Operando characterization of cathodic reactions in a liquid-state lithium-oxygen micro-battery by scanning transmission electron microscopy

Rechargeable non-aqueous lithium-oxygen batteries with a large theoretical capacity are emerging as a high-energy electrochemical device for sustainable energy strategy. Despite many efforts made to understand the fundamental Li-O2 electrochemistry, the kinetic process of cathodic reactions, associated with the formation and decomposition of a solid Li2O2 phase during charging and discharging, remains debate. Here we report direct visualization of the charge/discharge reactions on a gold cathode in a non-aqueous lithium-oxygen micro-battery using liquid-cell aberration-corrected scanning transmission electron microscopy (STEM) combining with synchronized electrochemical measurements. The real-time and real-space characterization by time-resolved STEM reveals the electrochemical correspondence of discharge/charge overpotentials to the nucleation, growth and decomposition of Li2O2 at a constant current density. The nano-scale operando observations would enrich our knowledge on the underlying reaction mechanisms of lithium-oxygen batteries during round-trip discharging and charging and shed lights on the strategies in improving the performances of lithium-oxygen batteries by tailoring the cathodic reactions.

diffraction 15 , ab initio calculations 16 , ex situ transmission electron microscopy (TEM) and scanning electron microscopy (SEM) 17 , and so on. However, the direct correspondence of Li 2 O 2 nucleation, growth and decomposition to electrochemical discharge/charge in the liquid-state Li-O 2 system has not been well established owing to the lack of operando characterization with real-space visualization. The rationale of the basic cathodic reaction kinetic process, related to those questions, is still in dispute. Therefore, the direct observations of the nucleation, growth and decomposition of Li 2 O 2 , accompanying with simultaneous electrochemical measurements of discharging/charging response, is highly desired to answer those fundamental questions and thus the reaction kinetics of the Li-O 2 batteries.
TEM with a high spatial resolution has been demonstrated as a powerful tool for in situ characterization of electrochemical and chemical reactions 12,[18][19][20] . In particular, the recently-developed micro-fabricated electrochemical cells (EC) provide a unique advantage for imaging electrochemical reactions while electrochemical measurements can be performed simultaneously. For lithium ion batteries, in situ EC TEM and scanning TEM (STEM) have been used to observe lithiation and delithiation dynamics in silicon nanowires and LiFePO 4 nanoparticle electrodes 21,22 . Although in situ TEM has been utilized to study the formation and/or decomposition of Li 2 O 2 in Li-O 2 batteries 12,23,24 , electrochemical processes of these studies were performed at a constant voltage and thus cannot provide the information on the electrochemical correspondence of the nucleation, growth and decomposition of Li 2 O 2 to discharge/charge overpotentials. Moreover, the weak contrast of Li 2 O 2 under TEM mode, especially interfered by liquid surroundings in EC, limits the spatial resolution for detecting the formation/decomposition of light Li 2 O 2 under non-equilibrium high-rate charge/discharge conditions. In contrast, the state-of-the-art spherical aberration corrected STEM with enhanced contrast by a high angle annular dark field (HAADF) detector is a promising technique for characterizing basic electrochemical reactions in liquid-state at a high spatial resolution of ~1 nm 19 . In this study we employ in situ EC HAADF-STEM to investigate the nucleation, growth and decomposition of Li 2 O 2 in a basic liquid-state Li-O 2 system during discharging and charging