Formation of Chemically Reacted Glasses. The synergistic effects of oxygen additions on the properties of the NPSO SEs were examined on different compositions of x = 0, 0.15, 0.30, 0.60 that were synthesized via the high-energy MCM and systematically characterized. The XRD patterns (Supplementary Fig. S1) show that all the raw materials became amorphous after MCM, as no diffraction peaks of the starting materials were detected. Hayashi et al. and we have observed that unreacted Na2S and P2S5 were observed for MCM times shorter than those used here and that unreacted Na2S was observed for Na2S contents higher than those used here.22,30 Full chemical reaction between the Na2S and P2S5 and not just amorphization of the starting materials was proven though the chemical spectroscopies, infrared (IR), Raman, nuclear magnetic resonance (NMR), and X-ray photoelectron spectroscopy (XPS). The DSC curves in Supplementary Fig. S2 further reveal that all of the amorphous samples are glassy exhibiting a glass transition (Tg) around 200°C. For the oxysulfide SEs (x > 0), the working range, or the difference between Tg and Tc, becomes larger with increasing oxygen contents, indicating adding oxygen results in greater resistance to crystallization.
While complete understanding of the complex mechano-chemical processes taking place during high-energy MCM are as yet not fully known, it is generally agreed that MCM generates high local temperatures during the rapid ball to material to wall and ball to material to ball collisions and generates rapid thermal quenching after the collision as the materials are thermalized back to the near room temperature of the reaction vessel. The combination of high local temperature and the rapid quenching are believed to be responsible for the formation of amorphous materials that also exhibit a liquid state formed glassy Tg. However, just as in MQ systems, the quenching rate is not always sufficient to yield a fully amorphous glass and the presence of small amounts of crystalline phases can be detected in some systems, which was recently observed in the Na2S–P2S531 and Li2S–P2S532 systems.
To examine in more detail the possible formation of small amounts of fine-grained crystalline phases in the NPSO SEs during the MCM process, high-energy synchrotron XRD patterns were collected and shown in Fig. 1a. All synthesized SEs feature two broad halos superimposed with weak Bragg peaks, indicating that a small number of fine-grained crystalline phases do form and are embedded in the majority of an otherwise glassy matrix. The TEM image and SAED pattern (Supplementary Fig. S3) show that the crystalline phase is tetragonal Na3PS4 (t-Na3PS4) with an average particle size of ~ 3 nm. However, as seen in Fig. 1a, when x reaches 0.6, these crystallization processes weaken as evidenced by the fact that the t-Na3PS4 peak has almost disappeared. However, it is now replaced by the appearance of crystalline Na2S that may arise from incomplete chemical reaction.
The reason for the appearance Na2S at the higher O substitution was examined by Raman spectroscopy since it is sensitive to sulfur-containing units. As seen in Fig. 1b, the primary feature peak25 centered at 420 cm− 1 is assigned to the stretching mode of the PS4 unit, which is a non-bridging sulfur (NBS) unit, identified as P0 24,30 where the superscript 0 is the number of bridging oxygens (BOs) on this short range order (SRO) unit. The Gaussian fitting of this mode (Supplementary Fig. S4) suggests that with the oxygen incorporation, a small population of the original P0 units disproportionate into P1 units (containing one bridging sulfur (BS), e.g. P2S7) with the liberation of sodium sulfide (2P0 → 2P1 + Na2S)24 to balance charge. This disproportionation reaction is supported by the observation of Na2S peaks in the synchrotron XRD results. The mode centered at 380 cm− 1 is assigned to the P2S6 unit, which possesses a homopolar P–P bond, that decreases in concentration with the oxygen addition33. The formation process of the P2S6 defect unit generates residual sulfur species (e.g. sulfur element) that show a characteristic Raman mode arising from the –S–S– linkage and occurs at 470 cm− 1. As indicated from both Fig. 1a and 1b, except for a small amount of crystalline phases of Na3PS4 and Na2S, which we estimate to be in the range of < 5%, the main composition of the NPSO SEs is glass.
