Mapping Structure-Composition-Property Relationships in V- and Fe-Doped LiMnPO 4 Cathodes for Lithium-Ion Batteries

: A series of LiMn 1 − x − y Fe x V y PO 4 (LMFVP) nanomaterials have been synthesized using a pilot-scale continuous hydrothermal synthesis process (CHFS) and evaluated as high voltage cathodes in Li-ion batteries at a production rate of 0.25 kg h − 1 . The rapid synthesis and screening approach has allowed the speci ﬁ c capacity of the high Mn content olivines to be optimized, particularly at high discharge rates. Consistent and gradual changes in the structure and performance are observed across the compositional region under investigation; the doping of Fe at 20 at% (with respect to Mn) into lithium manganese phosphate, rather than V or indeed codoping of Fe and V, gives the best balance of high capacity and high rate performance.


■ INTRODUCTION
Lithium-ion batteries (LIBs) currently offer the highest energy densities of all commercial rechargeable battery technologies. Current cathode technology in LIBs relies largely on cobalt oxide-based materials, and there has been much interest focused on developing more inexpensive and sustainable materials composed principally of less toxic metals. The olivine family of materials (especially LiMPO 4 , where M = Fe, Mn, or others) have been heavily researched as candidate Li-ion battery cathodes because of their reasonable theoretical capacity (170 mA h g −1 for LiFePO 4 ), thermal stability, environmental inertness and high cycle life. 1 LiMnPO 4 (LMP) is of interest as a higher energy density cathode compared with LiFePO 4 (LFP) [theoretical energy density value of 697 for LMP vs 586 W h kg −1 for LFP]. This difference is due to the higher potential of the Mn 2+ /Mn 3+ couple compared with the Fe 2+ /Fe 3+ (4.1 V vs Li/Li + as opposed to 3.45 V vs Li/Li + ). 2 However, LMP has lower electronic conductivity than LFP (10 −10 S cm −1 for LMP compared to the range ∼10 −7 to 10 −10 S cm −1 for LFP). 3 There are also electron−lattice interactions in LMP (from the Jahn− Teller effect), which causes significant lattice deformations, 4 thereby limiting its achievable capacity and cyclability. 5 Manganese dissolution in the electrode has also been reported as an issue during operation, leading to Mn plating onto the Li anode and a decline in electrochemical performance. 6 For these reasons, the electrochemical performance of pure (undoped) LMP is very much limited, and generally requires a higher proportion of conductive carbon in the electrode compared to LFP (>20 wt%) or nanosizing the particles or carbon coating particle surfaces to attain significant reversible Li + intercalation. Rangappa et al. reported a specific capacity of 153 mA h g −1 at a discharge rate of C/100 and a value of 62 mA h g −1 at 0.5 C for LMP made via a liquid phase batch process. 7 Wang et al. achieved a specific capacity of 145 mA h g −1 at a discharge rate of C/20 and a value of 113 mA h g −1 at 1 C for LMP made using a batch polyol synthesis method. 8 More recently, Zheng et al. reported a specific capacity of 152 mA h g −1 at C/20 and 60 mA h g −1 at 5C, 9 using a solid state synthesis route from a mixture of Mn 2 O 3 and MnO 2 . Yoo et al. synthesized 3D macroporous LiMnPO 4 flakes via a colloidal template, which when tested as a cathode achieved 162 mA h g −1 at C/10 and 110 mA h g −1 at 10C. 10 All of the aforementioned electrodes contained a high proportion of carbon (generally 17 wt% or more), significantly reducing tap density and therefore, reducing the likelihood for commercial development. Kwon et al. achieved a capacity of 165 mA h g −1 at C/40 and 66 mA h g −1 at 1C with an electrode with only 10 wt% carbon content. 11 Clearly, the literature shows that pure LMP as a cathode material is limited, particularly at high C rates.
Doping LMP with transition metals is a promising approach to improve electrochemical performance. Doping with cations of oxidation states >2+ also brings the potential advantage of generating Li + vacancies as a charge-compensation mechanism, and is a well-established method of improving the performance of both LFP and LMP. 12 It has been observed that replacing the manganese ion with another divalent cation (such as Zn 2+ , Mg 2+ , or Fe 2+ ), leads to improved capacity retention at high discharge rates, which was attributed to increased electronic conductivity and reduced particle size. 12 Martha et al. developed LiMn 0.8 Fe 0.2 PO 4 /C cathodes, which achieved a specific capacity of 162 mA h g −1 at C/10 and >90 mA h g −1 at 10C. 13 14 As well as LMP, LFP has also been doped with transition metals, such as V, resulting in dramatic improvements in highrate performance, either because of favorable structural changes in the olivine lattice/improvements in conductivity or due to the formation of highly conductive secondary vanadiumcontaining phases on the LFP surface. 15−19 The latter approach has also been used to improve the performance of LMP cathodes in a Li-ion battery, by formation of a V-doped LiMnPO 4 material or a composite cathode comprising of a highly conductive Li 3 V 2 (PO 4 ) 3 phase mixed with LMP. 20−22 The majority of syntheses of LFP and LMP in the literature consist of batch solid-state or hydrothermal or wet chemical methods. 9,21,23−25 Solid-state methods to make such materials into high performance cathodes usually involve high temperatures, lengthy heat-treatments and multiple processing steps, including grinding. 9,21 In contrast, batch hydrothermal reactions typically require lower synthesis temperatures while often needing to incorporate surfactants to reduce particle size and control morphology, all of which adds additional cost and complexity to the process. Such methods also suffer from batchto-batch variations and are challenging to scale-up.
Continuous hydrothermal flow synthesis (CHFS) methods possess some advantages compared to batch-type synthesis routes, as continuous processes are often more flexible and can allow independent control over reaction variables (such as pressure, reaction temperature and residence time). Furthermore, the CHFS process can convert solution precursors to solid products (dispersed in the process liquid) over timescales of a few seconds, often generating kinetic (metastable) products. With CHFS methods, this rapid conversion to products (typically <1 s) is achieved by mixing aqueous metal salts at room temperature with a flow of supercritical water (in a well-defined mixer arrangement, Figure S1−2), rapidly forming nanoparticles via simultaneous hydrolysis and dehydration. 26,27 The speed and simplicity of CHFS, has led to the development of high throughput continuous hydrothermal routes for rapid manufacture of libraries of nanomaterials, to allow exploration of structure−property-compositional relationships. 28−32 The approach was previously successful in the production of large or partial phase diagrams for nanomaterials, such as the Ce−Zr−Y−O and Ce−Zn−O systems. 28−30 Using a similar rapid synthesis approach, libraries of doped nanomaterials were developed, for example, ZnO 31 and Zn−Ti oxide photocatalysts, 32 and Eu-doped yttria phosphors. 33 In addition to the direct synthesis of nanomaterials libraries, highthroughput CHFS was used to make nanocoprecipitated oxides, used as precursors for the direct solid state (heat-treat with no grinding) synthesis of complex oxide libraries, for example, doped La 4 Ni 2 FeO 10-δ mixed ion conductors. 34 In many of these reports, the use of fast screening or parallel testing methods were used to quickly develop structure− property-composition relationships for the nanomaterials/ products.
Historically, continuous hydrothermal processes have largely been used to investigate the production of metal oxides. More recent work has begun to explore sulfides, 35 and also phosphates. 27,36−38 To the best of the author's knowledge, there are no peer-reviewed reports of continuous hydrothermal flow synthesis of LMP or its doped analogues, although LFP has been reported with varying degrees of performance as a Liion battery cathode. 39−41 A series of LiMn 1−x−y Fe x V y PO 4 (LMFVP) nanomaterials were synthesized using a pilot-scale CHFS process in an attempt to optimize the specific capacity within compositions with high Mn content, and enable comparisons between the merits of the two dopants.

