High-performing iron phosphate for enhanced lithium ion solid state batteries as grown by atomic layer deposition

Atomic layer deposition (ALD) is an excellent tool for realisation of uniform coating of cathode materials on highly 3D-nanostructured microbatteries. We have developed an ALD-process for deposition of iron phosphate, FePO4, as a cathode material and characterised its electrochemical properties towards a lithium metal anode. Thin films were deposited between 196 and 376 °C using the precursor pairs: trimethyl phosphate (TMP, Me3PO4) with both H2O and ozone (O3) as an oxygen source, and Fe(thd)3 (Hthd = 2,2,6,6-tetramethyl-3,5-heptanedionate) with O3. The as-deposited films are amorphous and crystallize to trigonal FePO4 after heat treatment in air at 600 °C. The amorphous FePO4 films were characterised electrochemically proving exceptional cyclability and capacities almost reaching the 100% theoretical value (178 mA h g−1) for 1 hour charge–discharge rates.


Introduction
Lithium ion batteries are currently used in numerous mobile applications ranging in size from smaller electronic devices to cell phones, power drills and electric vehicles.Common to these applications is the need for materials with high gravimetric and volumetric energy densities along with large power density.Lithium ion batteries already full many current demands, however, for enabling future applications even higher densities and smaller footprints are required.][3] The latter approach is relatively demanding as it requires a production technique that provides pin-hole free lms on large aspect ratio structures.5][6] Currently, we report an ALD-process for deposition of thin lms of FePO 4 along with electrochemical characterization of such lms as cathode materials in Li-ion batteries.
In bulk, iron phosphate, FePO 4 , is implemented as a cathode material with a theoretical specic capacity of 178 mA h g À1 upon 1 mol of lithium intercalation.It is made of cheap and abundant elements and is environmentally benign.However, it is difficult to obtain the full theoretical capacity of FePO 4 due to poor kinetics of the lithium intercalation/deintercalation process.Nevertheless, this limitation is reported to be overcome by nanostructuring. 7,8e electrochemical properties of FePO 4 are strongly affected by the atomic arrangement.0][11] This is likely due to the instability of tetrahedral Fe 2+ in trigonal FePO 4 . 101][12][13][14] The electrochemical capacity of amorphous hollow nanospheres of FePO 4 is reported to be as large as 174 mA h g À1 at C/10 and 150 mA h g À1 at 1C, 8 whereas its trigonal crystalline counterpart only shows 40 mA h g À1 at C/10. 12 However, the capacity of crystalline olivine FePO 4 is reported to be 154 mA h g À1 at 1C. 15 Although thin lms of LiFePO 4 have been subject to numerous studies, 14,16,17 reports on unlithiated FePO 4 are rather limited.Thin lms of amorphous FePO 4 are reported to be grown by rf-sputtering and are used in all-solid-state batteries (Li/LiPON/FePO 4 ) with an overall capacity of 21 mA h cm À2 , which we estimate to roughly correspond to 6.6 mA h g À1 in gravimetric capacity. 18he cathode materials that previously have been made by ALD are mainly unlithiated transition metal oxides such as TiO 2 , 19 MnO 2 , 20,21 24,25 and NiO. 26he lithiated cathode materials reported so far are LiCoO 2 (ref.27) and LiFePO 4 . 28he eld of phosphate materials deposited by ALD is relatively limited.The rst ALD deposition of phosphates was that of aluminium phosphate, AlPO 4 , in the 1990s 29,30 40 and FeCp 2 /O 2 . 41n this work we rst describe the FePO 4 deposition process, supported by structural and morphological characterisation by X-ray diffraction (XRD) and atomic force microscopy (AFM).Second, we present electrochemical characterization data for the amorphous as-deposited lms by galvanostatic measurements and cyclic voltammetry.Compared to previous work these lms exhibit enhanced electrochemical performance as cathodes for Li-ion batteries.

Experimental
The lms were grown in an F-120 Sat reactor (ASM Microchemistry Ltd) using Fe(thd) 3 , Me 3 PO 4 (97.0%,Fluka), H 2 O (18 MU cm À1 ) and O 3 as precursors.Fe(thd) 3 was synthesized from FeCl 3 (99.0%,Fluka) and Hthd (98%, Fluka) as described in ref. 42.The source temperature for Fe(thd) 3 was 115 C, and Me 3 PO 4 , H 2 O and O 3 were introduced into the reactor at room temperature using a carrier gas for the Me 3 PO 4 source.Ozone was generated by feeding O 2 (99.999%,AGA) into an ozone generator (OT-020, OzoneTechnology, or 803N, BMT Messtechnik GMBH).The ow rate of ozone during its pulses was ca.500 cm 3 min À1 .The pressure in the reactor was kept below 3 mbar using a total N 2 ow (99.9995%) of 500 cm 3 min À1 .The lms were deposited on substrates of p-type Si(111), soda lime glass and 316 Stainless steel (CR20).