at a constant current density, together with synchronized electrochemical measurements. Figure 1a shows the schematic diagram of the Li-O 2 micro-battery for in situ STEM observations. A micro-channel was sandwiched between two micro-chips with ~30 nm thick electron-transparent Si 3 N 4 membranes to form a flowing liquid cell. The top chip was patterned with two Au pads as the electrodes for electrochemical measurements. One Au pad loaded with LiFePO 4 nanoparticles was used as the counter electrode. The formation and decomposition of Li 2 O 2 during discharge-charge cycles were recorded on the other Au pad which acted as the working electrode (Fig. 1b). 1.0 M LiClO 4 in dimethylsulfoxide (DMSO) saturated with oxygen was used as the electrolyte and fed into the cell at a low flowing rate of 1 μL/s through an external liquid pump. The flowing electrolyte keeps near constant oxygen and Li ion concentrations during electrochemical measurements and STEM observations at a very low flow velocity (smaller than 0.05 nm/s) of the electrolyte in the vicinity of the Au cathode, calculated by the finite element analysis (Fig. S2). The influence of the flowing electrolyte at the low velocity on the cathodic reaction kinetics is ignorable as verified by invisible mechnical motion of small Li 2 O 2 nanoparticles. The assembled liquid Li-O 2 micro-cell was loaded into a TEM holder and connected to an electrochemical workstation. To achieve high-quality images, a Cs-corrected STEM mode with a mass-sensitive HAADF detector was employed for the operando observations. The spatial resolution of the Cs-corrected STEM liquid cell was estimated to be ~1 nm 19,25 . Additionally, HAADF-STEM images were taken at a relatively lower resolution of 512 by 512 pixels and a short scanning time of 5 μs/pixel to avoid electron beam irradiation and sample drift and to capture more frames during observations. Under the imaging condition, the electron dosage is as low as 0.53 e − /Å 2 with low magnification of 50k, which is several orders of magnitude lower than that of the typical high-resolution Cs-corrected STEM (~10 5 e − /Å 2 ) 26,27 (Table S1). The experimental parameters of the low-dose STEM can be found in Supplementary Materials. To test the electrochemical performance of the Li-O 2 micro-battery inside TEM, the cyclic voltammetry (CV) measurements were performed as shown in Fig. 1c. The CV curves well trace the reversible cathodic reactions and are consistent with those of conventional Li-O 2 batteries 28,29 , demonstrating that the electrochemical functions of our Li-O 2 micro-battery are essentially identical to these of conventional coin batteries although their apparent configurations look different. A time-capacity dependent potential evolution curve during galvanostatic discharge and charge at a constant current density of 1.3 μA/cm 2 was recorded during HAADF-STEM observations (Fig. 1d). In spite of the slightly high charge/ discharge overpotentials due to large series resistance and complex configuration of the liquid cell setup 12,30 , the characteristic voltammetric profile of the micro-battery is consistent with those of conventional Au based Li-O 2 batteries 6,31,32 . Moreover, the galvanostatic charge/discharge profiles do not show obvious changes when the electron beam for STEM imaging turns on and off (Fig. S3). Moreover, we have verified that the electrolyte and reaction product Li 2 O 2 are stable under the low-dose STEM mode in the experimental time scale (Supplementary Materials). Therefore, the influence of electron beam irradiation on the cathodic reactions during our operando observations should not be observable. Under the applied galvanostatic discharge-charge cycles, the detailed electrochemical reactions on the Au cathodes of the Li-O 2 micro-battery can be monitored simultaneously by STEM.