Chemical Short-Range Order. 31P MAS–NMR was used to gain further insights into the glassy phase of these NPSO SEs by examining the local structure around the phosphorus glass forming cation. Deconvolution of the 31P NMR spectra (Fig. 1c) shows that the glass Na3PS4 (x = 0) is composed mainly of PS4 (82 ppm) and P2S6 (108 ppm) units, which is consistent with the Raman spectra described above. With the incorporation of oxygen, three new peaks attributed to the formation of mixed oxysulfide and oxide units, PS2O2 oxysulfide (63 ppm), PSO3 oxysulfide (32 ppm), and PO4 oxide (3 ppm) units can be clearly observed34. The peak for the expected but missing PS3O oxysulfide units is nearly indiscernible since it has essentially the same chemical shift as the PS4 units29.
However, following the methods we proposed before24,35, this unit and its relative fraction of all of the SRO units was determined by careful spectral deconvolution and corrected for charge balance among the Na+ ions and the various PS4 − xOx anions. The composition dependence of the population of the various SRO structural units in these oxysulfide NPSO SEs is given in Fig. 1d and shows that as expected the fractions of mixed oxysulfide units dramatically increases with oxygen doping level and become the dominant species when x reaches 0.6. The appearance of PS4 − xOx oxysulfide SRO units suggests that the oxygen has been incorporated into the P–S tetrahedra unit, which is expected to improve the chemical stability and mechanical strength of the glass network over that of the pure sulfide SE.
FTIR spectroscopy was further applied to explore the chemical bonding between phosphorus and sulfur and phosphorous and oxygen. The FTIR spectra shown in Fig. 1e can be divided into three sections corresponding to terminal P–S− (400–600 cm− 1), BO P–O–P (600–950 cm− 1) and terminal non-bridging oxygens (NBOs) P–O− or P = O (950–1200 cm− 1) modes, respectively36,37. Detailed peak assignments are listed in Supplementary Table S1. It is evident that oxygen incorporation into the glass matrix leads to a slightly increased fractions of P–O− and P = O bonds and a particularly significant increase in the fraction BO P–O–P bonds, where the O atom is bridging between two phosphorus atoms. Further evidence of the formation of BOs on adding oxygen to Na3PS4 can be found in O 1s XPS spectra (Fig. 1f), where the BO with the binding energy of 532.5 eV38,39 accounts for the higher fraction of the oxygen (NBOs) for larger x values in NPSO SEs. Spectral deconvolution of Na3PS3.4O0.6, for example, shows that more than 90% of the added oxygen atoms are present as BOs in the glass. This behavior is completely consistent by our previous work40 where in a similar Li2GeS4 − xOx glass system, the added oxygen was also found to preferentially replace S in the BS positions until all of the BS were eliminated. It was not until there were no more BS to replace did the oxygen form NBOs. The results here for the NPSO series reveal a more complex mechanism for the added oxygen to form BO because in the present system, there are no BSs to begin with. Hence, the added O forces the formation of the disproportionation reactions described above. To maintain charge balance in the series, the formation of a BO requires the formation of “free” Na2S in direct proportion to the amount of O added. While overall charge balance appears to be maintained in this system, the formation of BOs through the addition of O also forms increasing chain units in the form of P1 units. Kmiec et al. has observed this same behavior in the slightly lower modifier Na4P2S7 − xOx system35. Using the SRO composition map similar to Fig. 1d, Kmiec calculated the average chain length that forms through the formation of BOs in these systems. The network chain connection of the structural units creates an interconnected glass network that produces glasses with increasing packing densities, mechanical moduli, and stronger chemical bonding that improves the chemical stability of SEs.
