Structure and lithium‐ion mobility in Li 1.5 M 0.5 Ge 1.5 (PO 4 ) 3 (M = Ga, Sc, Y) NASICON glass‐ceramics

Lithium-ion batteries played such an important role in the technological revolution that took place in the past two decades that their developers received the 2019 Nobel Prize in Chemistry.1 To date, commercial lithium-ion batteries employ liquid-organic electrolytes which present both operational drawbacks like restricted cyclability (due to electrode corrosion) and health hazards caused by leakage of the electrode.2‒5 All-solid-state batteries are believed to be the next step in the evolution of batteries as they avoid such drawbacks.2‒5 Therefore, optimized solid-state electrolytes with high ionic conductivity are required. One of the most promising solid electrolyte systems found in the literature is based on glass-ceramics in the Na-Super Ionic Conductor (NASICON) phase field.2,3 The first NASICON glass-ceramic systems were reported and patented by J. Fu in 1997.6‒9 Typical compositions are Li1+xAlxM2−x(PO4)3, and Li1+x+yAlyM2−ySixP3−xO12 systems where M = Ti and Ge. In the NASICON structure, tetravalent M ions are octahedrally bonded by corner-sharing oxygens to phosphate tetrahedra. Li+ ions can occupy two sites named M1 (sixfold coordinated located between MO6 octahedra) and M2 (eightfold coordinated and located between two MO6 columns). In the LiGe2(PO4)3 parent compound, the M1 sites are fully occupied while the M2 sites are vacant, and at elevated temperatures, ionic transport occurs by hopping between these sites.2,3,10 To improve the ionic conduction, aliovalent substitution of M4+ by Mʹ3+ ions having slightly larger ionic radii11 than M4+ can be employed. This procedure widens the bottleneck window Received: 22 November 2019 | Revised: 22 December 2019 | Accepted: 22 December 2019 DOI: 10.1111/jace.16998


| INTRODUCTION
Lithium-ion batteries played such an important role in the technological revolution that took place in the past two decades that their developers received the 2019 Nobel Prize in Chemistry. 1 To date, commercial lithium-ion batteries employ liquid-organic electrolytes which present both operational drawbacks like restricted cyclability (due to electrode corrosion) and health hazards caused by leakage of the electrode. [2][3][4][5] All-solid-state batteries are believed to be the next step in the evolution of batteries as they avoid such drawbacks. [2][3][4][5] Therefore, optimized solid-state electrolytes with high ionic conductivity are required. One of the most promising solid electrolyte systems found in the literature is based on glass-ceramics in the Na-Super Ionic Conductor (NASICON) phase field. 2,3 The first NASICON glass-ceramic systems were reported and patented by J. Fu in 1997. [6][7][8][9] Typical compositions are Li 1+x Al x M 2−x (PO 4 ) 3 , and Li 1+x+y Al y M 2−y Si x P 3−x O 12 systems where M = Ti and Ge. In the NASICON structure, tetravalent M ions are octahedrally bonded by corner-sharing oxygens to phosphate tetrahedra. Li + ions can occupy two sites named M1 (sixfold coordinated located between MO 6 octahedra) and M2 (eightfold coordinated and located between two MO 6 columns). In the LiGe 2 (PO 4 ) 3 parent compound, the M1 sites are fully occupied while the M2 sites are vacant, and at elevated temperatures, ionic transport occurs by hopping between these sites. 