Material characterization
The thickness of the deposited lms was analyzed by spectroscopic ellipsometry using an alpha-SE ellipsometer (J.A Woollam Co. Inc.) and by X-ray reectivity (XRR) using an AXS D8 Discover (Bruker).The X-ray reectivity data were tted using the GenX soware 43 and the ellipsometer data were tted to a Cauchy function.XRR was further used to determine the density of the deposited lms.The deposited lms were studied using X-ray diffraction in grazing incidence (GI-XRD) and q-2q conguration using a D5000 (Siemens) diffractometer equipped with a Göbel mirror providing parallel Cu-Ka radiation.For the GI-XRD measurements u ¼ 0.5 was used as the incident angle.The lm roughness was analyzed by atomic force microscopy (AFM) using a XE-70 (Park Systems) in non-contact mode.X-ray uorescence (XRF) was used to determine the composition of iron and phosphorus in the deposited lms using a Philips PW2400 spectrometer and UniQuant analysis soware. 44

Electrochemical characterization
The electrochemical properties of the deposited samples were investigated in CR2032 coin cells.Cell assembly was carried out in an Ar-lled glove box with water and oxygen levels below 0.1 ppm.Metallic lithium (99.9%,Sigma-Aldrich) was used as an anode and FePO 4 lms as-deposited on steel substrates were used as a cathode.The active mass was calculated based on thickness and density from XRR.The electrolyte consisted of 1 M LiClO 4 (99.99%,Sigma-Aldrich) in a 1 : 1 mixture of ethylene carbonate (EC) (99%, Sigma-Aldrich) and dimethyl carbonate (DMC) ($99% Sigma Aldrich).Electrochemical properties were measured at room temperature using a MPG2 (Biologic).Cyclic voltammetry measurements were carried out in the range 2.3-4.0V with a scanning rate of 0.1 mV s À1 .The voltage was kept constant for 30 minutes at the two endpoints of 2.3 and 4.0 V.The cyclability and specic capacity of the amorphous FePO 4 lms were investigated by taking galvanostatic measurements.The galvanostatic data were measured during charge-discharge cycles at a current rate close to 1C (181 mA g À1 ) over the voltage range 2.0-4.0V with 5 min relaxation time aer each cycle.

Results
The system for deposition of FePO 4 is based on individual subcycles managing the iron and phosphorous content:  1, and prove that all precursors saturate within the selected pulse times.
An optimised pulsing scheme was used throughout this investigation: 2s Fe(thd) 3 -1.5spurge -4s O 3 -5s purge -4s TMP -3s purge -8s (H 2 O + O 3 ) -5s purge.The temperature dependency of the growth was investigated by maintaining a 1 : 1 supercycle of Fe and P and varying the deposition temperature.A total of 1000 supercycles were applied and the achieved growth rates and compositions are given in Fig. 2.
The Fe-content in the lm increases steadily with the deposition temperature, and reaches an equimolar ratio at 350 C.However, the overall growth rate remains almost constant in the temperature range of 246-360 C. Above 360 C, large visible gradients suggest decomposition of at least one of the precursors.
The possibility of tuning the deposited composition was investigated by applying different combinations of subcycles.The individual subcycles were alternated in a manner that maximized the intermixing of the two components.The total number of subcycles was kept at 2000 and the deposition temperature was maintained at 246 C. The results show a notable variation in growth rate with deposited composition, with a maximum at the equimolar pulsing ratio, Fig. 3.The deposited compositions varied from ca. 40% Fe with respect to the Fe + P content, to ca. 70% for a 4 : 1 pulsing ratio of Fe : P. Fig. 3 shows that an equimolar ratio of Fe and P is obtained in the lms for a pulsing ratio of 3 : 2. This pulsing ratio was maintained in the following deposition processes of FePO 4 .
At 246 C the deposited lm thickness varies linearly with the number of supercycles without any sign of nucleation barriers, Fig. 4. The lm density remains quite constant, around 3 g cm À3 , Fig. 4.
A 100 nm thick lm deposited at 246 C was investigated by X-ray diffraction as-deposited and aer annealing at 600 C in air, Fig.The topography of the as-deposited 100 nm thick lms and of lms annealed at 660 C for 15 min in air shows rather smooth surfaces with a root mean square (RMS) roughness of ca.0.6 nm, that increases to ca. 1.6 nm upon annealing, Fig. 6.The lms were deposited in the range 220-325 C.
The electrochemical activity of an amorphous 46 nm thick as-deposited FePO 4 lm (deposited at 246 C) using 400 supercycles was characterized by cyclic voltammetry and galvanostatic measurements, Fig. 7 and 8.
Previous work reports that addition of a conductive coating is required to achieve acceptable capacities for both amorphous FePO 4 and crystalline LiFePO 4 at current rates of 1C and higher. 8,14The ALD deposited thin lms tested in this work contained no such conductive additives.We have therefore also chosen a current rate close to 1C for the galvanostatic measurements to investigate whether the inherent poor kinetics of FePO 4 would be compensated by a lm thickness in the nanometer range.
The cyclic voltammetry in Fig. 7 shows clear signs of reduction and oxidation taking place at 2.9 and 3.2 V, respectively.However, the features are not well dened in potential, which stems from the amorphous character of the lm.
From the galvanostatic measurements (Fig. 8) the initial electrochemical capacity of the material was 159 mA h g À1 whereaer it steadily increased on cycling until it saturated at 175 mA h g À1 aer 230 charge-discharge cycles.Thereaer, the capacity began gradually to decline.However, the capacity retention aer 600 cycles (165 mA h g À1 ) is still 3% higher than the initial capacity.The electrochemical loss between charging and discharging is 0.5% at the beginning and 1% aer 600 cycles, indicating good reversibility.