Homogeneous Glass with Improved Mechanical Properties. A defect-free and mechanically robust SE is a prerequisite for successfully cycling with Na metal anodes11. Consequently, the morphological structure and the mechanical properties of NPSO SEs were investigated. For comparison, the widely studied heat-treated Na3PS4 glass-ceramic SE (HT-Na3PS4) was also studied25,41. Obvious pores and grain boundaries are clearly observed in the SEM images of pelletized HT-Na3PS4 (Fig. 2a–b). These surface defects (pores and grain boundaries) are believed to induce dendrite formation, subsequent dendrite propagation 42–44, and eventual short-circuit through the SE, as demonstrated in Supplementary Fig. S5. Na3PS4 glass SE, however, shows fewer structural pores compared to HT-Na3PS4.In sharp contrast to both of these SEs, the surface of the NPSO glass SE appears to be absent of pores from the surface through to the interior. To the best of our knowledge, this defect-free fully dense glassy morphology induced by simple low-cost room temperature uniaxial pressing is observed here for the first time in any SE fabricated by simple cold-pressing. Such a perfect morphology is presumably closely related to the outstanding formability of NPSO. As shown in Fig. 2b and Fig. S5, this phase is nearly fully densified at a pressure as low as 150 MPa. In comparison, Na3PS4 glass and HT-Na3PS4 couldn’t achieve similar relative density even when much higher pressures were applied during cold-pressing. The excellent formability of these oxysulfide SEs is attributed to the synergistic effects of mixed P2S5 and P2O5 glass formers, which not only build a close-packed glass network with abundant BO units but also facilitate the local sintering process of the powders as demonstrated by more interparticle adhesion and necking compared to the pure sulfide SEs particles(Supplementary Fig. S7). Overall, for oxysulfide SEs, their unique formability and defect-free structure will undoubtedly enhance their mechanical strengths and reduce the likelihood of dendrite-induced short-circuit when using Na metal anode, as described in Scheme 1a.
To quantify the mechanical properties of SEs, two critical parameters: Young’s modulus E and hardness H were measured using nano-indentation technique45. Typical load-displacement curves in Fig. 2 and Supplementary Fig. S8 show that the HT-Na3PS4 pellet experiences a sudden increase of indenter penetration during the loading process, which is not found in glassy NPSO SEs. This is referred to as “pop-in” and is associated with the generation of a crack45,46. This behavior indicates that the HT-Na3PS4 SE is more brittle than the glassy counterparts. Such cracking of course is certainly unfavorable for the preparation and practical application of SEs. Further, benefiting from the homogeneous property of the glassy materials, the NPSO SEs display very small standard deviations for E and H as seen in the bar chart in Fig. 2d. Moreover, the oxygen doping results in an increase of E and H of SEs, which supports the observation that oxygen doping produces more BO units in the glass network, thus providing more robust mechanical properties. The E and H of Na3PS3.4O0.6 glass were measured to be 20.9 ± 0.7 GPa and 1.0 ± 0.1 GPa, respectively, which is the highest among the NPSO series and even higher than those of the reported sulfide-based Li-ion and Na-ion SEs prepared by hot-pressing18,47,48. Assuming that Poisson’s ratio ν is ca. 0.3 according to the study from Sakuda et al.49, the shear modulus G of Na3PS3.4O0.6 glass is ca. 8.0 ± 0.3 GPa, which is believed to be sufficient to suppress dendritic penetration of Na metal as predicted by the Monroe and Newman criterion50.
Electrochemically Stable SEs Against Na Metal. As described above, the chemical and electrochemical stability of SEs against Na metal anodes is critically important for developing high-performance ASSSBs and for this reason we have examined the Na − NPSO interface using time-dependent electrochemical impedance spectroscopy (EIS) and XPS. Na/NPSO/Na symmetric cells were fabricated to monitor the evolution of the EIS spectra and the area specific resistance (ASR) over 5 hours resting time and this is shown in Fig. 3a and Supplementary Fig. S9, respectively. The obtained EIS spectra can be divided into three semicircle regions51 and the fitting parameters are listed in Supplementary Table S2. The high-frequency semicircle represents the combined bulk and grain boundary resistance and capacitance (Rb + Rgb, Cb + Rgb) of the SEs; the mid-frequency semicircle with the characteristic capacitance of 10− 6~10− 7 F arises from the interfacial resistance and capacitance (Ri, Ci) between Na and SE52; and the low-frequency one corresponds to the charge-transfer resistance and capacitance(Rct, Cct). As seen from Table S2 and Fig. S9, the added oxygen has a positive effect on improving the SEs’ chemical stability against Na metal. Not only do the Ri and Rct resistances decrease significantly, but their increase after contact with Na metal becomes negligible. Further, Fig. S9 shows that the ASR becomes less significant when more oxygen is doped. Indeed, it is noteworthy that for the Na3PS3.4O0.6 SE, not only does the ASR increase essentially vanish, it further presents a negligible change of Rb and Rct, but also indistinguishable Ri, suggesting an intimate and ohmic contact that forms a stable interface between Na3PS3.4O0.6 and Na. In contrast, all of the other SEs display a Ri varying with time, indicating an unstable Na − SE interface.