2,3,10 To improve the ionic conduction, aliovalent substitution of M 4+ by Mʹ 3+ ions having slightly larger ionic radii 11 than M 4+ can be employed. This procedure widens the bottleneck window between the two sites and introduces additional Li + ions in the NASICON structure for charge compensation, which can be accommodated at the M2 sites. At higher x contents, a redistribution of populations takes place. 2,3 In the last years, a large number of studies have focused on the Li 1+x Al x Ge 2-x (PO 4 ) 3 system characterizing structural aspects and lithium-ion mobility [12][13][14][15][16][17] as well as technological applications. [18][19][20] Further compositional optimization has been attempted by combining isovalent (Sn 4+ and Ti 4+ ) and aliovalent substitution (Al 3+ and Cr 3+ ) approaches [21][22][23] in an effort to decrease the activation energy of M1 → M2 hopping. Aliovalent substitution by trivalent ions that are substantially larger than Ge 4+ has been rarely reported in LiGe 2 (PO 4 ) 3 glass-ceramics, apart from the most recent work of by Zhang et al. 24 reporting successful substitution of Ge 4+ by Ga 3+ . Here we are reporting the results of our recent efforts of substituting Ge 4+ by Ga 3+ , Sc 3+ , and Y 3+ in LiGe 2 (PO 4 ) 3 -based NASICON glass-ceramics probed by X-ray powder diffraction, impedance spectroscopy, and solid-state NMR. Such substitutions have been recently studied for the analogous LiTi 2 (PO 4 ) 3 glass-ceramics 25 and (unsuccessful) yttrium substitution was attempted in LiGe 2 (PO 4 ) 3 by the solid-state reaction. 26

PROCEDURES
Stoichiometric mixtures of Li 2 CO 3 (Vetec, 99%), GeO 2 (Sigma, 99.99%), NH 4 H 2 PO 4 (Sigma, ≥98%), Ga 2 O 3 (ABCR, 99.99%), Sc 2 O 3 (ABCR, 99.9%), and Y 2 O 3 (REEtec, 99.9941%) were used to synthesize the glass precursors. Raw materials were heated at 700°C for 2 hours to eliminate water, CO 2 , and to decompose the ammonium dihydrogen phosphate. The mixtures were then melted at 1200°C for 2 hours in a platinum-gold crucible, quenched in a cylindrical mold and annealed at 450°C for 2 hours to relieve thermal stress. The resulting glasses were sliced into 3.0 mm thick discs. Glass transition onset (T g ) and crystallization peak (T p ) temperatures were determined by DSC analysis (Netzsch DSC404) applying a heating rate of 10°C/ min. The results are summarized in Figure 1 and Table 1. Fully crystallized samples were produced by heat treatment of the precursor glasses at the crystallization temperature for 3 hours. For Li 1.5 Ga 0.5 Ge 1.5 (PO 4 ) 3 precursor glass (LGGP), heat treatment was employed at 606°C for 3 hours and the resulting crystallized sample was labeled LGGP-C-606. For Li 1.5 Sc 0.5 Ge 1.5 (PO 4 ) 3 precursor glass (LSGP), heat treatments were done on separate samples at 600°C for 1 hour, 690°C for 1 hour, and 832°C for 3 hours. Labels are LSGP-C-600, LSGP-C-690, and LSGP-C-832, respectively. Finally, for Li 1.5 Y 0.5 Ge 1.5 (PO 4 ) 3 precursor glass (LYGP), heat treatment was performed at 728°C for 3 hours and the crystallized sample was labeled LYGP-C-728.
X-ray powder diffraction patterns at room temperature were taken for all synthesized samples in a Rigaku Ultima IV X-ray diffractometer operating with Cu K α radiation between 5° and 100° in steps of 0.02°. An integrator time of 1 second was used. Rietveld refinements of the experimental powder patterns were done using the software MAUD to acquire the lattice parameters, the volume of the unit cell, and the ratios of the crystallized phases. 27 The refinements assume an R3c space group.
