Review of current progress in non-aqueous aluminium batteries Renewable and Sustainable Energy Reviews

Research on aluminium batteries is rapidly gaining momentum as a potential alternative to established battery chemistries such as lithium ion. Aluminium is abundant, recyclable, and due to its three-electron redox reaction it offers the potential for high specific energy and power. It can also be used as a metal negative electrode due to its dendrite-free plating behaviour at relevant conditions in room temperature ionic liquid electrolytes. With these electrolytes, a battery made entirely of abundant elements can also be envisaged. However, in order to commercialise aluminium batteries, researchers must still overcome formidable challenges, because no positive electrode material has yet demonstrated efficient reversible storage of aluminium ions. This review paper pro- vides a critical summary of the research to date. We present a discussion of the chemistry of the electrolytes, the deposition and dissolution behaviour of aluminium, and the various cathode materials that have been attempted. We also place non-aqueous aluminium batteries in context with other battery systems and provide an outlook of future research direction, and potential future applications. Using clear graphics to explain the various concepts, we intend this review to provide a broad and clear introduction to the field for researchers new to the area.


Introduction
In the effort to move away from fossil fuels, humanity urgently needs affordable, sustainable, long cycle life and high-performance energy storage. The integration of large percentages of intermittent renewable energy sources into electrical grids requires large scale storage to balance supply and demand [2]. Meanwhile, electric vehicles are creating demand for extremely high specific energy batteries to facilitate ranges comparable to current petrol and diesel vehicles (at least 320 km for about 50% of consumers to consider the range acceptable [3]). These must also become more affordable, as approximately half of the current electric vehicle cost is the battery pack [4]. To date, only nations that heavily subsidise their purchase (i.e. Norway) have achieved widespread uptake of electric vehicles [5]. It has been estimated that for electric cars to become price competitive in 50% of the US market, battery pack cost would have to fall from around $230/kWh in 2017 to $150/kWh [3].
Lithium-ion batteries currently dominate for new-build grid storage, electric vehicles, and electronic devices due to their outstanding specific energy and acceptable specific power. This position is reinforced by economies of scale and technological maturity resulting from nearly four decades of development for electronic devices and more recently for electric vehicles [6]. A common fear is that lithium supplies will soon run out; this fear is misplaced as mineral deposits are plentiful. However, price is rising because demand is outstripping supply from the cheaper brine extraction methods. These brines are typically found in the Andes (Chile, Bolivia) and the USA, and provide around half of the world's lithium, with increasing proportions coming from more expensive mineral deposits. Because this is making mineral deposits economic to mine, a lithium mining boom may subsequently drive down price [7].
Another major issue with lithium-ion batteries is the use of cobalt and nickel in the commonly used lithium-cobalt-oxide (LiCoO 2 ), nickelmanganese-cobalt (NMC) and nickel-cobalt-aluminium (NCA) positive electrodes. Lithium-ion batteries consume over 50% of the world's primary cobalt production, of which around 66% is mined in the Democratic Republic of the Congo. There is a well-reported history of human rights abuses in the nation, including widespread use of child labour [8]. Additionally, despite the fact that lithium-ion batteries contain valuable quantities of cobalt, nickel, lithium, iron, copper and aluminium, and that failure to recycle them can lead to environmental damage due to leaching of the toxic components, the difficulty of extracting these elements is so great that significant research effort is still required to make recycling economically viable [9]. It is likely that improvements in recycling technology combined with rises in the price of the constituent elements of lithium-ion batteries will eventually cause recycling to become economically viable, however, and recycling might then provide a substantial proportion of the raw materials for new lithium-ion batteries.
Fire risk is another major issue with lithium-ion batteries. There is inherent flammability associated with lithium-ion and the organic solvents in the electrolyte [10,11]. If the batteries experience an internal short circuit due to damage or dendrite formation, or are heated past a critical temperature, they may undergo exothermic breakdown, termed thermal runaway. This is a violent thermal event that often results in the destruction of the entire battery pack and associated device or vehicle [12]. This problem can be even worse for large scale recycling of lithium batteries where large quantities of cells are stored in one place and purposefully dismantled: a thermal runaway of a single cell can then trigger the destruction of the entire store [9].