The time sequential HAADF-STEM images during discharging are presented in Fig. 2a-h. To enhance the weak contrast of the reaction products of light Li 2 O 2 , color correction with blue, white and red is used. For reference, the original grey HAADF-STEM images are shown in Fig. S15. The regions with a bright white contrast represent the Li 2 O 2 phase while the very bright rim of the gold electrode in Fig. 2a may be from the edge effect of the electrode because it keeps unchanging with discharging and charging. During the first 10 seconds of the discharging reaction, a diffuse layer with continuous bright contrast emerges, as marked by white dashed line in Fig. 2b. Although discrete Li 2 O 2 particles cannot be identified from the diffuse layer by HAADF-STEM, correspondingly, the discharge overpotential reaches the discharge plateau with continuous increase of capacity. It is known that the solubility of Li 2 O 2 in DMSO is very low 7 . The absence of the Li 2 O 2 phase on the electrode surface and diffuse layer suggests that the discharge reaction is carried out by the one-electron reduction of O 2 with the formation of lithium superoxide (LiO 2 ) intermediates which have high solubility in DMSO. Since the Scientific RePORts | (2018) 8:3134 | DOI:10.1038/s41598-018-21503-w contrast of HAADF images arises from the high-angle electron scattering, the white contrast of the diffuse layer evidences the enhanced electron scattering from the electrolyte in the vicinity of charged electrode. Thus, the enhanced contrast of the diffuse layer most likely originates from the charged or polarized molecules of the electrolyte in electric fields, which may have higher structure ordering by polarized molecule alignment or higher density from reduced inter-molecule distances. This assumption is supported by the fact that the diffuse layer can also be observed around the Au electrodes with a pure DMSO electrolyte without the additives of O 2 and Li ions (Fig. S5). The thickness of the diffuse layer gradually increases with discharge time and reaches the maximum of about 80-90 nm before the precipitation of solid Li 2 O 2 phase (Fig. S4). When discharging for about 16 seconds, nanoparticles gradually become visible within the diffuse zone and preferentially start from the topmost surface of the gold electrode. Significantly, the precipitation of isolated nanoparticles in the electrolyte without direct contact with the electrode can also be observed, as marked by arrows in Fig. 2c and d. These observations indicate that the formation of solid Li 2 O 2 phase is most likely by a non-Faradaic disproportionation reaction lagging behind the one-electron ORR. The earlier precipitation of Li 2 O 2 in the vicinity of Au electrode surfaces is also in agreement with the reaction mechanism as the formation of Li 2 O 2 by the disproportionation reaction is expected to be controlled by LiO 2 diffusion. With continuously discharging, the thickness of the Li 2 O 2 layer keeps increasing from ~90 nm to ~160 nm mainly by the formation of new nanoparticles from the diffuse layer ( Fig. 2d-g). The newly-formed nanoparticles often connect with pre-existed ones and form a network like porous structure. After ~60 seconds, the discharge products forms a ~160 nm thick Li 2 O 2 layer on the Au cathode, which is composed of loosely interconnected nanoparticles with open pore channels (Fig. 2h).
The formation of Li 2 O 2 has been suggested to follow a two-step reaction, which includes the one-electron reduction of O 2 to form the lithium superoxide (LiO 2 ) intermediates (Eq. 1) and subsequently either a non-Faradaic disproportionation reaction (Eq. 2) or a second reduction reaction (Eq. 3) to form the final Li 2 O 2 discharge product 5,14 : The operando STEM observations provide compelling evidence that the formation of Li 2 O 2 is through a hybrid electrochemical and chemical process, i.e, the one-electron reduction of O 2 to LiO 2 by Eq. 1 and followed by the non-Faradaic disproportionation of LiO 2 with Eq. 2 under the current experimental conditions. The hybrid mechanism allows Li 2 O 2 nanoparticles to directly precipitate from electrolyte, rather than layer-by-layer growth on the electrode surface. The operando observation is also consistent with the reaction mechanism proposed by Bruce et al. The high-donor-number solvent DMSO with a high solubility of LiO 2 facilitates the formation of Li 2 O 2 through a solution pathway 28 . It has been reported that the disproportionation of LiO 2 to Li 2 O 2 has a half-life of tens of seconds 14 , close to the time scale of the discharge in our in situ STEM experiments. Therefore, the direct second electron reduction pathway by charge transfer (Eq. 3) appears unfavorable at a low constant discharging current. As shown in the time sequential HAADF-STEM images, the formation of Li 2 O 2 is dominated by the nucleation process in the initial stage. The soluble LiO 2 produced at the electrode surface diffuses into the electrolyte and forms Li 2 O 2 by the disproportionation reaction (Eq. 2). When the concentration of Li 2 O 2 exceeds the solubility limit, crystallites precipitate from the electrolyte as the initial Li 2 O 2 nuclei and gradually grow up during continuous discharging. The smallest Li 2 O 2 precipitates, which can be identified by the STEM observations, are about 3-4 nm (Fig. S6), which could be the upper bound of the critical size of Li 2 O 2 precipitates. The formation of the network structure with interconnected Li 2 O 2 particles may result from heterogeneous nucleation of Li 2 O 2 nanoparticles during which the pre-existed particles act as the nucleation sites to reduce the nucleation barrier of newly-formed Li 2 O 2 particles, or attachment growth of the nanoparticles. The open channels in the network structure enable the gold electrode to contact with electrolyte. In this way the LiO 2 intermediates can be continuously generated from the electrode by ORR at a nearly constant overpotential until the electrode is completely passivated by a solid Li 2 O 2 film which blocks the contact between the gold electrode and electrolyte (Fig. S13) 33 . The crystal structure and chemistry of the Li 2 O 2 phase formed during the operando observations are verified by separate selected area electron diffraction (SAED) and electron energy loss spectroscopy (EELS) analysis at a low acceleration voltage of 120 kV (Fig. S12).