To identify the interphase composition that apparently grows with time for the x ≠ 0.6 glassy NPSO SEs, Na metal was detached from the symmetric cells and the surface of SEs was probed by the XPS. Compared to the surfaces of the pristine SEs (Supplementary Fig. S10), Fig. 3 shows that the surfaces of the Na3PS4 − xOx, x = 0, 0.15, and 0.3, SEs exhibit two new pairs of doublets for P 2p spectra and S 2p spectra at lower bonding energy after contacting with Na metal. These new pairs of doublets correspond to the reduced phosphide and sulfide species, respectively, and indicates that these electrolytes have been reduced by Na metal. According to the theoretical calculation from Tian et al.12 and the experimental study from Wenzel et al.13, these reduced species are Na2S and Na3P, respectively, the latter of which is a mixed ionic and electronic conductor that can cause continuous reduction of the electrolyte. Thus, for these SEs an unstable interphase grows between the Na anode and SE, impeding the Na stripping/plating process. Similar phenomena have been observed in other pure sulfides and selenides, e.g. Na3PS4, Na3SbS4, Na3PSe4, and while they all have high ionic conductivities, their propensity to form these unstable interphases make them unsuitable in Na-metal batteries14.
In contrast, Fig. 3 shows that these same XPS signals from these reduced species are nearly indiscernible above the baseline noise in the S2p and P2p spectra for the Na3PS2.4O0.6 SE, validating the negligible ASR increase from EIS measurement. To our knowledge, this is the first sulfide-based defect-free fully glassy SE that forms a stable non-reacting interphase directly against Na metal.16 We assert here that the outstanding chemical stability of Na3PS3.4O0.6 in contact with Na metal is contributed by: first, Na3PS3.4O0.6 has a more robust glass network than the other SEs due to the existence of more BOs units which have higher electronic binding energies; second, and in a similar way in comparing NBO to NBS moieties, oxide and oxysulfide units also have higher electronic binding energies than pure sulfide units; third, even if NBOs and BOs do react with Na metal, the terminal oxygen in P − O− may react with Na metal and form a thin electronic insulating interphase (Na2O) as suggested by Gao et al. for solid-state electrolytes53 and by Seh et al. for liquid electrolytes54.
Electrochemically Stable and High Na + Ion Conductivity Tri-Layer Composite SEs. Figure 4a shows the temperature-dependence of the Na+ ionic conductivities that were obtained from the Nyquist plots (Supplementary Fig. S11) for the Na3PS4 − xOx, x = 0, 0.15, 0.30, 0.60. From Supplementary Fig. S12, it is intriguing that the ionic conductivities exhibit a 6-fold increase with the initial addition of oxygen, x = 0.15, rendering the conductivity of the Na3PS3.85O0.15 SE as high as 2.7×10− 4 S cm− 1 and activation energy as low as 41.5 kJ mol− 1. This anomalous increase in the ionic conductivity may be associated with two distinctly different features of the effects of the addition of oxygen to the base Na3PS4 glassy SE. First, when we doped oxygen into the comparable Li2GeS2 − xOx glassy SE in a previous study40, we observed that the added oxygen increased the free volume available to the mobile Li+ ion for conduction. This led to an increase of the ‘doorway’ radius, rD, between Li+ ion sites, thereby lowering the overall activation energy and increasing the conductivity28. While these two systems are different, the compositional similarity and the similar sharp increase in the conductivity suggest a similar underlying mechanism. Second, as evidenced in the synchrotron XRD data, Fig. 1a., the existence of highly conductive t-Na3PS4 crystalline phase25 may add a second highly conductive parallel phase, although less than 5% fraction, that gives rise to the overall higher conductivity.