All these reasons motivate research into other battery chemistries: even if alternatives cannot match lithium-ion on performance, perhaps they can offer cheaper, safer, more environmentally friendly and ethical options for applications such as grid storage. Aluminium is an attractive metal for a battery system for several reasons. It is low density, affordable, and is the most abundant metal in the Earth's crust [13]. It forms a trivalent cation (Al 3+ or Al(III)), releasing three electrons per ion relative to monovalent lithium's single electron [14]. As a result, aluminium has a specific energy of 3.0 Ah g − 1 , slightly lower than lithium metal at 3.9 Ah g − 1 , and also has an energy density of 8.0 Ah cm − 3 [13], higher than lithium metal at 2.06 Ah cm − 3 [15]. Additionally, aluminium can be repetitively stripped and deposited without forming dendrites that might cause short circuits [16], unlike lithium metal [17,18]. The currently used electrolytes in aluminium batteries -room temperature ionic liquids (RTILs) -are non-flammable [19], and there is no need for rare elements. There would be a strong economic case for recycling: recycled aluminium requires only 5% of the energy needed for primary production [20], and disassembly would be straightforward due to the non-flammable electrolyte and a plain aluminium electrode that could be relatively easily removed.
Aluminium batteries naturally find themselves compared to other lithium-ion alternatives such as alkali-ion (sodium and calcium), and multivalent metal batteries (magnesium, zinc). Sodium-ion batteries currently attract the most research attention of any lithium alternative battery due to the similarity to lithium-ion battery operation and the abundance of sodium. However, the larger size of the sodium ion causes poor cycle life, rate capability, and specific energy/energy density. Sodium is also a reactive metal, and the electrolytes are generally organic, so are also flammable. The specific energy is worsened by the inferior potential of Na|Na + compared to Li|Li + . The reversible and durable storage of sodium ions in the negative electrode is a major hurdle as sodium does not intercalate properly into graphite [21]. A prototype 18650 (18 mm diameter, 65 mm lengthstandard lithium-ion cell size) cylindrical sodium-ion cell has been demonstrated with an energy density of 90 Wh kg − 1 compared to commercial lithium-ion at 200-250 Wh kg − 1 [22,23].
Beyond sodium-ion, calcium batteries are in their infancy. The current generation of electrolytes cause a solid electrolyte interphase (SEI) layer to form on the surface of the calcium negative electrode. This causes very slow calcium deposition and dissolution. Additionally, little progress has been made on positive electrode selection [24]. Zinc secondary batteries use a metallic zinc negative electrode, but despite great research effort still suffer from dendrite formationa problem that can be avoided in batteries with a flowing electrolyte [25][26][27][28][29]. Finally, magnesium plates without dendrite formation, but the magnesium negative electrode causes extremely high internal resistance. No electrolyte with a full range of suitable properties has been found, and reversible positive electrode materials have yet to be identified [30]. As will be discussed, batteries with an aluminium negative electrode have been demonstrated that do not suffer from any of these issues.
This review is necessarily limited to non-aqueous aluminium batteries; however, it is worth mentioning research taking place on aluminium batteries outside this remit. Aluminium batteries with aqueous electrolytes are being explored (benefits include very high ionic conductivity and non-corrosive, non-toxic electrolytes). Conventional aqueous electrolytes where water is the solvent are not competitive on specific energy, partly because the deposition of aluminium occurs at a more negative potential than the evolution of hydrogen [31] (− 1.76 V vs. standard hydrogen electrode (SHE) [32]), so an anodic intercalation reaction at a less negative potential must be used [33]. These aqueous systems perform more like supercapacitors, demonstrating specific energy of 15  and negative electrode masses [15]. A more promising system has recently been reported using supersaturated solutions of AlCl 3 in water to allow deposition of Al 3+ onto an aluminium negative electrode with a graphite positive electrode storing chloroaluminate species. The super-saturation of the electrolyte prevents the evolution of hydrogen and provides an operation and performance similar to aluminium-graphite batteries (see Section 5.1.1), while addressing the cost issue of ionic liquid electrolytes [34]. Similarly work continues on metal-air batteries including aluminium-air [35]. Metal-air batteries can be characterised by offering very good performance as a primary battery and theoretical specific energy and energy density that exceeds any existing secondary battery. However, all metal-air batteries face the same serious issues in recharging (including irreversible consumption of the electrodes), and so currently cannot be described as being rechargeable [36].