Impedance spectroscopy was performed in a Novocontrol Alpha Analyzer coupled with a Novotherm furnace, operating at temperatures between 30°C and 140°C (accuracy: ±0.1°C) in the F I G U R E 1 DSC thermograms of LGGP (A), LSGP (B), and LYGP (C) precursor glasses. Glass transition temperatures (T g ) and crystallization peak temperatures (T p ) are listed in Table 1. Thermograph A shows that LGGP precursor glass crystallizes only one phase while LSGP and LYGP precursor glasses crystallize more than one phase. T A B L E 1 Glass transition temperature (T g ) and crystallization temperature (T p ) acquired from the DSC thermograms in Figure 1  10 7 -0.1 Hz frequency range. Voltage amplitudes of 300 mV were applied. For these electrical measurements, parallel surfaces of glass-ceramics were previously sputtered with gold during 300 seconds. Total ionic conductivity data (σ) was calculated from where L is the sample thickness (around 3 mm) and A is the area of the electrode in contact with the sample. R is the total resistance acquired from the Zꞌ vs Zꞌꞌ impedance plots using the WinFIT software. High-resolution 31 P, 71 Ga, and 45 Sc NMR spectra were measured in a Bruker Avance 600 spectrometer operating at 14. LGGP precursor glass and its glass-ceramic (sample LGGP-C-606), using the pulse sequence of Larsen et al. 28 Pulse widths of 0.75 and 1.5 µs (30° and 60° flip angles, respectively), a nutation frequency of 111 kHz, and a recycle delay of 0.1 seconds were used. For these experiments, 150 echoes were acquired with 5 × 10 5 scans. Chemical shifts were referenced to powdered ScPO 4 (−48.2 ppm 29 against 1 mol/L ScCl 3 solution) and GaP (307 ppm 30 against 1 mol/L Ga(NO 3 ) 3 solution), respectively. Static 7 Li NMR experiments were done in an Agilent DD2 600 MHz spectrometer interfaced with a 5.6 T magnet using a commercial triple channel 4 mm probe between 160 K and 440 K. Single-pulse experiments employed pulse lengths of 1 µs with relaxation delays between 0.1 and 60 seconds. Chemical shifts were referenced to a 1 mol/L LiCl solution. Experimental spectra were analyzed using Gaussian and/or Lorentzian line shapes to extract the full width at half maximum of the spectra. All data processing and parameter extraction was done using DMFIT software. 31

| Structure
X-ray powder patterns of all the precursor glasses ( Figure 2) present no sharp diffraction peaks while the patterns for the crystallized samples ( Figure 3) present peaks related to crystalline LiGe 2 (PO 4 ) 3 (PDF2 card number 80-1922) indicating the crystallization of a NASICON phase. Phase-pure NASICON materials were only obtained for the gallium-substituted compound, except for a small unidentified peak near 2θ = 17° for the sample LGGP-C-606. For crystallized samples containing scandium, X-ray diffraction patterns showed that the sample LSGP-C-600 ( Figure 3) is still in the glassy state.
The LSGP-C-690 sample showed a dominantly crystallized NASICON phase while sample LSGP-C-832 showed multiple phases, including LiScP 2 O 7 (PDF2 card number 47-931) and GeO 2 (PDF2 card number 85-1515). Finally, peaks assigned to YPO 4 (PFD2 card number 83-658) and Li 4 P 2 O 7 (PDF2 card number 3-222) were identified in the X-ray powder pattern of sample LYGP-C-728. Table 2 lists the lattice constants, unit cell volumes, and percentages of the crystalline NASICON phases determined by Rietveld refinements (Figure 3) as well as for pristine LiGe 2 (PO 4 ) 3 NASICON. 31 Approximate mass fractions of the other crystalline phases as deduced from XRD are (30.7 ± 0.2)% for LiScP 2 O 7 ; (5.7 ± 0.2)% for GeO 2 ; (23.0 ± 0.2)% for YPO 4 and (2.8 ± 0.2)% for Li 4 P 2 O 7 . The data shown in Table 2  resonance line around −25 ppm is observed, reflecting phosphorus environments dominated by two bridging oxygen species in the glass phase (P 2 units). For the crystallized samples, however, the spectra are characterized by sharp resonance lines. Table 3 summarizes the results of the spectral deconvolutions conducted. The LGGP-C-606 sample exhibits a typical spectrum expected for statistical substitution of Ge 4+ by Ga 3+ with partially resolved contributions from P 4 4Ge,0Ga , P 4 3Ge,1Ga , P 4 2Ge,2Ga , and P 4 1Ge,3Ga units. 