhomogeneities through better understanding of the nucleation process.