Mode of operation of non-aqueous aluminium batteries
The most promising aluminium batteries work on a different principle to lithium-ion batteries. Lithium-ion batteries operate according to the well-known rocking-chair mechanism where lithium ions travel from the negative to the positive electrode as the battery is discharged [37] with the electrolyte acting only as a transport media. Ideally, aluminium batteries would work in a similar way because this would require minimal quantities of electrolyte. The positive electrode would largely determine the specific energy and energy density of the battery. However, attempts to store aluminium ions in a positive electrode have at best demonstrated unacceptably low discharge potential, capacity, coulombic efficiency and cycle life (see section 'Proposed positive electrodes storing Al(III)'). This is due to the trivalency of the aluminium ion, causing aluminium to have a strong charge interaction with the host material, and preventing reversible charge storage [38][39][40][41].
The main approach taken to avoid this problem is to store monovalent AlCl 4 − and Al 2 Cl 7 − anions from the RTIL electrolyte in the positive electrode instead of Al(III). At partial charge a mix of species is stored, but as the battery is fully charged the Al 2 Cl 7 − is depleted, so only AlCl 4 − is stored [13,42]. This leads to a fundamental difference in battery operation compared to lithium-ion batteries because these species are stored in the positive electrode during charge, not discharge [13]. Therefore, when aluminium batteries are charged, Al 2 Cl 7 − species in the electrolyte are split into aluminium ions, which are deposited, and AlCl 4 − , which is stored at the positive electrode. The full-cell equation is shown by Equation (1) [43]: The mechanism of a battery with an aluminium negative electrode, graphite positive electrode and a Lewis acidic AlCl 3 -[EMIm]Cl electrolyte is shown in Fig. 1 Because both electrodes in the aluminium battery store species during charge and release them during discharge, the electrolyte must hold all the electroactive species when the cell is fully discharged. Because the electrolyte can only support a certain range of composition, the specific energy and energy density of the battery are dependent on the amount of electrolyte [42], unlike rocking-chair batteries where only enough electrolyte is required to saturate the electrodes. This has led some researchers to suggest that the electrolyte is actually the negative electrode and the batteries should be named aluminium-chloride batteries [42]. However, we simply refer to these systems as aluminium batteries [45]. The dependence on the quantity of electrolyte means that some authors have started to calculate energy density of aluminium batteries using the following formula. F is Faraday constant (26.8 × 10 3 mAh mol − 1 ), x is the number of electrons involved in reducing 1 mol of AlCl 3 (0.75), r is the maximum molar ratio of the electrolyte (2), Q C is the specific capacity of the positive electrode (in mAh g − 1 ), and M(AlCl 3 ) and M(EMImCl) are the molar masses of AlCl 3 and EMImCl (133.34 g mol − 1 and 146.62 g mol − 1 respectively) [42].
This accounts for the entire volume of electrolyte required and the mass of active positive electrode in the discharged state (it is assumed that the fully discharged battery will have no active aluminium at the negative electrode at all). This limits the specific capacity of aluminium batteries to 48.6 mAh g − 1 (in the limit of a positive electrode with infinite specific capacity) [13]. Using the highest value yet reported of 192 mAh g − 1 , for a PEDOT positive electrode, gives a full battery capacity of 39 mAh g − 1 [46].

Electrolytes for non-aqueous aluminium batteries
Over the history of aluminium electrochemical deposition, electrolyte research focus has progressed from organic solvents [47] to room temperature ionic liquids (RTILs). Even within these categories, the choice of electrolytes that are able to reversibly deposit aluminium over a useful electrochemical range is extremely limited [48]. RTILs are a significant improvement over organic solvents [49], which are volatile, flammable and only allow low current densities [33,50], and to our knowledge organic electrolytes have never been used in aluminium battery research.