24 In contrast to reference 24, no GaPO 4 impurities were observed at the x = 0.5 substitution level. Figure 5 shows the 31 P MAS NMR spectra of the scandium-containing samples. Precursor glass (LSGP) and sample LSGP-C-600 present a broad resonance like the LGGP sample. The spectrum of sample LSGP-C-690 contains features related to the substitution of Ge 4+ by Sc 3+ in the NASICON structure. In this case, numerical deconvolution of the spectrum was able to identify P 4 4Ge,0Sc , P 4 3Ge,1Sc , and P 4 2Ge,2Sc units as well as a major broad feature assigned to 31 P nuclei in a residual glassy phase. For sample LSGP-C-832, the NASICON signal attributed to the P 4 4Ge,0Sc sites at −44.9 ppm is accompanied by various unassigned resonances (P1-P2, see Table 3) and one at −15 ppm due to LiScP 2 O 7 . 33 P2 may be assigned to P 2 units (phosphate units with two bridging oxygens) in [Sc(PO 3 ) 3 ] n polyphosphate chains. 34 Figure 6 presents the 31 P MAS NMR spectra of samples LYGP and LYGP-C-728. While the precursor glass only produces a broad signal, the spectrum of sample LYGP-C-728 consists of three prominent resonances: two overlapping signals at −45.2 and −44.1 ppm may be assigned to P 4 4Ge,0Y phosphate units 13,35 whereas the one at −12.8 ppm belongs to YPO 4 . There is no indication of Ge 4+ -> Y 3+ substitution. Finally, the 31 P MAS NMR spectra of samples LGGP-C-606, LSGP-C-832, and LYGP-C-728 present two signals above −10 ppm which can be ascribed to the two P 1 units of Li 4 P 2 O 7 . Figure 7 and Table 4 summarize the 45 Sc MAS NMR results. For the precursor glass, a broad asymmetric resonance at 11.6 ppm with a large average quadrupole coupling constant C Q of 12.5 MHz dominates the spectrum. This signal was simulated using the Czjzek model, which is based on a distribution of chemical shifts and quadrupolar coupling constants (C Q ). 36 This resonance is related to 45 Sc nuclei dispersed in the glass matrix. A resonance at −42.6 ppm is also observed with low intensity. Similar resonances appear in the spectrum of sample LSGP-C-600 but at 8.7 and −55.0 ppm, respectively. For sample LSGP-C-690, only an asymmetric signal is observed at 43.3 ppm, which is attributed to the six-coordinate Sc atoms on the Ge sites in the NASICON structure Finally, the 45

Sample
Fraction NMR signals for sample LSGP-C-832 were deconvoluted into four components (Table 4), reflecting the effects of the first-and second-order quadrupolar perturbations on the Zeeman levels. 71 Ga MAS NMR experiments were attempted for both LGGP and LGGP-C samples at MAS speeds of 45 and 10 kHz, respectively ( Figure 8). Three resonances were observed in the spectrum of sample LGGP at 106.9, 21.4, and −38.6 ppm. These signals were assigned to four-, five-, and six-coordinated gallium species and described by the Czjzek model; in addition 71 Ga{ 31 P} rotational echo double resonance results conducted in reference 24 show that all of these species have Ga-O-P connectivity. For sample LGGP-C-606, two resonances at 143.6 ppm and −3.5 ppm were observed and described by the Czjzek model. The first one can be ascribed to the four-coordinated gallium while the latter is assigned to the six-coordinated gallium, which is expected for Ga 3+ substituting Ge 4+ in the NASICON structure. 24 This assignment was also proven by 71 Ga{ 31 P} REDOR experiments. 24 The content of the five-coordinate gallium is rather low. Deconvolution of the spectrum also showed that the six-coordinated gallium has a larger average quadrupolar coupling constant (C Q = 3.6 MHz) than the four-coordinated gallium species (C Q = 1.7 MHz). Table 5 summarizes the results obtained by the deconvolution of the experimental 71 Ga MAS NMR spectrum. 71 Ga static QCPMG experiments were performed for both LGGP and LGGP-C-606 samples (Figure 8). From the spectra shown in Figure 8, one can see that the precursor glass gives rise to a very broad spikelet pattern centered at around 100 ppm. Its overall spread in frequencies is consistent with the MAS-NMR spectrum indicating that the MAS NMR spectrum is representative of the entire Ga inventory and there is no broad 71 Ga signal component that would go undetected in the MAS-NMR spectrum. This spectrum resembles the wideband uniform-rate smooth truncation QCPMG (WURST-QCPMG) spectrum of the model compound β-Ga 2 O 3 , which also contains four-and six-coordinated gallium atoms. 37,38