The formation kinetics of Li 2 O 2 was investigated by measuring the evolution of the size and number of Li 2 O 2 particles during discharging. As shown in Fig. 3a and b, the average particle size via discharge time is parabolic while the particle number quickly increases within first 20 seconds and then remains nearly constant, indicating that the formation of solid Li 2 O 2 phase is via a nucleation-growth process. From the initial discharge plateau to the end of cut-off capacity, the mean particle size increases from ~10 to ~35 nm within ~60 seconds at the average growth rate of 0.4 nm/s. The low growth rate is in line with the low current density used in this study. We also estimated the total volume changes of Li 2 O 2 with discharge time by assuming the Li 2 O 2 nanoparticles have an approximate spherical shape (Fig. 3c). Since the liquid layer is very thin (~120 nm), the density of Li 2 O 2 particles is not very high and the contrast difference between the solid Li 2 O 2 particles and electrolyte is large, most Li 2 O 2 nanoparticles can be counted during the in situ observations, although a small number of nanoparticles may be missing because of the overlap along the electron beam direction in the early stage of discharging. Interestingly, except the first 10 second corresponding to the formation of the diffuse layer, the Li 2 O 2 volume has a linear relation with discharge time and thereby capacity, which is consistent with the smooth discharge curve at a constant current density (Fig. 1d). After the initial stage of Li 2 O 2 formation, the thickness of the Li 2 O 2 layer keeps nearly unchanged during continuous discharging (Fig. 3d) and the increase of the charge capacity is by the growth of individual Li 2 O 2 nanoparticles. The nearly constant Li 2 O 2 layer thickness could be associated with the relatively stable concentration gradient of LiO 2 around the electrode because the contour of the Li 2 O 2 layer shows the similarity with the local curvature of the Au electrode. Since Li 2 O 2 is almost insoluble in the non-aqueous DMSO electrolyte 7 , Li 2 O 2 precipitates once it forms by disproportionation of LiO 2 . The growth of Li 2 O 2 particles is more likely controlled by the discharge reaction rate, namely LiO 2 formation rate and discharge current density, after a stable LiO 2 concentration gradient is formed in the vicinity of the electrode. Both the linear relation of Li 2 O 2 volume vs. time and the parabolic curve between Li 2 O 2 particle size and time are in good agreement with the galvanostatic discharge mode which has a constant LiO 2 formation rate and thus Li 2 O 2 formation rate (Li 2 O 2 volume per unit time) under a near equilibrium condition.