A suitable electrolyte for aluminium deposition must fulfil a number of criteria. The solute must be highly soluble in the solvent, but should not be held in solution so strongly that aluminium cannot be deposited. Reversible aluminium deposition/dissolution must occur within the electrochemical potential window bound by the limits at which electrolyte decomposition, electrode reaction with solvent, or other undesirable reactions occur. Finally, the electrolyte must have low viscosity at operating temperatures, in order to provide high ionic conductivity and mobility, so that the electrolyte does not contribute with a high resistance to the battery. It must also plate the aluminium negative electrode reversibly without dendrite formation [33,51].
It is worth clarifying the differences between ionic liquids, molten salts, and deep eutectic solvents. Ionic liquids are liquid salts, consisting exclusively of weakly coordinated complex ions that are liquid below 100 • C or at room temperature. Molten salts are liquefied salts at high temperatures (~800 • C) and deep eutectic solvents are mixtures with a melting point much lower than their anionic and cationic components have on their own. The type of ionic liquids that are liquid at room temperature are called RTILs, and as batteries typically operate relatively near to this temperature, these are suitable candidates for battery electrolytes [19]. In fact, it is loosely conventional to describe any molten salt that is liquid below 100 • C as a RTIL, but it is important to recognise that this distinction is arbitrary [52].
While tailoring of RTILs with various additives has been attempted for purposes such as reduced moisture sensitivity, these have not yet been explored in aluminium batteries. Extra additives would lower the specific energy, and it makes sense to avoid moisture completely through experimentation in inert environments [64]. Difficulties with AlCl 3 -[EMIm]Cl are mainly in its preparation, which must be done in an inert atmosphere due to the hygroscopicity of AlCl 3 , and slowly due to the exothermic nature of the mixing process. AlCl 3 is low cost and widely available, but [EMIm]Cl is still a relatively niche component with a small supply chain, although it is made of abundant elements and large-scale manufacturing could probably make it economical. It is also hard to purify, and there is a lack of detailed information on its toxicity due to its use not yet having become widespread [60,62,63]. Chloroaluminate Lewis acidic RTILs are highly corrosive, and AlCl 3 reacts violently with water and is a severe irritant [49]; however these difficulties are mitigated by handling in a glovebox.
For optimal battery performance, moisture content should be as low as possible, again implying use of a glovebox. Even a low level of moisture significantly impairs coulombic efficiency due to HCl and Cl 2 evolution [43]. As more water is added, the ionic nature of the mixture is reduced because large hydrogen-bonded clusters and networks formed of water molecules begin to form, separating the ions from each other and significantly altering the chemistry [65]. Water reduces the electrochemical potential window of RTILs; for example, from 4.10 V to 1.95 V in [BMIM][BF4] when 3 wt% water is added. Finally, water can also react with ILs to create different electroactive species [66].
RTILs have a wide liquidus range that goes down to around − 80 • C [63], and AlCl 3 -[EMIm]Cl is thermally stable to +95 • C; above which highly coloured decomposition products form [33,67,68]. While this wide temperature range seems appealing for batteries, the viscosity and hence ionic conductivity of RTILs is strongly dependent on temperature [69]. At room temperature, a typical aqueous lead acid electrolyte conductivity is around 730 mS cm − 1 , while ionic liquids and lithium-ion electrolytes have conductivities in the region of 10-20 mS cm − 1 [19]. It is possible to enhance the conductivity and reduce the viscosity of these melts without changing the types of species formed by adding organic solvent [55,70], however this would reduce the 'greenness' of the mixture. Over the full liquidus range, RTILs have very low vapour pressure. This provides low volatility and is a major reason why ionic liquids are considered 'green solvents'. This is attributable to high cohesive energy density, caused mainly by strong Coulombic interactions between the components of the ionic liquids [71].