| Lithium-ion dynamics
Lithium-ion dynamics were characterized by impedance spectroscopy and variable-temperature static 7 Li NMR for the crystallized samples. Figure 9 shows a representative complex F I G U R E 4 31 P MAS NMR spectra of samples LGGP (A) and LGGP-C-606 (B). Black curves are the experimental data and red curves show the simulations. At (B), P 4 (4-n)Ge,nGa units are highlighted in shades of blue and correspond to P 4 4Ge,0Ga , P 4 3Ge,1Ga , P 4 2Ge,2Ga , and P 4 1Ge,3Ga following the increase of the ppm scale (from dark to light blue).
T A B L E 3 31 P chemical shift values found by numerical deconvolution of the experimental 31 P MAS NMR spectra. Sites generically labeled were not assignable with certainty. Pʹ 4 4Ge,0M indicates the highly crystalline pristine LiGe 2 (PO 4 ) 3 NASICON phase. Estimated errors are 0.2% for the fractions, 0.1 ppm for δ iso , and 3 Hz for the linewidths

Linewidth/ Hz
LGGP -100 −24.8 5160 LGGP impedance plot of Ge-substituted glass-ceramics. Note that the semicircles corresponding to samples LSGP-C-832 and LYGP-C-728 are highly flattened, indicating at least two electrical contributions attributed to grain and grain boundary, which are strongly overlapped. Those samples are highly resistive at 40°C, note the intercept of the semicircle with the x-axis on the 10 6 Ω.cm scale. The inset of Figure 9 shows an incomplete semicircle at the left side for sample LGGP-C-606, indicating the need for employing frequencies higher than 10 7 Hz, which are outside the experimental limits of the equipment. This sample also exhibits the lowest resistivity, note that the x-axis´ scale is within the 10 3 Ω.cm range. In all cases, it was not possible to separate grain and grain boundary contributions. Therefore, the discussion of electrical conductivity is limited to the total conductivity of samples, which includes grain and grain boundary contributions. Temperature-dependent σ values were analyzed assuming an Arrhenius behavior ( Figure 10) (σ = σ 0 exp{−E a /k B T}) and the values for the activation energy (E a ), ionic conductivity at room temperature (σ RT ), and the pre-exponential constant (σ 0 ) are summarized in Table 5. The gallium-containing sample LGGP-C-606 shows the highest total ionic conductivity at room temperature among the analyzed samples (1.1 × 10 −4 S/ cm) and the lowest activation energy. Conductivities of the other analyzed samples are several orders of magnitude lower.
Sample LSGP-C-600, which remained essentially in the glassy state, shows the lowest ionic conductivity at room temperature, 7.9 × 10 −11 S/cm ( Table 6). In order to probe the lithium mobility at the local level in our samples in more detail, temperature-dependent 7 Li NMR experiments (between 160 and 440 K) were measured under static conditions for LGGP-C-606, LSGP-C-690, LSGP-C-832, and LYGP-C-728 glass-ceramics (see Figure 11 for the representative case of LGGP-C-606). The temperature dependence of the full width at half maximum (FWHM) of the curves is plotted in Figure 12. For the samples containing scandium and yttrium, one can see the typical behavior of motional narrowing, where the FWHM is almost constant at low temperature (rigid lattice regime) and rapidly decreases above an onset temperature T onset,  reaching a constant plateau value at high temperature, corresponding to the fast motion regime. For sample LGGP-C-606, the ion dynamics are so fast on the NMR timescale (kHz) that the rigid lattice regime could not be reached within the experimentally accessible temperature range.