Further additions of oxygen monotonically increase the activation energy thereby decreasing the conductivity. Consistent with our previous studies,28 the added oxygen collapses the structure thereby decreasing the free volume available for conduction and increasing the activation energy through the formation of further BO units. Further decreasing the conductivity is the formation of a very poorly conductive Na2S phase formed through the disproportionation reaction described above. As a result, the Na3PS3.4O0.6 SE shows a lower conductivity of 8.2×10− 5 S cm− 1 at 60°C. Furthermore, it has been recently reported that high electronic conductivity of SEs originating from internal defects (cracks, pores, grain boundaries) is responsible for Li dendrite formation and growth in SEs55,56. Figure 4b compares the electron conductivity of Na3PS3.4O0.6, Na3PS4 and HT-Na3PS4 measured by two ion-blocking electrodes at 60°C. Na3PS3.4O0.6 shows two orders of magnitude lower electronic conductivity than that of HT-Na3PS4, supporting the unique defect-free microstructure of oxysulfide SE which will undoubtedly reduce the likelihood of dendrite-induced short-circuit.
However, the superb formability of the Na3PS3.4O0.6 SE can provide a means to fabricate a thin SE layer, the resistance of which could be very low. An ideal SE for ASSSBs requires both high ionic conductivity and excellent mechanical and chemical stability. To create such an SE, therefore, a tri-layer architecture with the most ionically conductive Na3PS3.85O0.15 in the middle and the most mechanically and chemically stable Na3PS3.4O0.6 on the outside was designed and is shown in the inset of Fig. 4c. From the SEM images shown in Supplementary Fig. S13, an entirely all-glass SE separator without any discernible voids or grain boundaries was attained owing to the excellent formability and densification. The cyclability of the tri-layer electrolyte was studied in a symmetric Na|trilayer-SE|Na cell using gradient-current and constant-current tests, as shown in Fig. 4c and Fig. 4d–e, respectively. As seen in Fig. 4c, the voltage profile of the tri-layer separator shows an ohmic response, V = IR; however, sudden voltage spikes occur when the current exceeded 2.3 mA cm− 2, which was determined to be the critical current density (CCD) for the tri-layer SE. 2.3 mA cm− 2 is the highest reported CCD value for sulfide-based Na-ion SEs, which is also comparable to the state-of-art CCD value of Li-ion based sulfide SEs. The increased capability of the oxysulfide SEs to resist Na dendrite formation and penetration is attributed to their homogeneous and porosity-free structure and robust mechanical properties. These observations agree with Porz et al.’s11 conclusions on the failure mechanisms of SEs toward a metal anode where they observed that Griffith’s flaws in the surface of the SEs drive dendritic growth and propagation through SEs.
Figure 4d and 4e show that a symmetric cell with a tri-layer electrolyte can stably cycle for several hundred hours without short-circuiting at current densities of 0.2 mA cm− 2 and 0.5 mA cm− 2, respectively. The inserts to Figs. 4d and 4e show that very flat voltage profiles are observed at each cycle over the range of time that the cell was cycled. This behavior indicates that the Na plating and stripping processes are very stable at the Na3PS3.4O0.6–Na interface, consistent with the discussion above. As seen, the charge-transfer resistance for Na3PS3.4O0.6–Na is only ~ 10 Ω·cm2. Fig. S14 compares this work and some of the state of art reports on Na|SEs|Na symmetric cells for different categories of reported Na-ion conducting SEs. Three key parameters, which determine the energy density, power density, and cycling life of practical full cells are shown in this figure; the capacity of Na metal plated per cycle (area of the circle), current density, and cycling time. It is clear from this figure that the composite tri-layer oxysulfide SE developed in this work significantly extends the cycle time and current density performance for SEs cycling Na in symmetric cells. In particular, our composite tri-layer SE sets new standards for higher current density and longer cycling times in ASSSBs.