■ RESULTS AND DISCUSSION
Herein, we report the synthesis of LMP and the manganeserich region of the LiMn 1−x−y Fe x V y PO 4 phase diagram (as highlighted in Figure 1a). Pure LiMnPO 4 and the Fe-or Vdoped variants (with nominal compositions shown in Figure  1b) were all synthesized with an amorphous carbon coating using a pilot-plant CHFS process. 26 The rapid formation of a large number of samples (17) in the phase-space allowed the comparison of structure and performance, which has not previously been achieved for these compounds. The pure LiMnPO 4 /C sample was named LMP, and the iron/vanadiumdoped samples (also coated with amorphous carbon) were named LMFVP(x,y), where x and y are the nominal percentages of iron and vanadium, respectively (as a proportion of total transition metal content). The nominal content refers to the metal at% present in the precursors, normalized to a total of 100%. These samples were all heat-treated (to graphitize the carbon coatings) to give the corresponding samples named ΔLMP and ΔLMFVP(x,y). The as-prepared samples after freeze-drying were generally fine gray-blue or gray-pink powders, and heat-treatment of these samples, gave black powders. The properties of the surface carbon was analyzed using Raman spectroscopy for both the as-prepared and heattreated samples. Prior to heat-treatment, a very broad feature in the range 200−2000 cm −1 was observed, corresponding to a high proportion of disordered carbon, although some graphitic sp 2 carbon was evident from the presence of the G band ( Figure S3). Analysis of the samples after heat-treatment confirmed the conversion of the carbon coating to a graphitic carbon coating, with only the characteristic D and G bands observed ( Figure S4). The small feature observed at 950 cm −1 was consistent with the symmetric stretching vibration of PO 4 .
Powder X-ray diffraction (XRD) patterns of the as-prepared samples all exclusively showed the pure olivine LiMnPO 4 structure (good match to JCPDS reference pattern number 00-077-0178, Figure S5a−c). Significant peak-shift and broadening was observed in the as-prepared olivine structures as the