To investigate the decomposition reaction of Li 2 O 2 , galvanostatic charging was conducted after the Li 2 O 2 formation (Fig. 1d). The corresponding evolution of Li 2 O 2 particles during charging is shown in Fig. 2i-p. Within the first 10 seconds, noticeable shrinkage of the Li 2 O 2 nanoparticles takes place at the Au electrode/Li 2 O 2 interface (Fig. 2i,j), indicating that the preferential decomposition of Li 2 O 2 takes place from those directly contacting with the electrode. The process is accompanied by a quick increase of the charge potential (Fig. 1d) with the gradual loss of the direct contact between Li 2 O 2 nanoparticles and the electrode. This observation is consistent with previous in situ TEM studies 12,23 and evidences that the decomposition of Li 2 O 2 starts from the electrode surface at lower charging potentials. The kinetics of Li 2 O 2 oxidation is apparently limited by the charge transport between the electrode and indirectly contacted Li 2 O 2 particles. With further charging, noticeable dissolution of Li 2 O 2 particles that reside away from the gold electrode (up to 160 nm) can be observed at high charge potentials ( Fig. 2k-n). This phenomenon is different from the previous observations that the decomposition of Li 2 O 2 is carried out by the collapse of Li 2 O 2 films to keep direct contact between Li 2 O 2 and electrode for interfacial oxidation operated at constant voltages and high current densities 23 . In our experiment, the charge and discharge are conducted at a constant current density of 1.3 μA/cm 2 , similar to the conventional battery testing, and the charge reaction is under a near electrochemical equilibrium state with a relatively lower reaction rate in comparison with that of ref. 23 . The Li 2 O 2 oxidation involves charge transfer, i.e.
during which considerable electron transport in the porous Li 2 O 2 film is required for the decomposition of solid Li 2 O 2 via OER during charging. It is known that bulk Li 2 O 2 is insulator and electrons cannot tunnel into Li 2 O 2 deeper than ∼5-10 nm 34 . However, if the Li 2 O 2 phase formed by ORR during discharging is an insulator, it is difficult to explain the operando observations of continuous oxidation of Li 2 O 2 nanoparticles far away from the electrode. Thus, one possible explanation is that the Li 2 O 2 nanoparticles formed by the solution precipitation may be electrically conductive, or at least Li 2 O 2 particle surfaces are metallic as theoretically predicted by Siegel et al. 16,35 . As the reaction product layer consists of inter-connected Li 2 O 2 particles, the particle surfaces and the interfaces could provide conductive paths for the decomposition of Li 2 O 2 at Li 2 O 2 /electrolyte interfaces. However, after partial dissolution, many Li 2 O 2 particles become isolated with the loss of the surface conductive pathway (Fig. 2k). Significantly, the operando observation reveals that the subsequent charging proceeds via continuous dissolution of the isolated Li 2 O 2 particles while the charge potential smoothly increases (Fig. 1d), showing similar behavior in the final charging stage of conventional Li-O 2 batteries. Therefore, the isolated residual Li 2 O 2 nanoparticles are not "flotsam" but contribute to the charge recovery by OER. Except the hypothesis of the conductive Li 2 O 2 , the operando observations can also be interpreted by a hybrid chemical and electrochemical process, i.e, the dissolution of Li 2 O 2 by the reverse non-Faradaic disproportionation reaction of Eq. 2: Li 2 O 2 + O 2 → 2LiO 2 , and OER by the one-electron oxidation of LiO 2 by reverse Eq. 1: LiO 2 →Li + + O 2 + e − . In this case, the charging kinetics is controlled by the non-Faradaic dissolution of Li 2 O 2 , which depends on the solubility of both LiO 2 and oxygen in the electrolyte. Although recent observations of a redox mediator assisted Li 2 O 2 decomposition provides an additional clue on the existence of non-Faradaic disproportionation reaction 36 , confirmation of this hypothesis still needs more electrochemical and theoretical evidences. After charging for 58 seconds, all the Li 2 O 2 particles completely disappear but leave behind a diffuse layer around the Au electrode (Fig. 2p), similar to the initial stage of discharging. The much stronger intensity of the diffuse layer in Fig. 2p than that in Fig. 2a suggests that the formation and distribution of the diffuse layer are associated with the applied potentials of the electrode. It