It has been shown above that the Na3PS3.4O0.6 SE can be combined with Na3PS3.85O0.15 to create a stable all-glass separator. In a similar manner, using the tri-layer architecture, Na3PS2.4O0.6 can be used for protecting other highly conducting but chemically unstable (against Na metal) SEs such as glass-ceramics, e.g. HT-Na3PS4, or ceramics, e.g. Na3SbS4. An example of this is shown in Supplementary Fig. S15 for Na3PS3.4O0.6|HT-Na3PS4| Na3PS3.4O0.6. Due to the ease at which the Na3PS3.4O0.6 SE deforms under pressure to form a homogeneous interface, excellent interfacial contact can be achieved between these two SEs. Furthermore, Fig. S15 shows that the Na|trilayer SE|Na symmetric cells also show very stable Na plating/stripping profiles at 0.2 mA cm− 2 for up to 500 h and up to 240 h at 0.5 mA cm− 2.
All Solid-State Na–S Full Cells. The excellent Na3PS2.4O0.6–Na stability enables the fabrication of ASSSBs, of which the ambient-temperature Na–S battery is one of the most promising because of its very low cost and high specific energy. On the basis of the above study that demonstrated the stability of the tri-layer SE in a Na metal symmetric cell, a Na–S battery having the architecture of S–Ketjen Black (KB)–Na3PS3.85O0.15|Na3PS3.85O0.15|Na3PS3.4O0.6|Na was designed and tested at 60°C. Notice here, in the full cell configuration, the second layer of Na3PS3.4O0.6 on the cathode side is not needed because CV measurements, see Fig. S16, shows that the higher conducting Na3PS3.85O0.15 SE is stable to the oxidizing potentials of 3.3 V for a sulfur cathode. For this reason, the SE separator could be simplified to a bi-layer structure of |Na3PS3.85O0.15|Na3PS3.4O0.6|, the slightly poorer conducting but very stable Na3PS3.4O0.6 SE was used to create a stable interface to Na metal.
Figure 5a shows the voltage profiles for the cell operated within the voltage window of 1.0 to 3.0 V. The cell shows a high initial discharge capacity of 1280 mAh g− 1, which is 76% of the theoretical capacity of sulfur (Na→Na2S: 1675 mAh g− 1) and much higher than that, 558 mAh g− 1, of the conventional high-temperature Na–S battery57. This is because the Na–S battery described here is an all-solid-state system that does not have the discharge limitation of Na→Na2S4 that the high-temperature Na–S battery exhibits. The first cycle coulombic efficiency is 92%, indicating that the polysulfide shuttle phenomenon commonly found in a liquid electrolyte-based cell is not observed in the current system. The cell then improves to ~ 100% coulombic efficiency after first initial five cycles. Upon further cycling, the reversible capacity stabilizes at about 1000 mAh g− 1 after 40 cycles with capacity retention more than 80% (Fig. 5b). These values are significantly better than those of the reported Na–S batteries using oxide or polymer SEs and Table S3 summarizes these previously reported solid-state Na–S batteries58–65 operated near ambient temperature in comparison to that reported here. The discharge voltage profile shows two plateaus located at 1.9 V and 1.25 V vs. Na+/Na with a single slope in between. The average discharging potential is 1.42 V, which is higher than those of other pure sulfide SE-based Na–S batteries that use Na15Sn4 or Na3Sb alloy anodes.
The rate capability of our Na–S battery was evaluated by examining the capacity by cycling at increasing current densities. As seen in the Supplementary Fig. S16, the cell can deliver specific capacities of 1116, 908 and 574 mAh g− 1 at current densities of 0.10, 0.20 and 0.35 mA cm− 2, respectively. After the current was returned to 0.10 mA cm− 2, the cell capacity also reverted to a value close to the original one and the cell cycled stably for another 150 cycles (Fig. 5c). The significantly improved Na–S battery cycling performance is attributed to the excellent interface stability, which enables Na metal to stably plate/strip at high rates, the superior formability property of oxysulfide SEs, which ensures the consistently good contact with sulfur and carbon during cycling, and the overall solid nature of the SEs that completely stops any polysulfide shuttle between the anode and the cathode. Therefore, the oxysulfide-based Na–S battery described here shows the highest specific energy density among all currently reported all-solid-state Na–S batteries (Fig. 5d). For these reasons, we believe these oxysulfide composite SEs may offer an entirely new and successful approach to the development of low cost, high energy density, safe, and high cycle life Na-based solid-state batteries.