A lithium–oxygen battery based on lithium superoxide

Batteries based on sodium superoxide and on potassium superoxide have recently been reported. However, there have been no reports of a battery based on lithium superoxide (LiO2), despite much research into the lithium–oxygen (Li–O2) battery because of its potential high energy density. Several studies of Li–O2 batteries have found evidence of LiO2 being formed as one component of the discharge product along with lithium peroxide (Li2O2). In addition, theoretical calculations have indicated that some forms of LiO2 may have a long lifetime. These studies also suggest that it might be possible to form LiO2 alone for use in a battery. However, solid LiO2 has been difficult to synthesize in pure form because it is thermodynamically unstable with respect to disproportionation, giving Li2O2 (refs 19, 20). Here we show that crystalline LiO2 can be stabilized in a Li–O2 battery by using a suitable graphene-based cathode. Various characterization techniques reveal no evidence for the presence of Li2O2. A novel templating growth mechanism involving the use of iridium nanoparticles on the cathode surface may be responsible for the growth of crystalline LiO2. Our results demonstrate that the LiO2 formed in the Li–O2 battery is stable enough for the battery to be repeatedly charged and discharged with a very low charge potential (about 3.2 volts). We anticipate that this discovery will lead to methods of synthesizing and stabilizing LiO2, which could open the way to high-energy-density batteries based on LiO2 as well as to other possible uses of this compound, such as oxygen storage.

. These studies also suggest that it might be possible to form LiO 2 alone for use in a battery. However, solid LiO 2 has been difficult to synthesize in pure form 18

because it is thermodynamically unstable with respect to disproportionation, giving Li 2 O 2 (refs 19, 20). Here we show that crystalline LiO 2 can be stabilized in a Li-O 2 battery by using a suitable graphene-based cathode. Various characterization techniques reveal no evidence for the presence of Li 2 O 2 . A novel templating growth mechanism
involving the use of iridium nanoparticles on the cathode surface may be responsible for the growth of crystalline LiO 2 . Our results demonstrate that the LiO 2 formed in the Li-O 2 battery is stable enough for the battery to be repeatedly charged and discharged with a very low charge potential (about 3.2 volts). We anticipate that this discovery will lead to methods of synthesizing and stabilizing LiO 2 , which could open the way to high-energy-density batteries based on LiO 2 as well as to other possible uses of this compound, such as oxygen storage.
The crystalline LiO 2 reported here was made electrochemically using a cathode based on reduced graphene oxide (rGO) with added iridium (Ir) nanoparticles. Initially graphene oxide (GO) was prepared by a modified Hummer's method 21,22 . The Ir-rGO composite was then made by a hydrothermal reduction method and characterized ( Supplementary Fig. 2). Scanning electron microscopy (SEM) images of the pristine rGO and Ir-rGO composite ( Fig. 1a and b, respectively) reveal porous three-dimensional (3D) networks of rGO composed of wrinkled 2D rGO sheets. Figure 1c and d shows transmission electron microscopy (TEM) images of the Ir nanoparticles on rGO, indicating that the well-dispersed Ir nanoparticles decorated on rGO are very small (<2 nm), with evidence for the presence of some small Ir clusters (circled in Fig. 1d). A backscattering image ( Supplementary  Fig. 1) shows some scattered larger Ir particles of about 500 nm in size, which may be due to agglomeration of the smaller nanoparticles, and fast Fourier transform analysis of high-resolution (HR)-TEM images ( Supplementary Fig. 1) show that the nanoparticles are Ir. An X-ray photoemission spectroscopy (XPS) analysis ( Supplementary Fig. 1) indicates the Ir surface is only partially oxidized.
The performance of the rGO and Ir-rGO cathodes was examined using a Swagelok-type cell composed of a lithium metal anode, electrolyte (1 M LiCF 3 SO 3 in tetraethylene glycol dimethyl ether (TEGDME)) impregnated into a glass fibre separator, and a porous cathode. A current density of 100 mA g −1 was used for both discharge and charge, and the cell was run with a capacity limit of 1,000 mA h g −1 to avoid side reactions. The specific capacity (mA h g −1 ) and the current density (mA g −1 ) are based on the active materials of the O 2 electrodes. Figure 2a and b shows voltage profiles for the Ir-rGO and rGO cathode architectures, respectively. The Ir-rGO discharge product shows a very low charge potential of ~3.2 V that rises to 3.5 V over 40 cycles leading to more than 85% efficiency in this system (Fig. 2a). The voltage profile of the rGO cathode shows a much larger charge potential of ~4.2 V with a lower efficiency of ~67% (Fig. 2b).