| DISCUSSION
From the X-ray powder patterns, one can see that substitution of Ge 4+ by Ga 3+ and Sc 3+ was successful, although in the latter case a multiphase crystallization pattern was observed in the DSC analysis for LSGP precursor glass and some extra peaks were also observed in the diffraction pattern. As previously shown, the exact degree of Ge substitution x in the NASICON phase can be obtained from the 31 P MAS-NMR spectra using the expression 2,13,14,24 where I n refers to the fractional area of each P 4 (4-n)Ge,nM signal component. Using the data shown in Table 3, values of x = 0.53 and 0.38 are calculated for LGGP-C-606 and LSGP-C-690, respectively. For samples LSGP-C-832 and LYGP-C-728, the situation is unclear from the solid-state NMR spectra, even though the small unit cell volume expansion of the NASICON structure observed by XRD (as shown in Table 2) suggests a substitution level of x well below 0.1. Overall, our data indicate successful substitution of Ge 4+ by Ga 3+ and Sc 3+ , although the gallium substitution appears to be more efficient since all of the batched gallium content enters the NASICON structure in our samples. For the scandium-containing samples, higher crystallization temperatures appear to promote the crystallization of the LiScP 2 O 7 phase, which has a structure similar to LiGe 2 (PO 4 ) 3 , where a scandium-centered octahedron is linked to tetrahedral phosphate units by corner-sharing oxygens. 39 R. Kahlaoui et.al. 33 studied the substitution of Ti 4+ by Sc 3+ in Li 1+x Sc x Ti 2-x (PO 4 ) 3 (0 ≤ x ≤ 0.5) and reported that the real value of x calculated from the 31 P MAS NMR data starts to deviate from the nominal value for x nominal = 0.2. In addition, the authors found that the real value for x nominal = 0.5 is 0.33, which is in good agreement with the value of 0.38 found here for sample LSGP-C-690. The substitution of Ge 4+ by Y 3+ was not achieved as shown by the X-ray powder pattern and the 31 P MAS NMR spectra. Arbi et.al. 26 reported the substitution of Ti 4+ by R 3+ (R = Al, Ga, Sc, In, Y) in Li 1.2 Ti 1.8 R 0.2 (PO 4 ) 3 glass-ceramics and showed that the Y 3+ substitution was indeed the only one that did not work. In the case of gallium and scandium, they observed that the efficiency of the substitution was the same for both trivalent ions (x = 0.16 and 0.18 for R = Ga and Sc respectively). However, although basically all gallium and scandium enter the NASICON structure of LiTi 2 (PO 4 ) 3 in their case, one should note that the authors used a lower content of trivalent ions than the ones reported in the present contribution.  Unsuccessful Y 3+ substitution into the Li 1+x Y x Ge 2-x (PO 4 ) 3 system with x = 0.2 and 0.4 was also reported by Yamamoto et al. 25 The success of the gallium substitution is also shown by 71 Ga MAS NMR, where almost 94% of the 71 Ga MAS NMR spectral area is due to six-coordinated gallium on the Ge site. The other 6% are found in four-coordinated sites that might be due to substitution of phosphorus by gallium in the tetrahedral site  42 The interpretation of the 45 Sc MAS NMR spectra in terms of coordination number is more complicated because in the case of six-coordinated Sc, the chemical shift values can vary by more than 100 ppm. 29 However, literature data indicate a tendency of six-coordinated scandium toward positive chemical shift values and of eight-coordinated scandium toward negative chemical shift values. 29,43 Based on the correlation suggested by Bräuniger et al. 43 for a limited set of Sc compounds (ScPO 4 , Sc 2 O 3 , ScOF, CaSc 2 O 4 , and Li 3 Sc(BO 3 ) 2 ) with coordination numbers of 6, 7, and 8, it may be tempting to assign the resonance in the glassy state to seven-coordinated scandium. On the other hand, comparison with other data on crystalline scandium phosphates 29 suggests that scandium is more likely to be six-coordinated in the glasses. For sample LSGP-C-690, the Czjzek analysis indicates an average chemical shift of 43.3 ppm which is consistent with octahedrally coordinated Sc. The distribution width of 20.9 ppm reflects the intrinsic disorder in the NASICON phase, which is expected based on the random substitution of Ge by Sc.
