Principle of operando NDP Li-metal
NDP of Li relies on the capture reaction of a thermal neutron with a 6Li atom, which results in two charged particles that, based on conservation of momentum and energy, have a well-defined energy: 4He (2044 keV) and 3H (2727 keV). By placing a detector at some distance from a Li-containing battery as shown in Fig. 1a, the energy loss of these charged particles can be related to the depth of the capture reaction. Hence, the original position of the 6Li can be determined through the stopping power of the material. Because each measured 4He and 3H particle represents one Li in the system of interest, NDP measures the Li number density with great accuracy as a function of depth. Based on this ability NDP has been used to monitor Li distributions in thin film solid state micro-batteries52, in Sn and Al electrodes51,53, in LiFePO4 electrodes54,55, and Li-metal plating at the interface of a garnet solid electrolyte56.
In the present research, operando NDP is used to investigate Li plating and stripping on a Cu current collector in symmetric Li-metal cells employing a LiPF6 ethylene carbonatedimethyl carbonate (EC/DMC) liquid electrolyte. Because natural Li consists of only 7.5% of the 6Li isotope, enriched Li-metal was used (95% 6Li and 5% 7Li), which is assumed not to influence the metal and plating reactions. By bringing the electrolyte in contact with excess of enriched Li-metal, also the electrolyte is enriched up to 95% 6Li and 5% 7Li. Because the 4He particles are not able to pass the Cu current collector, only the 3H particles are detected during the present experiments. The depth as shown in Fig. 1b is measured starting from the interface of the Cu current collector with the ambient atmosphere. The interface of the Cu current collector with the interior of the cell is observed by the appearance of Li, and is positioned at ~ 12 μm, reflecting the Cu current collector thickness. The 12 μm thick Cu foil results in a loss of 1040 keV of the initial 2727 keV of the 3H. In Fig. 1b operando NDP is shown for four subsequent electrochemical plating and stripping cycles at 1 mA cm–2 up to a plating capacity of 1 mAh cm–2 during which continuously 1 min NDP measurements are acquired. During plating the detected Li density on the 12 μm thick Cu current collector reflects the anticipated appearance of Li, whereas the Li density decreases during the stripping current as anticipated. The background and intensity correction as described in the methods sections result in the quantitative measurement of the Li concentration as a function of depth. Figure 1b shows the Li density normalized to Li-metal; hence, it reflects the fractional density with respect to Li-metal. The maximum fractional density is ~ 0.3, which, assuming this all to be Li-metal, results in a minimum porosity of the deposited Li of ~ 0.7 (70%). This demonstrates that plating results in a highly porous Li-metal film, consistent with the plated capacity, 1 mAh cm–2, which corresponds to a solid Li-metal film thickness of ~ 5 μm, in this case distributed over at least a 20 μm thick layer as observed in Fig. 1b. From the evolution of the Li density, the growth rate can be determined, amounting 0.23 μm min–1 for 0.1 fractional density and 0.11 μm min–1 for 0.2 fractional density. Very little quantitative operando information exists on the Li density and porosity of electrochemically plated Li, illustrating the added value of NDP for the characterization of Li-metal-plating reactions. The depth resolution for these systems is ~ 70 nm, which is dictated by the stopping power of the materials along the path of the 3H between the Li position and the detector. Therefore, NDP results in the Li density, and approximate porosity of the deposited Li-metal with a relatively good spatial depth resolution.