The discharge product resulting from the Ir-rGO cathode was examined using SEM, differential electrochemical mass spectroscopy (DEMS), high-energy X-ray diffraction (HE-XRD), TEM and Raman spectroscopy with the results shown in Figs 2 and 3. The SEM image in Fig. 2c shows the Ir-rGO cathode after discharge (~2.75 V) from the first cycle (1,000 mA h g −1 capacity). This image indicates that the discharge product resulting from the Ir-rGO-based cathode consists mainly of nanoparticles with needle-or rod-like morphology, although the presence of other shapes such as cubic cannot be ruled out. This needle-or rod-like morphology is also observed in the TEM image of a part of the discharge product, which appears to be on the surface of the Ir-rGO nanostructures (Fig. 2c inset). An SEM image after charging shows that the nanoparticles have disappeared ( Supplementary  Fig. 3). The discharge product from the rGO-based cathode has a range of morphologies, including toroids and nanoparticles ( Supplementary  Fig. 3). The Ir-rGO discharge product produced by a current density of 100 mA g −1 was characterized by DEMS during the first charging cycle by on-line monitoring of the number of evolved O 2 molecules. The experiment was performed using high current densities (1,000 mA g −1 and 640 mA g −1 ) for charging to enable measurement of the evolved O 2 . The DEMS results at the higher current density are shown in Fig. 2d. Analysis of the data in Fig. 2d gives an average O 2 formation rate of 1.3 × 10 −9 mol s −1 , resulting in an e − /O 2 ratio of 1.00. A similar (1.00) e − /O 2 ratio was also obtained for the experiment with a current density of 640 mA g −1 (Supplementary Fig. 18). Additionally, negligible amounts of CO 2 and H 2 gases were generated during the DEMS experiments ( Fig. 2d and Supplementary Fig. 18). A DEMS experiment was also carried out during discharge, and gave an e − /O 2 ratio of 1.02 ( Supplementary Fig. 18, Supplementary Table 3). These results are consistent with LiO 2 as the main discharge product, and provide evidence for the absence of other products (for example, Li 2 O 2 , LiOH and Li 2 CO 3 ). The DEMS results for LiO 2 are similar to DEMS measurements on a NaO 2 battery that gave an e − /O 2 ratio of 1.00 for discharge and 1.02 for charge 2 .

Letter reSeArCH
The HE-XRD pattern in Fig. 3a for the discharge product on the Ir-rGO cathode (1,000 mA h g −1 capacity) during the first cycle shows peaks corresponding to crystalline LiO 2 ((101), (111), (120)), and no evidence for peaks corresponding to Li 2 O 2 . The identification of the LiO 2 peaks is based on a theoretical XRD pattern derived from the DFT (density functional theory)-predicted crystalline LiO 2 structure ( Supplementary Fig 4) from refs 19 and 23, as no experimental XRD pattern has been reported. The LiO 2 structure is orthorhombic ( Supplementary Fig. 4). For comparison, NaO 2 is cubic at room temperature and orthorhombic at <196 K, whereas KO 2 is tetragonal at room temperature. Some amorphous LiO 2 cannot be ruled out on the basis of the XRD results. The standard XRD pattern of Li 2 O 2 was used to determine the absence of Li 2 O 2 . The Raman spectra of the discharge product of the Ir-rGO cathode in Fig. 3b show the presence of a peak at 1,123 cm −1 , consistent with the range of values that have been observed for superoxide stretching frequencies (Supplementary Table 1). It is also consistent with the Raman peak at 1,156 cm -1 observed 1 for NaO 2 . There is also a peak at 1,505 cm −1 that has recently been attributed to the strong interaction between LiO 2 and a graphitic carbon surface 9 . In contrast, the HE-XRD pattern ( Supplementary Fig. 3) for the discharge product on the rGO cathode without Ir added (1,000 mA h g −1 capacity) during the first cycle shows peaks corresponding to both crystalline LiO 2 ((101), (111), (120)) and Li 2 O 2 ((101), (102), (103), (110)). Evidence for both LiO 2 and Li 2 O 2 components in the discharge product has also been reported in other studies [9][10][11][12][13][14][15] , although none are based on XRD characterization, which is made possible by the use of high-energy X-rays at the Advanced Photon Source of Argonne National Laboratory. When the Ir-rGO cell is run to deep discharge of 2.2 V and ~9,500 mA h g −1 capacity, the HE-XRD data shows evidence for the presence of LiO 2 , Li 2 O 2 and LiOH with a toroidal morphology ( Supplementary Fig. 5).We also note that there have been some previous studies 24,25 on rGO and rGO with Au nanoparticles that showed formation of Li 2 O 2 , but no report of LiO 2 in a Li-O 2 cell.