Discussion
As the Me 3 PO 4 precursor does not yield thin lm growth without pulsing of cation precursors, it is interesting to note that the phosphate composition, Fig. 3, saturates at around 60% P. Similar behaviour has earlier been observed for this TMP precursor.In ref. 31 Putkonen et al. reported a saturation of P/(Ca + P) ¼ 46% for the Ca-P-O system, and in ref. 32  With respect to the cyclic voltammetry measurements, Fig. 7, the capacity of the rst discharge is higher than the following reversible discharge capacities.This is likely attributed to SEIlayer formation and possibly other surface effects.Further in the rst discharge, the peak of intercalation of lithium ions occurs at 2.84 V.In the second discharge the peak is shied to 2.87 V, whereas it remained stable at 2.95 V in all subsequent cycles.It is therefore likely that the material rearranges during intercalation and that the increased discharge potential is due to slight changes in the Fe 2+ /Fe 3+ local structure.The charge peaks on the other hand remained stable at 3.2 V.
Fig. 5a shows that the as-deposited lms, on which the electrochemical measurements were carried out, are amorphous.Since it has not previously been observed that amorphous FePO 4 crystallizes as a result of Li intercalation/ deintercalation during electrochemical cycling, it is likely that the lms investigated here also remained amorphous during the electrochemical measurements. 7,10,12,46,47If the lms did indeed crystallize during cycling, we would have expected a decrease in capacity rather than the observed increase based on that all known crystalline phases of FePO 4 have less electrochemical activity than the amorphous phase. 14Furthermore, Fig. 7 shows that the amorphous FePO 4 lms exhibit reversible intercalation and deintercalation of lithium at about 3 V, which is an expected value for the Fe 2+ /Fe 3+ redox pair in an amorphous phosphate environment. 7No other peaks occur in the CV-plot that could indicate other redox active species.The lack of distinct peaks in the CV-plot supports the presence of amorphous materials, 25 and also indicates single phase reactions.This is consistent with what was previously observed for a-FePO 4 . 7,10,12,46he high capacity observed in this work is likely due to the fact that the investigated lms are only 46 nm thick, which limits the impact of the inherent poor electrical and ionic conductivity of FePO 4 .Amorphous materials have previously been shown to support the stress generated by lithium insertion/desertion better than crystalline lms, due to a higher exibility. 25,48To our knowledge, the type of increase in capacity during cycling we observe in our work has only been observed previously for nanostructures 49,50 and amorphous lms. 48We therefore suggest that the observed increase in capacity during cycling is likely linked to the amorphous structure of the lm.We speculate that the exibility of the amorphous structure allows for necessary rearrangements during lithium intercalation/deintercalation that activates more of the material, resulting in excellent cyclability.Furthermore, the advantage of an amorphous phase was also shown by K. Le Van et al. in ref. 25.
They could reversibly intercalate 2.9 mole of Li into 1 mole of amorphous V 2 O 5 in a potential window between 4 and 1.5 V versus Li/Li + .In contrast, intercalating more than 1 mole of Li into 1 mole of crystalline V 2 O 5 results in irreversible phase changes that reduces the cycling performance. 51