is worth noting that after Li 2 O 2 nanoparticles completely dissolve at 58 seconds (Fig. 2p) the charging continues until 60 seconds (Fig. 1d). The short charging time gap could be caused by that the remained Li 2 O 2 nanoparticles are too small to be seen from the low magnification STEM images, the oxidation process of Li 2 O 2 takes place at other location out of the observation area, or the dissolved LiO 2 in the electrolyte contributes to the continuous charging by the one-electron oxidation reaction. The measurements of Li 2 O 2 evolution during charging reveal that the structural evolution of these particles is almost the inverse process of Li 2 O 2 formation in discharging with the nearly symmetric plots shown in Fig. 3, although the reaction kinetics and reaction pathways for discharging and charging are very different. On the basis of the operando observations (Movie S1), the ORR and OER processes under round-trip discharging/charging cycles at a constant current density are schematically illustrated in Fig. S16. We noticed that Luo and his co-workers very recently reported in situ observations of cathodic reactions in a solid-state Li-O 2 battery using environmental transmission electron microscopy 24 . This work reveals new insights Scientific RePORts | (2018) 8:3134 | DOI:10.1038/s41598-018-21503-w into oxygen reduction and oxygen evolution reactions in the solid-state Li-O 2 system. However, it is fundamentally different from our work on the cathodic reactions in the liquid-state Li-O 2 battery. Electrochemically, the utilization of different electrolytes (liquid versus solid) has rendered them as essentially different systems in the following aspects: (1) the chemistry of the discharge products is different between the two observations. In the study of Luo  Although it was claimed to be oxygen-rich environment, the pressure is far below the atmospheric pressure of a practical solid-state Li-O 2 battery. (2) In our operando experiments, the charging and discharging were conducted at galvanostatic mode with a constant current density of 1.3 μA/cm 2 , similar to the coin battery testing. In contrast, in situ study in the solid-state cells by Luo et al. was discharged or charged by applying constant potential biasing and therefore the electrode reactions occurred at a far-from-equilibrium state. Therefore, our experimental setup better resembles the condition of practical batteries and hence the observed phenomena in our operando study would be more translatable for normal real-world batteries. Therefore, these two works investigate quite different systems and, we believe, both are of great scientific significance and would contribute to better understandings of the Li-O 2 electrochemistry. In particular, the comparison between these two studies would provide insights to how the state of electrolytes affects the fundamental electrochemical reactions and the operation of Li-O 2 batteries.
In summary, we have developed a liquid-state Li-O 2 micro-battery for operando characterization of fundamental cathodic reactions under aberration-corrected STEM (STEM) synchronized with electrochemical measurements. The direct observations of Li 2 O 2 nucleation, growth and decomposition at known discharge/charge overpotentials shine lights on the underlying mechanisms of cathodic reactions at a constant current density. The formation of porous Li 2 O 2 layer through solution precipitation by the non-Faradaic disproportionation reaction of intermediate LiO 2 is critical to keep a constant or near constant discharge potential for a large capacity. The dissolution of resultant Li 2 O 2 nanoparticles during charging is mainly contributed by the Li 2 O 2 /electrode interfacial oxidation at the early stage of the charging reaction. We propose that the hybrid chemical and electrochemical process would be responsible for the oxidation of Li 2 O 2 particles away from the electrode surfaces. Inspired by the operando observations, we have developed a full performance nanoporous graphene based Li-O 2 battery by utilizing stable high-donor-number organic solvent DMSO as the electrolyte to enhance the solution mechanism of Li 2 O 2 formation, the large surface area of nanoporous graphene for multiple heterogeneous nucleation/growth of small and porous Li 2 O 2 nanoparticles, and an oxidation redox mediator to effectively reduce the charge overpotential by improving the charge transfer between insulating Li 2 O 2 and electrode 37 .