ACS Combinatorial Science
Research Article vanadium levels increased, which may be indicative of increasing lattice strain, and comparatively reduced shift (with no broadening) was observed with increasing Fe content ( Figure S5b and d). There was consistent variation in the lattice parameters and unit cell volume with composition, as shown in Figure 2a−d and Table S2, calculated from LeBail fits of the X-ray Diffraction patterns ( Figures S6−22). It can be seen that increasing Fe and V dopant levels led to a smooth reduction of the a and b lattice parameters, whereas the V dopant had the most significant effect on the c parameter, with increased V substitution increasing the c parameter across all compositions. This is consistent with the distortion observed with V substitution in LFP by the authors and others, and implies successful incorporation of V into the olivine structure, which is still a matter of some debate in the literature. 18,42,43 This systematic change in parameters could only be observed due to the wide phase-space observed and the number of samples analyzed.
The samples were all heat-treated to graphitize the carbon and remove any antisite defects. However, a limitation in vanadium solubility after heat-treatment was observed, with an Li 3 V 2 (PO 4 ) 3 impurity phase forming for certain dopant levels as shown in the XRD patterns (Figure 2e, Figure S5e and f). Interestingly, the additional presence of iron in the V-doped LMP samples biased the formation of a different impurity phase, Fe 2 P 2 O 7 , compared to the samples doped with V only (Figure 2f, Figure S5f). The location of the two weak and broad peaks at 13.2 and 13.8°in 2θ, resulted in a tentative assignment for the iron pyrophosphate (Fe 2 P 2 O 7 ) phase (match to JCPDS reference pattern number 00-076-1672). This result was somewhat surprising, as it suggests the iron dopant was less stable in the structure than vanadium in these materials. The presence of aliovalent dopants in the olivine structure has traditionally been discounted, 43 although there is an increasing body of evidence that aliovalent doping is possible in both LiFePO 4 and LiMnPO 4 . 16,18,20,42,44 Indeed, the high doping degree achieved in sample LMFVP(0,20) has not been achieved for LMP previously, and indicates the phase was only accessible due to the relatively low synthesis temperature used herein (335°C) and rapid conversion from precursor to product in the process.
Field-emission scanning electron microscope (FE-SEM) images of selected heat-treated samples showed they consisted of fused networks of agglomerated particles approximately 100 nm in size (Figure 3). A significant amount of faceting was observed in sample ΔLMFVP(20,0), which appeared cuboidlike, and may reflect changes in the surface energies of the samples with doping. The small particle size achieved herein reflected the high degree of supersaturation and rapid nucleation achieved when the precursors were combined with supercritical water in CHFS.
The heat-treated samples were made into electrodes in Li-ion coin half-cells and cyclic voltammetry (CV) tests were conducted at a scan rate of 0. The Fe-doped samples generally exhibited significantly better rate capability in constant-current constant-voltage tests compared with the pure ΔLMP sample. In contrast, V-doping was found to have less influence on cathode performance ( Figure 5, Table S3). The heat map of the sample capacities generated at C/2 and 5C, respectively, showed the dominant beneficial effect of iron doping on the discharge capacity of LMFVP. This was surprising given the well-documented apparent positive effect of the inclusion of V in the olivine lattice or the presence of a Li 3 V 2 (PO 4 ) 3 conductive impurity. 20−22 However, the ratio Li 3 V 2 (PO 4 ) 3 :LiMnPO 4 was lower in this study (maximum possible was 1:8 for Li 3 V 2 (PO 4 ) 3 :LiMnPO 4 ) than those typically reported in the literature (that are typically 1:4), 21,22 and could explain this apparent contradiction. To our knowledge, this is the first combinatorial analysis of two beneficial dopants in the nanoparticle LiMnPO 4 system to give an accurate comparison.