Impact of current density and salt concentration
The Li density profiles in Fig. 1b show that repeated cycling leads to less-dense, and thicker Li deposits. The low-density tails of these profiles indicate highly porous Li morphologies that extend far into the electrolyte. Because Li-metal reduces the EC/DMC electrolyte57,58, resulting in a heterogeneous mixture of inorganic compounds (Li2O and Li2CO3 for carbonate electrolytes) and organic polymers16,59, the observed Li density is the sum of Li-metal and Li in the SEI at the surface of the Li-metal. NDP is unable to distinguish the chemical nature of the observed Li. For comparison with the NDP experiments, ex situ SEM measurements were performed that show the micro structured morphologies of the deposited Li-metal in Fig. 2a–h at different deposition capacities. Similar to other investigations, at 0.1 mAh cm–2 and at 0.2 mAh cm–2 Li deposit “hotspots” appear. At a capacity of 0.5 mAh cm–2 elongated Li deposits as well as mossy Li structures are formed and at 1.0 mAh cm–2 ~ 200 nm wide needle-like tips are observed. Figure 2f shows that after stripping, a porous microstructure remains. The tilted SEM image at 1 mAh cm–2, Fig. 2g, clearly shows a porous morphology including dendrites extending ~ 13 μm from the Cu current collector into the electrolyte. This is in good agreement with the average thickness of the Li density observed with operando NDP, where the first plating and stripping cycle at mAh cm–2 is displayed in Fig. 2i. The dense Li region observed by NDP in Fig. 1b corresponds to the more dense mossy Li deposits observed by SEM, whereas the low-density tails extending far into the electrolyte corresponds to the dendrites observed by SEM. After stripping, the SEM shown in Fig. 2f displays difficult to characterize morphologies, most likely representing a mixture of SEI species and “dead” Li-metal. The evolution of the inactive Li, both “dead” Li-metal and Li in the SEI during cycling as observed by NDP is discussed in more detail in the next section.
From the 2D Li density in Fig. 2i, the plating and stripping process appears to be asymmetric. This can be more clearly visualized by subtracting subsequent Li densities in time, resulting in the change in Li density shown in Fig. 2j. This provides direct insight in the distribution of the plating and stripping, resulting in a positive change in Li density during plating (yellow/red), and a negative change in Li density during stripping (blue). During plating at 1 mA cm–2 Li deposition is localized within 5 μm in depth, the position of which progressively moves away from the Cu current collector into the electrolyte. This is consistent with root growth of mossy Li pushing the moss progressively into the electrolyte, as observed by optical analysis11,12,13,14,18. The localized growth indicates that at these conditions the root growth dominates, as compared with the thickening of the mossy whiskers, because thickening would result in a homogeneous Li-plating activity. In contrast, the stripping activity is more homogeneously distributed over the thickness of the deposited Li-metal film. This indicates thinning of the mossy Li whiskers throughout the deposited Li-metal film. Thinning is likely to result in regions of Li-metal that are disconnected from the Cu current collector, so called “dead” Lithium, which lowers the capacity upon cycling. Thereby, the average plating and stripping activity observed by operando NDP provides a very direct and operando view on the average growth and stripping mechanism, indicating that the fundamental origin of “dead” Li is the homogeneous stripping activity throughout the depth.
The local ion concentration is considered to be one of the key parameters that determine the growth mechanism and Li-metal morphology under plating conditions. As a consequence, the Li-metal morphology will depend on the current density. At current densities that lead to ion depletion at the metal surface, inhomogeneous Li-metal deposition in the form of dendrite formation is induced17,18,19. Although this prediction appears to works well above the critical current, dendrites are also observed at much lower current densities such as studied at present11,12,13,14, and also observed in Fig. 2a–h. How the Li-metal mossy/dendrite density distribution depends on the current density and the Li salt concentration can be directly assessed by operando NDP. More dendritic Li-metal growth may be expected to result in less-dense Li-metal films extending further away from the current collector as suggested by optical studies11,12,13,14.
In Fig. 3, operando NDP is shown during the first plating/stripping cycle at both 0.5 and 2 mA cm–2 for 2 and 0.5 h, respectively, (both resulting in 1 mAh cm–2 plated capacity). This reveals that a larger current density results in more compact Li-metal plating, which is difficult to assess by comparing the SEM images in Fig. 2g, h. For mossy/dendritic features this result is hard to validate with for instance SEM or optical techniques, and at first appears to contradict that larger currents are expected to result in more dendritic growth, and hence less-dense Li-metal films. The present operando NDP indicates that there is a strong relationship between the current and the resulting Li-metal density, even at these relatively low current densities. The plating and stripping activity shown in Fig. 3 demonstrates that plating occurs much more localized and closer to the current collector at 2 mA cm–2 compared with 0.5 mA cm–2 current density. This may be explained by current density dependent Li-metal nucleation behavior. Pei et al.15 have shown that the Li nuclei size in ether electrolytes is inversely proportional to the overpotential and that the number of Li nuclei is proportional to the cubic power of the overpotential, following classic nucleation theory. The much higher overpotentials at 2 mA cm–2, see Fig. 3, can therefore expected to result in denser plating at larger current densities based on a much larger amount of Li-metal growth centers, in particular in the vicinity of the Cu current collector. Another aspect is the SEI, formation of which will be more severe at larger overpotentials, that will influence the local properties at the current collector. Given the much less-dense dendritic growth at larger current densities, these findings suggest the existence of an optimal current, resulting in the largest plated metal density, motivating further systematic studies on the relationship between the Li-metal morphology and the current density.