AlCl 3 -[EMIm]Cl chemistry
The speciation behaviour of AlCl 3 -[EMIm]Cl is largely responsible for the operation of the aluminium battery [49], so warrants detailed discussion. While [EMIm]Cl will always dissociate to [EMIm] + , the type of chloroaluminate species formed, and the Lewis acidity, depends on the molar ratio of AlCl 3 to [EMIm]Cl. The parameter r is the molar ratio of AlCl 3 to [EMIm]Cl. For example, r = 2 means 2 mol of AlCl 3 to 1 mol The deposition of aluminium is only possible from r > 1.1 liquids [72], where a large proportion of Al 2 Cl 7 − is present. The saturation limit of AlCl 3 in [EMIm]Cl occurs at r = 2 and is shown in (c) [38]. As discussed, in the best performing aluminium batteries, chloroaluminate species are stored in the positive electrode during charge [42]. This causes r to increase as the battery is discharged, reaching a maximum when the battery is fully discharged and a minimum when the battery is fully charged. It is only possible to deposit aluminium for r values between 1.1 and 2 [42,72]. This influences the chloroaluminate ions available to store in the positive electrode. At r = 1.1, the battery is fully charged and Al 2 Cl 7 − is depleted, so only AlCl 4 − is stored [42] while at partial charge levels both are present [13]. The balance of species present is driven by Equations (3)-(5). K is equilibrium constant and the given values are specific to AlCl 3 -[EMIm] Cl at 40 • C [73].
The very high equilibrium constant of Equation (3) shows that when Cl − and AlCl 3 are mixed in equal proportions, the result will be the overwhelming formation of AlCl 4 − . The intermediate equilibrium constant of Equation (4) shows that when AlCl 4 − is combined with additional AlCl 3 , a mixture of AlCl 4 − and Al 2 Cl 7 − is formed. Finally the low equilibrium constant of equation (5) shows that when further AlCl 3 is added to a solution already rich in Al 2 Cl 7 − , there is a low tendency for Al 3 Cl 10 − to form [74]. These equations combine to give Lewis acid/base behaviour for AlCl 3 -[EMIm]Cl that depends on the concentration of AlCl 3 to [EMIm]Cl, whereby two amphoteric, uncharged solvent molecules undergo a Lewis acid transfer, as shown in Equation (6) [73].
In a Lewis neutral AlCl 3 -[EMIm]Cl liquid, the electrochemical limits are due to EMIm + being reduced and AlCl 4 − being oxidised. This gives the widest electrochemical windowfrom around − 2 V to +2.7 V vs. Al| Al(III), providing the electrolyte is free of moisture. For a Lewis acidic liquid however, Al 2 Cl 7 − reduces at 0 V to deposit aluminium on the electrode. In Lewis basic liquids, EMIm + becomes the cathodic limit as Al 2 Cl 7 − is not present, but the presence of Cl − means that Cl − evolves before AlCl 4 − , reducing the anodic limit to about − 1 V vs. Al|Al(III) [73].
Because EMIm + is reduced before AlCl 4 − , aluminium cannot be deposited from AlCl 4 − within the electrolytic limits of the electrolyte, which is why only acidic electrolytes can be used for aluminium deposition/dissolution [67]. Above the discharge potential window of the electrolyte, side reactions involving Cl 2 evolution from the electrolyte begin to occur, particularly above 2.6 V vs. Al|Al(III) [43,75]. A graph showing the electrochemical potential range of basic, neutral and acidic AlCl 3 -[EMIm]Cl is shown in Fig. 3 [73].

RTIL gel electrolytes
Extending the electrochemical stability window of the ionic liquid electrolyte to -2 V vs. Al|Al(III) would stabilise the [EMIm] + sufficiently to allow deposition of aluminium from AlCl 4 − . Because the required quantity of electrolyte drives the specific energy and energy density of non-aqueous aluminium batteries storing chloroaluminate species at the positive electrode, this could substantially improve overall battery performance.