Conclusion
Nanostructuring of battery materials is a promising way to improve the performance of Li-ion batteries so they can meet the future demands.ALD shows large potential as an enabling technology for 3D-structured Li-ion batteries.We have developed an ALD process for deposition of electroactive FePO 4 and characterized its electrochemical performance.The as-deposited material reveals excellent performance when used as a cathode in Li-ion batteries with a reversible capacity at 1C-rate as high as 98% of the theoretical capacity.In addition the material shows outstanding cyclability and aer 600 chargedischarge cycles at 1C-rate the initial capacity has increased by 3%.Based on these observations nanostructured amorphous FePO 4 shows great promise as a cathode for Li-ion batteries.

Fe:
Fe(thd) 3 -purge -O 3 -purge P: Me 3 PO 4 -purge -(H 2 O + O 3 ) -purge ALD type growth was veried by basis in a supercycle of Fe-P deposition during 1000 cycles at 246 C by systematic variations of the length of the individual pulse cycles.The basic cycle and purge times for the parameter that was not under investigation were set at specic values: 3s Fe(thd) 3 -6s purge -6s O 3 -6s purge -6s TMP -6s purge -6s (H 2 O + O 3 ) -12s purge.The choice of deposition temperature and purge times was based on prior experience with the Fe 2 O 3 (ref.36) and LaPO 4 (ref.32) systems.The purge times were not subject to further optimization in the current study.The results from the investigation of the pulse times are shown in Fig.

5 .
The amorphous as-deposited lm crystallizes into trigonal FePO 4 (ref.45) aer annealing.The absence of impurity faces in the obtained diffractograms further indicates an equimolar ratio between Fe and P throughout the whole lm.

Fig. 1
Fig. 1 Growth rate per supercycle of 1 : 1 pulsing of Fe : P, as a function of the pulsing times of (a) Fe(thd) 3 and O 3 and (b) Me 3 PO 4 and (H 2 O + O 3 ), based on ellipsometry measurements.

Fig. 2
Fig.2Growth rate per subcycle and composition of deposited films as a function of deposition temperature for a pulsing ratio of 1 : 1 between Fe and P as measured by XRR and XRF, respectively.

Fig. 3
Fig. 3 Growth rate per subcycle and composition as a function of the percentage of Fe-subcycles as measured by XRR and XRF, respectively.

Fig. 4
Fig. 4 Film thickness and density measured by XRR as a function of the number of supercycles with 3 : 2 pulsing between Fe and P.

Fig. 5 X
Fig. 5 X-ray diffractograms of a 100 nm thick FePO 4 film.(a) As-deposited measured with GI-XRD; (b) and (c) heat treated for 2 h at 600 C in air and measured with XRD and GI-XRD, respectively.The vertical lines represent trigonal FePO 4 PDF 29-0715.

Fig. 6
Fig. 6 RMS roughness as measured with AFM as a function of deposition temperature for 100 nm as-deposited and annealed films on Si(111).

Fig. 7 Fig. 8
Fig. 7 Cyclic voltammetry curves of the amorphous FePO 4 thin film in the voltage range 2.3-4.0V at a scan rate of 0.1 mV s À1 .
using trimethyl phosphate (TMP, Me 3 PO 4 )/H 2 O and P 2 O 5 /H 2 O as phosphorous sources.Films in the Ca-P-O system were deposited using Ca(thd) 2 , TMP and O 3 . 31Recently ALD of the phosphate-containing materials LaPO 4 (ref.32) and LiFePO 4 (ref.28) has been described using the TMP/(H 2 O + O 3 ) phosphorus source together with La(thd) 3 /O 3 , Li(thd)/O 3 and Fe(thd) 3 /O 3 .Phosphates of both Al and Ti were reported recently using AlCl 3 and TiCl 4 , respectively, in combination with TMP, 33 and TiCl 4 with TMP/H 2 O. 34 Deposition of Li 3 PO 4 (ref.35) has recently been reported using TMP together with lithium hexamethyldisilazide or lithium tert-butoxide.
Sønsteby et al. observed a similar saturation of 55% in the La-P-O system.