As ion depletion after Sand’s time, at current densities that exceed the critical current density, has been shown to initiate Li-metal dendrite growth17,18,19, several strategies to prevent Li dendrite formation aim at avoiding ion depletion11,12,13,14. These strategies include raising the Li salt concentration in the electrolyte, inducing high transference numbers by solid electrolytes and the addition of inactive ions11,12,13,14. To investigate the influence of the LiPF6 concentration in the EC/DMC electrolyte, the molarity was increased from 1 to 2 molar. Figure 4 shows the comparison of the Li density between 1 and 2 molar LiPF6 electrolytes from NDP at different plating capacities and after stripping at 1 mA cm–2 during the first plating/stripping cycle. Even the small increase in concentration at this relatively small current density significantly increases the plating density resulting in a thinner deposited layer. Hence, from Fig. 4 we conclude that the salt concentration may not only reduce dendrite formation53, but also results in more compact mossy Li-metal plating. Concentrated electrolytes have been shown to reduce the thickness of the SEI layer60, however, how the density or porosity of Li-metal films quantitatively depends on the electrolyte concentration has not been reported.
Evolution of the total amount of Li during cycling
Mossy growth of metallic anodes needs to be avoided to minimize the amount of electrolyte interface, resulting in electrolyte decomposition and the formation of “dead Li”. Dead Li represents Li-metal that has no electric contact to the current collector, which is anticipated to remain on its original position fixed by the SEI that has formed around it. NDP allows monitoring the capacity loss owing to SEI formation as well as “dead” Li-metal formation over repeated cycling. By integrating the Li density profiles obtained by the operando NDP, the evolution of the total amount of Li can be obtained, which allows, for instance, to monitor the amount of inactive Li after each cycle, quantified by the Li mass after stripping. To investigate the influence of current density and cycling history, two operando NDP cycling experiments were performed, shown in Fig. 5a, one starting with five cycles at 0.5 mA cm–2 followed by five cycles at 2 mA cm–2 and the other experiment starting with five cycles at 2.0 mA cm–2 followed by five cycles at 0.5 mA cm–2 all up to a 1 mAh cm–2 plating capacity. In Fig. 5b, the integrated amount of Li in these experiments shows a marked difference for the two experiments, indicating that the initial plating current density has a significant impact on subsequent cycling.
Measuring the total amount of Li upon cycling allows to quantify the Li efficiency, here defined as the ratio between the amount of Li stripped (during oxidation of the investigated electrode) and the amount of Li plated (during reduction of the investigated electrode). The Li efficiency and the Coulombic efficiency during the two cycling experiments are shown in Fig. 5b. The Coulombic efficiency, the ratio of the integrating current during discharge and charge, does not allow to distinguish reactions that involve Li-ion transfer, including plating and stripping as well as SEI reactions, from those that do not involve Li-ion transfer, e.g., direct reduction/oxidation of the electrolyte. The Li efficiency allows to differentiate, by quantifying the amount of inactive Li, which is the sum of Li in the SEI, typically formed by one electron reduction of the EC/DMC electrolyte by the Li-metal57,58, and “dead” Li-metal. The difference between the Coulombic efficiency and the Li efficiency quantifies the amount of irreversible reactions that do not involve Li-ion transfer, such as direct electrolyte reduction. However, it should be realized that the total amount of inactive Li quantified by NDP is a combination of “dead” Li-metal, and (in)active Li in the SEI, as the chemical nature of Li cannot be distinguished with NDP.