Research Article
The performance of the LiMn 0.8 Fe 0.2 PO 4 cathode at a 5C charge/discharge rate was superior compared with other Fedoped samples herein at this C rate, and achieved a discharge capacity of 85 mA h g −1 (Figure 6), and maintained a stable discharge capacity of 125 mA h g −1 at C/2. This trend was reflected in the energy densities achieved by the samples (Table  1), where a smooth increase was observed with increasing Fe content. The improvement in high power performance, can be attributed to an increase in bulk electronic conductivity as suggested by others, 5 and also reduced strain in the delithiated olivine structure due to the reduced presence of Jahn−Teller distorted Mn 3+ ions, when the Mn is partially substituted with Fe. 5,12 While the proportion of carbon in the electrode was still too high compared to a commercial electrode (15 wt% overall), this is still comparatively low compared to similar LMP based materials in literature. 7,8,45 Combined with the semi-industrial scale of synthesis of these materials, the performance detailed herein for the best doped LMP samples represents a step forward in the development of LiMnPO 4 cathodes. Long-term cycling tests indicated reasonable stability of the material, with 70% capacity retention over 200 cycles at C/10 ( Figure S36). A Coulombic efficiency of 98% was also observed, implying minor

ACS Combinatorial Science
Research Article electrolyte degradation due to the high charge voltage (4.5 V) used.

■ CONCLUSIONS
A high-throughput nanomaterials synthesis approach was taken to develop high-performance carbon-coated LMFVP cathodes via a continuous hydrothermal pilot-scale process. The effect of vanadium and iron dopants were investigated across part of the tertiary phase diagram for the heat-treated olivines, where the iron dopant had the most beneficial effect on improving electrochemical performance. The combination of carboncoating, nanosizing and iron doping gave an optimized cathode material of LiMn 0.8 Fe 0.2 PO 4 , which achieved a reversible specific discharge capacity of 85 mA h g −1 at a high discharge rate of 5C. This was comparable to the best performances achieved in the academic literature for similar materials, especially given the low carbon content (15 wt% C) of the electrode. Importantly, the materials herein had the advantage that they were synthesized on a pilot-plant at process at a production rate of 0.25 kg h −1 rather than grams per day as demonstrated by the comparable literature, which offers potential for further scale-up and investigation of even larger cell formats or packs in the future.

■ EXPERIMENTAL PROCEDURES
Carbon-coated iron-and vanadium-doped lithium manganese phosphate samples (where the C was amorphous in the asprepared material) were synthesized using a pilot-scale CHFS reactor utilizing a patented confined jet mixer (CJM). A detailed description of the pilot-scale process can be found elsewhere, and a schematic is provided in the Supporting Information ( Figure S1). 25 The CHFS process for making nanomaterials can be summarized as follows: the precursors were prepared in two aqueous solutions. The first aqueous solution consisted of the following metal salts in the desired ratio; MnSO 4 Table S1). The second solution used in the process contained 0.8625 M LiOH·H 2 O (99+ %, Fischer Scientific, Loughborough, UK) in D.I. water for all experiments.
Both the metal salt and LiOH solutions were pumped to meet in a T-piece mixer (0.25 in. internal diameter) with 200 mL min −1 flow rate each. This combined mixture thus had a total flow rate of 400 mL min −1 and this was then delivered into the sidearms of the confined jet mixer (CJM, Figure S2), where it rapidly mixed with a flow of supercritical water at 450°C and 24.1 MPa (flow rate of 400 mL min −1 for this water when it was under ambient conditions) to give reaction temperature of ∼335°C under highly turbulent conditions (the design of the CJM is in the Supporting Information and is described fully