One promising way to achieve this is by the addition of a gelification component e.g. a polymer, though the amount must be carefully chosen as it will compromise the ionic conductivity [76][77][78][79][80]. The simplest approach is to introduce a small amount of non-reactive polymer into the ionic liquid, tangling the molecules of the ionic liquid, restricting their movement enough to give the electrolyte rubber-like or gel-like properties. These are referred to as gel electrolyte or ion gels, and suitable polymers include poly(phenylene oxide) (PPO), poly(methyl methacrylate) and poly(vinylidene fluoride) [77,81]. However, these mixtures can experience problems such as phase separations, where some of the ionic liquid becomes separated from the polymer, which can lead to leakage [77]. It has been shown that simply by adding 10 wt-% PEO to an r = 1 AlCl 3 -[EMIm]Cl electrolyte, the potential stability window of the electrolyte can be increased by 1 V, but the coulombic efficiency of aluminium deposition/dissolution is brought down to below 60% [82].
Alternatively, it is possible to polymerize a component of the ionic liquid itself, for example the imidazolium cation, which would produce a polymeric ionic liquid (PIL). These also have low ionic conductivity (<10 − 6 S cm − 1 ) relative to ILs (10 − 2 S cm − 1 ), raising the possibility of further tailoring by combining PILs with RTILs to gain a further intermediate compromise in electrolyte characteristics (10 − 3 to 10 − 5 S cm − 1 ). Due to being molecularly close to ILs in the first instance, PILs mix better with ILs than polymers, potentially helping to avoid some of the aforementioned weaknesses with polymer/IL combinations [77].

Deep eutectic solvents
A Deep Eutectic Solvent (DES) is a mixture of at least two components, which independently might be solids or liquids at room temperature, that when combined in a particular (eutectic) range of composition have a melting point up to 200 • C lower than either constituent on its own [63,83,84]. Similar to ionic liquids, they contain a combination of Lewis or Brønsted-Lowry acids and bases that when solvated in one another form a range of species that may be anionic and/or cationic [85]. DESs are made of readily available, abundant, cheap materials, and are synthesized through simple mixing of the components [86], generating only moderate heat [83]. By intent, many DESs are more environmentally friendly than ILs, as in addition to having low vapour pressure, they commonly contain no components of known or unknown toxicity [63,84]. Indeed, there is a subclass of DESs which are particularly benign, termed natural deep eutectic solvents (NADES), which consist only of bioavailable complexing agents such as choline, amino acids or their metabolites such as urea or sugars [84]. It may be that DESs are entering a period of rapid growth in research interest, as for ionic liquids in the 1990s [84]. DESs have already been used in a range of applications including electropolishing, synthesis of gold nanoparticles, metal oxide processing and biodiesel purification  [73]). [63].
DESs are little researched for aluminium deposition [83] despite the work of a few key authors [87] but initial research is somewhat encouraging, with aluminium deposition demonstrated from DESs including AlCl 3 -4-propylpyriinde (4-Pr-Py) [88], AlCl 3 -acetamide [89], and AlCl 3 -urea with 1,2-dichloroethane [90]. The lack of Al 2 Cl 7 − means that AlCl 3 -4-Pr-Py is less water sensitive than AlCl 3 -[EMIm]Cl [88]. DESs have reduced electrochemical redox potential window relative to RTILs [63,91] and poor ionic conductivity. Organic solvents could improve this but would make DESs less environmentally friendly [85,90]. The composition range for aluminium deposition is also less (r = 1.1 to 1.5 for AlCl 3 -acetamide) than for AlCl 3 -[EMIm]Cl [89]. While DESs are generally less affected by moisture than chloroaluminate ionic liquids [63], these benefits are limited for those DESs containing AlCl 3 , which to date is all those from which aluminium has been deposited [87]. While worse performing than RTILs at the moment, it is possible that DESs may see developments in the future that increase their appeal.

Aluminium deposition and dissolution
One of the key benefits of aluminium is its dendrite-free behaviour when deposited/dissolved at moderate current densities onto a suitable substrate such as aluminium, tungsten or planar vitreous carbon in chloroaluminate RTILs, a behaviour which it shares with magnesium [92]. The use of plain metal anodes is highly preferable to storing the active ions in a host negative electrode due to savings in mass, complexity and volume. Many metals used in batteries, including sodium, zinc and lithium, tend to form dendrites when the metal is repeatedly stripped and plated during charging [17,28,93], which has for example prevented the development of safe lithium metal batteries [17].