Interestingly, Fig. 5b, c demonstrate a profound history effect as the evolution of the Li density and both efficiencies depends on the sequence of applying 0.5 and 2 mA cm–2. During the first cycle of both current densities, a large amount of inactive Li is observed, presumably a sum of Li in the SEI and “dead” Li-metal, becoming smaller for each subsequent cycle up to the 5th cycle. As a consequence, the initially small Li efficiencies, increase during the first five cycles for both current densities. Also the Coulombic efficiencies increase, stabilizing ~ 80%, signifying continuous SEI formation. The main difference between the first five cycles is that the Li efficiency increases to values ~ 100% for 2 mA cm–2, which is significantly larger compared with the Li efficiency at 0.5 mA cm−2. This indicates that the Li stripping is more efficient at a larger current density. For both current densities, the Coulombic efficiency is well below the Li efficiency, demonstrating that direct reduction of the electrolyte plays a significant role.
Comparing the two experiments, the largest changes are observed when comparing cycles 6–10. When the slow 0.5 mA cm–2 cycling is followed by fast cycling at 2 mA cm–2, again a steep increase in amount of inactive Li is observed, similar to the initial five cycles. This is consistently reflected in the drop in Li efficiency observed in Fig. 6c, which goes along with a strong decrease in Coulombic efficiency. In contrast, when fast 2 mA cm–2 cycling is followed by slow cycling at 0.5 mAcm–2, there is no increase in inactive Li, in fact the amount of active Li increases, which implies that a small fraction of the inactive Li is reactivated, the origin of the larger than 100% Li efficiency, whereas the Coulombic efficiency remains constant. Clearly, slow cycling appears much more reversible after initial fast cycling as compared to initial slow cycling.
To gain more insight in the impact of current density and the cycling history on the distribution of active Li-metal and inactive Li (inactive Li-metal and Li in the SEI), the Li density profiles after each plating cycle and after each stripping cycle for both experiments are shown in Fig. 6. Comparing the densities after plating during the first five cycles at 0.5 and 2 mA cm–2, Fig. 6a, c, reflects more dense plating at the larger current density, consistent with the current dependency demonstrated in Fig. 3. After five cycles at 0.5 mA cm–2 a slightly larger inactive Li density is observed after stripping, see Fig. 6b, as compared with cycling at 2 mA cm–2, see Fig. 6d, in particular above 10 μm depth, consistent with the lower Li efficiency in Fig. 5c. This can be rationalized by the less-dense and more distributed Li plating at 0.5 mA cm–2, which can be expected to result in more SEI formation and a larger chance on the formation of inactive Li-metal upon stripping. After the initial five cycles the evolution of the Li density after plating is profoundly different as can be observed by comparing Fig. 6a, c. Cycling at 0.5 mA cm–2 during cycles 6–10 (after 2.0 mA cm–2 cycling) results in more compact plating compared with initial 0.5 mA cm–2 plating. Oppositely, 2.0 mA cm–2 plating during cycles 6–10, after 0.5 mA cm–2 results in less-dense and more distributed plating compared with initial 2 mA cm–2 plating. Also after the initial five cycles the evolution of the Li density profiles after stripping is very different, demonstrating a rapid buildup of inactive Li at 0.5 mA cm−2 over the full layer thickness, whereas it remains practically constant at 2 mA cm−2.