Further evidence that the discharge product is LiO 2 on the Ir-rGO cathode was obtained by an experiment in which Li was electrochemically added to the discharge product without the presence of O 2 (that is, O 2 was replaced by Ar). The voltage profile is shown in Fig. 3c for this discharge process, along with that of the initial discharge process (to 1,000 mA h −1 g). The HE-XRD of the resulting product with no O 2 in the cell is shown in Fig. 3d and reveals strong peaks from Li 2 O 2 , thus indicating a conversion of LiO 2 to Li 2 O 2 (Li + + e − + LiO 2 → Li 2 O 2 ) with ~96% of the theoretical capacity for this reaction attained. This is evidence for a reaction involving one electron per O 2 in the first cycle for 1,000 mA h g −1 capacity, and for no crystalline or amorphous Li 2 O 2 forming on the initial capacity-limited discharge. In contrast, no significant capacity for discharge in Ar is observed for the rGO cathode when a similar procedure is performed. We have also carried   Letter reSeArCH out electron paramagnetic resonance (EPR) measurements, and find a signal that is consistent with the presence of LiO 2 ( Supplementary  Fig. 7, Supplementary Table 2) in the initial discharge product.
The stability of the LiO 2 was investigated by carrying out HE-XRD measurements on Ir-rGO cathodes aged for different times in the presence of the electrolyte (Fig. 3a). After 12 hours at the end of both the first and second discharges the XRD patterns show only evidence for crystalline LiO 2 . When the discharge product is allowed to remain for seven days under the same conditions, both samples still show the signature of LiO 2 along with the presence of crystalline Li 2 O 2 . Thus, the HE-XRD measurements indicate that the crystalline LiO 2 formed with the Ir-rGO cathode is surprisingly stable for a relatively long period of time. Raman spectra ( Supplementary Fig. 8) measured as a function of time exhibit a decreasing intensity of the peak at 1,123 cm −1 as a function of time, which is consistent with the HE-XRD results. In addition, LiO 2 is still the dominant discharge product on the twentieth discharge cycle, indicating that LiO 2 is stable enough that it can be repeatedly charged and discharged for about 40 cycles with a very low charge potential (~3.2 V); this may open the way to a lithium-superoxide-based battery.
An explanation of the formation mechanism and stability of the LiO 2 found in this study requires an understanding of the growth and nucleation process, which is quite complex and beyond the scope of the present study. However, our results are consistent with mechanisms proposed in other recent studies. Some insight can be obtained from a postulated mechanism for the nucleation and growth of discharge products from various size-specific Ag clusters decorating carbon cathodes in a Li-O 2 cell 7 . In that case, the results are explained by a through-solution growth mechanism with sites for oxygen reduction reactions (ORRs) that are separate from the nucleation sites. Other researchers have also reported evidence for through-solution mechanisms [26][27][28][29] . In addition, the results of the Ag cluster study 7 suggest that availability of good ORR sites promotes a through-solution mechanism involving nucleation and growth of LiO 2 followed by disproportionation to Li 2 O 2 . Since both Ir and rGO are good ORR materials 30,31 , the Ir-rGO-based cathodes of the present study should result in a similar mechanism, that is, initial formation of LiO 2 .
The HE-XRD finding that rGO results in both LiO 2 and Li 2 O 2 in the discharge product can be accounted for by slow disproportionation. This explanation is supported by a recent theoretical study 17

Letter reSeArCH
shown the possibility of fast and slow disproportionation processes for LiO 2 depending on the LiO 2 cluster size. In addition, experimental evidence for slow disproportionation has been found in studies of Li-O 2 cell discharge products based on an activated carbon cathode 10 .
In contrast, the present HE-XRD results showing that only LiO 2 is present in the case of the Ir-rGO-based cathode material suggest that disproportionation is suppressed. The formation of only LiO 2 in the case of the present Ir-rGO cathode may be due to some aspect of that cathode that favours nucleation and growth of largely the crystalline LiO 2 phase, which prevents disproportionation (see theoretical calculations below). In the case of Ir-rGO, we noticed the formation of an Ir 3 Li intermetallic compound on the large Ir agglomerates seen in the backscattering image ( Supplementary Fig. 1), as shown Fig. 3f and g). We also noticed that some nanoparticles (needle or rod-like) are formed on the surface of these agglomerates during the first discharge (Fig. 3e). We note that the Ir 3 Li intermetallic compound has an orthorhombic lattice, a similar crystallographic lattice 32 to that of LiO 2 . It is possible that it may act as a template for growth of the crystalline LiO 2 , as has been found in template-controlled nucleation and growth of other crystalline materials 33 . We carried out DFT calculations on the interface between LiO 2 and Ir 3 Li, and found that some crystalline faces had good lattice matches ( Supplementary Fig. 13), as would be required for epitaxial growth of crystalline LiO 2 . There may be other intermetallic compounds that exhibit similar behaviour and will be the subject of further study. The schematic in Fig. 4 summarizes the novel templating process that may be responsible for the LiO 2 discharge product found for the Ir-rGO cathode material and the subsequent electrochemical reactions that it can undergo, that is, either further lithiation or further oxygen reduction. In contrast, the rGO cathode, which does not include Ir nanoparticles, probably has a different nucleation and growth mechanism resulting in a discharge product composed of both LiO 2 and Li 2 O 2 .