ACS Combinatorial Science
Research Article elsewhere). 25 The nanoparticles formed in or near the CJM exit had a residence time of ∼6.5 s before passing through a pipe-inpipe countercurrent heat exchanger and being cooled to near ambient temperature. The cooled slurry then passed through a back-pressure regulator (Swagelok KHB Series) which maintained the system pressure at 24.1 MPa, after which it was collected in a plastic container open to the atmosphere. The slurry settled after 1 h and the supernatant (containing unreacted precursors and byproducts) was siphoned off. The slurry was concentrated further using centrifugation (1500 rpm for 5 min), and the resultant wet paste was dialyzed in D.I. water until the conductivity of the liquid reduced below 150 μS. The cleaned paste was further concentrated with centrifugation (4500 rpm over 30 min) to give a clean, wet product, which was frozen and freeze-dried by slowly heating from −60 to 25°C , over 24 h under vacuum of <13 Pa. The freeze-dried powder was subsequently heat-treated from ambient temperature up to 700°C (held for 3h at this temperature), with a heating rate of 5°C min −1 under a flow of argon.
Powder XRD patterns were obtained on a Bruker D4 Endeavor diffractometer using Cu Kα radiation (λ = 1.54 Å) in the 2θ range 5 to 60°with a step size of 0.05°in 2θ and a count time of 4 s. LeBail analysis was performed using MAUD (Material Analysis Using Diffraction) software. 46 In addition, the Fe 2 P 2 O 7 impurity phase was identified using Mo−Kα radiation (λ = 0.71 Å) over the 2θ range 10−18°with a step size of 0.5°and step time of 20 s. The size and morphology of the heat-treated particles were determined by FE-SEM.
Raman spectra were collected on a Raman microscope system with a laser excitation wavelength of 514.5 nm; the laser power was set to 10% of full power for all samples. Scans were collected in the range 200−2000 cm −1 ; each individual scan lasted 30 s, and the sum of 4 individual scans were combined to give the final spectrum (total scan time of 2 min).
The Li-ion battery cathodes were prepared by mixing each heat-treated sample with conductive agent (carbon black, Super P, supplied by Alfa Aesar, Heysham, UK) and polyvinylidene fluoride, PVDF (PI-KEM, Staffordshire, UK) in such a way as to give an active material/carbon/binder dry mass ratio of 75:15:10 for all electrodes (inclusive of surface carbon from the synthesis and heat-treatment, which was in the range 2.5 to 6.3 wt% of the overall mass of the heat-treated samples) (Table  S1). To make inks, the PVDF was stirred in N-methyl-2pyrrolidone, NMP (>99.5%, Sigma-Aldrich, Steinheim, Germany) for at least 1 h at room temperature until it was fully dissolved. This solution was combined with the heat-treated sample and conductive agent and ball-milled for 1 h; the resulting slurry was cast on aluminum foil (PI-KEM, Staffordshire, UK) and dried in an oven at 70°C. 16 mm diameter circular electrodes were punched out and pressed with a pressure of 1.5 tons cm −2 . The mass loadings for all electrodes were in the range 2.3 to 2.7 mg cm −2 .
Two electrode 2032-type coin cells (Hohsen, Japan) were used to conduct electrochemical experiments, and were assembled in a dry room with a dew point of <−40°C. Lithium metal foil was used as the counter electrode (PI-KEM, Staffordshire, UK), and the separator (Celgard) was saturated with an organic electrolyte (1 M LiPF 6 in 1:1 ethylene carbonate/ethyl methyl carbonate, with 1 wt% vinylidene carbonate additive, supplied by BASF, Ludwigshafen, Germany).
Electrochemical measurements were performed using an Arbin Instruments battery tester at 20°C. The electrochemical performance was investigated by cyclic voltammetry in the range from 2.0 to 4.5 V vs Li/Li + at a scan rate of 0.05 mV s −1 . Galvanostatic charge/discharge cycling tests (specific current tests) were performed on LMP and all LMFVP compositions in the range of 2.0−4.5 V vs Li/Li + , applying a constant current, constant voltage (CCCV) test regime, with C-rates of C/2, 1C, 2C, and 5C being used during charge and discharge (where 1C value was assigned as 170 mA h g −1 ), and holding the voltage used was 4.5 V (at the end of the charge step) until the current had decreased to 3% of the charge rate. Long-term cycling tests were performed subsequently at C/10, using the same holding voltage regime. The specific current and specific capacity was calculated based on the mass of active material (i.e., pure or doped LiMnPO 4 ) for each printed electrode.

Author Contributions
The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

Notes
The authors declare no competing financial interest.

ACS Combinatorial Science
Research Article