These observations imply that cycling history has drastic impact on the plated Li-metal morphology upon subsequent Li-plating cycles. Based on the present observations, we propose the following mechanism, schematically shown in Fig. 6e–h. Less-dense plating at lower current densities, through less-dense nucleation15, results in more SEI formation upon cycling as there is more electrolyte volume available. The less-dense Li-metal morphologies are more susceptible for leaving inactive Li-metal upon stripping, through the more homogeneous stripping activity shown in Fig. 3, and consistent with the lower Li efficiency and larger inactive Li density after stripping comparing Fig. 6b, d. During subsequent plating at a higher current density, the less-dense SEI/inactive Li-metal morphology formed at the low current density promotes inhomogeneous plating, and hence rapid penetration through the SEI morphology, promoting dendrite formation into the electrolyte. This will expose Li-metal to fresh electrolyte and hence initiate further SEI formation and formation of “dead” Li-metal, explaining the rapid rise of the amount of inactive Li observed in Fig. 5b. In contrast, the denser nucleation and growth of Li-metal at 2 mA cm–2 is less susceptible to dead Li-metal formation and leaves less volume for SEI formation, which consequentially passivates, consistent with the stable Li and Coulombic efficiency in Fig. 5c. Perhaps the most interesting result is that this more dense morphology promotes dense plating near the current collector observed in Fig. 6c, even at lower current densities, also leading to activation of some of the inactive Li. The reactivated inactive Li may originate from reconnecting “dead” Li-metal, enabling subsequent stripping, or from reversible capacity stored in the SEI61.
Hence, we conclude that the current dependent Li-metal plating during the initial cycles templates the SEI that forms during the first cycles. The SEI morphology formed during these initial cycles, strongly influences the Li-metal-plating morphology on subsequent cycling. These results indicate the potential opportunities of initial cycling strategies to create in situ formed SEI morphologies that are more stable upon subsequent cycles.
Evolution of the Li density in the copper current collector
Although it is generally assumed that Li does not take up Cu, a small amount of Li uptake by Cu and other current collectors is known to occur53,62. Potentially, Li in the current collector can degrade both the structure and the electronic conductivity of the current collector53,62 both of which are crucial for the functioning of Li-ion batteries. That Li actually alloys with Cu is well known from the Cu-Li phase diagram63 from which an uptake of ~ 3.5 wt% Li may be anticipated at 100 °C, the lowest temperature reported. However, the Li kinetics in Cu is most likely to prevent this solubility limit to be reached. Although a few studies have reported small amounts of Li-metal in Cu53,62, the difficulty of measuring Li in current collectors and the absence of operando studies makes that current understanding of the uptake of Li in current collectors is very limited. A close look at the Cu current collector region in the operando NDP measurement shown in Fig. 1b surprisingly reveals a small amount of reversible Li uptake of the Cu current collector.
In Fig. 7a, the Cu region of the NDP measurement in Fig. 1b is shown, demonstrating Li uptake and release during four plating and stripping cycles at 1 mA cm–2. Li uptake in Cu was reported by post mortem analysis of the graphite anode using NDP53. In that case a limited penetration depth of ~ 1 μm was observed. Recently, chemical analysis showed the presence of ~ 10 μg after exposing Cu to Li-metal at 50 °C62. Because subsequent oxidation did not remove the Li in the Cu current collector, Li was concluded to be irreversibly trapped. The integrated amount of Li in the Cu, shown in Fig. 7b, demonstrates that a maximum of almost 20 μg cm–2 of Li is taken up by the Cu at a deposition potential of –0.18 V. This is largely removed during stripping (oxidation) at a potential of 0.2 V, leaving 4 μg cm–2 trapped in the Cu current collector within ~ 1 μm of the interface with the Li-metal anode. The amount and distribution of Li that is irreversible trapped appears quite similar to what was observed with post mortem analysis64. However, most of the Li, 16 μg cm–2, can be reversibly added and removed from the Cu, an phenomenon that to the best of our knowledge was not reported previously. Both the irreversible and reversible amount of Li uptake by the Cu current collector strongly depends on the plating potential, as demonstrated in Fig. 7d. At 0.2 mA cm–2, resulting in a plating potential of ~ –0.1 V, the irreversible and reversible amounts are both < 1 μg cm–2, but at 2 mA cm–2, resulting in a plating potential approaching –0.4 V, almost 25 μg cm–2 is incorporated, leaving ~ 4 μg cm–2 irreversibly trapped. Although these results clearly indicate that the plating potential drives the Li uptake of the Cu, further research is required to investigate the uptake kinetics and plating potential dependence. Interestingly, Fig. 7b indicates that subsequent cycling appears to reduce the reversible uptake of Li by Cu. At this stage it is not clear if this is due to induced changes in the grain boundaries, or SEI formation at the current collector that slows down the Li transport toward the Cu current collector. To further investigate the nature of the Li uptake by Cu operando X-ray Diffraction (XRD) was performed during the same electrochemical plating and stripping conditions, shown in Fig. 7c. The refined Cu lattice parameter during 15 plating/stripping cycles in Fig. 7c displays a variation as a consequence of the daily temperature changes in the laboratory (which based on the thermal expansion of Cu appears to be 4 °C). However, no lattice parameter changes due to the stripping and plating potentials are observed. This indicates that Li is not taken up significantly into the Cu crystal lattice, despite the large solubility limits indicated by the phase diagram, indicating that the Li diffusion through the Cu crystal is sluggish and does not occur at room temperature. Based on this, we propose that Li is primarily transported and taken up by the grain boundary regions in the Cu current collector, as schematically shown in Fig. 6e–h. This implies diffusion of Li over several micrometers through the Cu current collector via the grain boundaries, which requires large Li-ion mobility over the surface of Cu in the grain boundaries.