The kinetic stability of crystalline and amorphous LiO 2 was investigated using ab initio molecular dynamics (AIMD) and DFT calculations with the results shown in Fig. 4. The disproportionation rate will depend on several factors. One factor is the rate at which the O 2 leaves the surface. The DFT results in Fig. 4b indicate that the initial step of O 2 leaving the crystalline surface into vacuum has a barrier of ~0.9 eV based on a low-energy LiO 2 surface. Figure 4c shows that crystalline LiO 2 surfaces (that is, (101) and (111)) are thermally stable in vacuum at room temperature. For an amorphous surface, the barrier (~0.3 eV) is less than for the crystalline surface (Fig. 4d). From AIMD simulations, the presence of some solvent molecules adsorbed on the amorphous LiO 2 surface reduces O 2 desorption ( Supplementary  Fig. 14). This suggests that solvent on the LiO 2 surface could further suppress disproportionation of the crystalline phase. The electrolyte effect on disproportionation was investigated by allowing a sample from a 1,000 mA h g −1 discharge to age for 24 h in vacuum. Characterization of the sample by Raman spectroscopy, discharge in Ar, and charge potential shows a significant decrease of LiO 2 signature after ageing in vacuum, indicating that kinetics plays an important role in stabilizing the LiO 2 ( Supplementary Fig. 15) The Ir-rGO cathode also exhibits a low charge potential, which may be due to several factors. As shown in Fig. 4e, crystalline LiO 2 is a half-metal (on the basis of density functional calculations) and, thus, will have good electronic conduction, in contrast to insulating bulk Li 2 O 2 . Another factor is that Ir is known to be a good oxygen evolution catalyst 31 and interacts strongly with LiO 2 to form a good interface for electrical contact. These properties may explain why the discharge product formed on just rGO has a large charge potential, that is, it lacks the Ir nanoparticles. The Li-O 2 cell based on Ir-rGO cathode material also can cycle 40 or more times (Supplementary Fig. 9) before failure, similar to what has been found for Li 2 O 2 -based Li-O 2 cells, indicating that the lithium superoxide is not any more reactive towards the electrolyte than lithium peroxide. In addition, the low charge potential will lead to less side reactions. The failure of the cell could be due to oxygen crossover to the anode resulting in the anode being converted to LiOH, as evidenced by the corrosion of the anode ( Supplementary  Fig. 10) and, possibly the poisoning of Ir metal catalyst with cycling. When the cycled Li anode is replaced by a new anode, the cell cycles another 30 times (Supplementary Fig. 10).
The evidence presented here indicates that a Li-O 2 electrochemical cell based on a LiO 2 discharge product is possible with a reasonable cycle life, very high efficiency, and a good capacity. The performance characteristics of the cell based on LiO 2 are comparable to those of previously reported electrochemical cells based on KO 2 (ref. 3) and on NaO 2 (ref. 1), although some aspects-such as the charge and discharge potentials-differ. Problems with electrolyte stability and decomposition, as for the electrolytes used for other Li-O 2 systems, probably still remain, but they do not seem any worse than for those systems. There is little evidence of any side reactions in the Raman data for the first discharge cycle (Fig. 3b), or from Raman and Fourier transform infrared data after charging for up to 30 cycles ( Supplementary  Figs 11, 12), or from NMR data up to 20 cycles ( Supplementary Fig.  12), although there could be decomposition products that are not detected. The Fourier transform infrared and Raman results also confirm that the discharge product is not present after charging.
In summary, we have reported evidence it is possible to have a one-electron discharge process that forms only LiO 2 in a Li-O 2 electrochemical cell. This is different from the previous studies 9-15 that have provided evidence for both LiO 2 and Li 2 O 2 in the discharge product of Li-O 2 batteries with some cathode and electrolyte materials, and from studies 16 that have shown LiO 2 can be present in solution during discharge. The evidence for the existence of the LiO 2 comes from DEMS and HE-XRD data with no evidence for Li 2 O 2 being present. The results of TEM and density functional calculations indicate that a novel templating growth mechanism involving the use of Ir nanoparticles may be responsible for the crystalline LiO 2 growth. The LiO 2 formed in this way is stable enough to be repeatedly charged and discharged with a very low charge overpotential.
Online Content Methods, along with any additional Extended Data display items and Source Data, are available in the online version of the paper; references unique to these sections appear only in the online paper.