The highest overall room temperature conductivity achieved for our samples is 1.1 × 10 −4 S/cm for the essentially phase-pure sample LGGP-C-606. The corresponding   39 In the case of the LSGP-C-600 sample, the lowest ionic conductivity is expected, as the precursor glasses of NASICON glass-ceramics are usually more resistive than its respective crystallized samples. 8,44 In those glass ceramics in which no cation substitution takes place because the trivalent species forms non-ion conducting crystalline components such as YPO 4 , one would generally expect the ionic conductivity to be lower than in LiGe 2 (PO 4 ) 3 . In contrast, for sample LYGP-C-728, the overall room temperature conductivity is found to be two orders of magnitude higher than the reported one for the unsubstituted NASICON (LiGe 2 (PO 4 ) 3 7 Li NMR spectra, one can estimate the activation energy of the faster lithium-ion using the empirical expression proposed by Waugh and Fedin: T A B L E 6 Activation energy (E a ), total conductivity at room temperature (σ RT ), and pre-exponential factor (σ 0 ) acquired by assuming an Arrhenius behavior of the overall conductivity σ(T). Data for LiGe 2 (PO 4 ) 3 were taken from Ref. [ 21 ] F I G U R E 1 1 7 Li static NMR spectra of sample LGGP-C-606 at selected temperatures between 160 and 440 K. From this expression, activation energies were estimated as (0.33 ± 0.01) eV for sample LSGP-C-690, (0.36 ± 0.01) eV for sample LSGP-C-832, and (0.43 ± 0.01) eV for sample LYGP-C-728. It was not possible to estimate the activation energy for sample LGGP-C-606 since T onset could not be determined for this sample, as the low-temperature plateau could not be reached due to hardware limitations ( Figure 12). As frequently reported, the activation energies deduced from NMR lineshape analysis are significantly lower than those determined by electrical conductivity measurements, even though the compositional trends are the same. 47 The discrepancy most likely arises because the two techniques probe into ionic motion on different length scales. 48 In addition, there is most likely a distribution of mobilities and activation energies, resulting in a distribution of onset temperatures. Thus, by identifying those temperatures at which the linewidths get noticeably reduced, the Waugh-Fedin method tends to emphasize the contribution of those ions that are most mobile.

| CONCLUSION
Substitution of Ge 4+ by Ga 3+ , Sc 3+ , and Y 3+ in the LiGe 2 (PO 4 ) 3 NASICON phase was attempted using the glass-ceramic route. The results show that phase-pure Li 1.5 Ga 0.5 Ge 1.5 (PO 4 ) 3 could be obtained as also reported recently. 24 Substitution by Sc 3+ was achieved for the samples heat treated at 690°C for 1h (LSGP-C-690), although no phase-pure samples were obtained. No indication of any substitution of Ge 4+ by Y 3+ was observed in the NMR data. The different degrees of substitution can be explained on the basis of the differences in the ionic radii. While Ga 3+ ions (r = 62 pm) can still be accommodated substituting for Ge 4+ (r = 53 pm), those of Sc 3+ (r = 74.5 pm) and Y 3+ (90 pm) are significantly larger. Still, the possibility of at least partial substitution documented for Sc 3+ in this study reflects an impressive tolerance of the NASICON lattice for ionic size mismatch on the Ge sites. The highest room ionic conductivity (1.1 × 10 −4 (Ω.cm) −1 ) was achieved for Ga-substituted samples, which presented the NASICON as a single phase.