MD simulation of Li diffusion on Cu surfaces
To gain insight in the Li mobility on the Cu surface, molecular dynamics (MD) calculations based on density functional theory (DFT) were performed for Li at the dominant surfaces of Cu, the (100) and (111) surfaces, assuming that these surfaces can represent the Cu grain boundaries. First, evaluation of the energy of Li in the bulk of Cu with DFT demonstrates that Li insertion in Cu is highly unfavorable (–2.5 to –3.0 V vs. Li/Li+), whereas it is favorable at the surface (0.71 V and 0.58 V vs. Li/Li+ for (111) and (100), respectively) as set out in the Supplementary Methods in more detail. For the Cu(111) surface, Li stabilizes at a vertical distance of 2.1 Å from the surface copper atom layer, and the most stable positions suggest that Li is diametrically projecting either Cu atoms in the second or the third copper layer, as shown in Supplementary Figure 1. Thereby the distance to the nearest three Cu atoms is maximized to 2.585 Å, which is comparable to the Cu–Cu distances (2.57 Å). For Li at the Cu(100) surface the Li position is aligned with the Cu atoms in the second Cu layer, maximizing its distance from the first copper layer as shown in Supplementary Figure 1. To evaluate the Li mobility at the Cu surfaces, MD simulations were performed. For the Cu(111) surface, a single diffusing Li diffuses over the complete 92 Å surface within 75 ps as demonstrated by the integrated Li density shown in Fig. 8a, b. In Fig. 8c, the detected transitions between the stable p2 Li positions, shown as red lines, reveal a hexagonal diffusion pathway. Even transitions via the most unfavorable p1 position occur, which is positioned 100 meV in energy above the p2 positon. This demonstrates that the energy lands scape for Li diffusion over the Cu surface is very flat, promoting high diffusivity. Similar results are obtained for more adsorbed Li-ions and for the (100) surface as shown in Supplementary Figure 2 and 4, respectively.
The activation barriers can be calculated directly from the number of transitions, given the attempt frequency65,66. Assuming an attempt frequency of 1013 s−1, this results in energy barriers between 26 and 49 meV between the p2 positions. Based on the mean square displacement, shown in Supplementary Figure 3, the tracer diffusion was predicted to be 4.21 × 10–8 m2 s–1, indicating very rapid diffusion of Li on the surface of Cu. Details on the bulk and surface calculations can be found in the Supplementary Methods. These findings provide support that grain boundaries in the Cu current collector allow rapid Li transport, providing a rational for the observed lithiation of the Cu current collector in Fig. 7. This may give a hint that the Cu microstructure has an important role in the Li uptake of Cu current collectors. Minimizing the presence of Cu grain boundaries at the surface of the current collector is therefore brought forward to be a potential strategy to suppress Li uptake and degradation of Cu, and perhaps also other current collectors. The present operando NDP measurement provides the possibility to study reversible and irreversible Li uptake of Cu, or of any other current collector, under the various operational conditions. This direct view on Li in current collectors can be used to develop more understanding of Li-metal uptake in current collectors and support the development of strategies that aim at preventing current collector degradation.