When aluminium is deposited from AlCl 3 -[EMIm]Cl, dendrites only form above current densities of 100 mA cm − 2 [56]. This is due to plating occurring faster than diffusion can supply species from the electrolyte, causing preferential plating onto protruding parts of the electrode, leading to dendrite growth [27,94]. Fortunately, current densities are much lower than this in batteries (80 C can be achieved at 10 mA cm − 2 ) [95]. An Al-Ni 3 S 2 battery with a LiAlCl 4 -NaAlCl 4 -NaAlBr 4 -KAlCl 4 non-organic electrolyte functioned without dendrites at 70 mA cm − 2 and below, and still without dendrites but with a slightly less dense plating from 100 to 706 mA cm − 2 [96]. When electrodepositing aluminium from r = 2 AlCl 3 -[EMIm]Cl, dense plating has been achieved up to 70 mAh g − 1 [56,72]. Clearly, the choice of electrolyte is critical in obtaining a dendrite-free plating behaviour.
Aluminium negative electrodes in chloroaluminate electrolytes do not form a solid electrolyte interphase layer (a layer that forms on the electrode surface and separates it from the electrolyte) [97], so deposition and dissolution occur at the electrolyte/aluminium interface. Aluminium diffusion energy is less than 1/3 of that predicted by conventional diffusion mechanisms. This is due to an exchange-diffusion mechanism, where an aluminium atom with metallic bonding on the surface is extracted in exchange for another aluminium atom becoming covalently bonded [98][99][100]. This mechanism is the only exception known to the hard-sphere model of diffusion where an adatom will follow the lowest-energy diffusion route along a surface channel, and applies only to the (001) surfaces of platinum, nickel, aluminium and iridium [99]. Aluminium self-diffusion energy at 0.17 eV is nevertheless higher than either magnesium at 0.02 eV, which reliably plates without dendrite formation [101], or 0.14 eV for lithium and 0.16 eV for sodium, which do not [102]. However, the real benefit of the exchange-diffusion mechanism is that it has been shown in simulations to cause aluminium atoms to move down from protrusions and spread out across the plating surface in even layers, regardless of the diffusion energy [100].
Aluminium crystal structure is face-centred cubic (fcc) [92], which has a high coordination number of 12 [103], also supporting the development of a stable crystal structure through increased lattice stability. This is like magnesium, which also has a co-ordination number of 12 due to its hexagonal close packed (hcp) structure. In contrast, lithium and sodium are both body-centred cubic (bcc) with a lower co-ordination number of eight [93,104]. The mechanism is that higher co-ordination number metals have a higher driving force to re-organise into high-coordination configurations, which tend to be achieved better by a dense, even layer rather than through dendrite formation [105].
Most papers reporting on aluminium deposition from ionic liquids found that instantaneous nucleation occurred (from AlCl 3 -BMIC, AlCl 3 -[EMIm]Cl [56,106] and AlCl 3 -TMPAC [67]), with only one paper reporting progressive nucleation (also from AlCl 3 -[EMIm]Cl) [107]. When plated from ionic liquids, aluminium follows a hemispherical (3D) growth process, which means that as the initial nuclei grow on the surface, the electroactive species are drawn from the electrolyte, creating a diffusion zone with a deficiency of electroactive species in the electrolyte. The diffusion zone grows ahead of the hemispherical nuclei into the electrolyte (hence 3D growth), and creates a 'shadow' region of electroactive species depletion on the electrode face, creating an exclusion zone where nucleation can no longer take place so that new nucleation is forced to take place elsewhere [107]. The hemispherical nuclei grow by diffusion until they meet. At this point, they coalesce into a layer (going from a 3D growth mechanism to a 2D planar growth mechanism) [67].
Once a crystal structure is established in the layer, further growth of the layer may be limited by either diffusion or kinetic rates, whichever is slowest [108]. At all but the lowest current densities [97], aluminium layer growth is diffusion limited [67,97,107]. Aluminium can be deposited either directly onto a suitable substrate such as vitreous carbon or an aluminium seed layer [42]. High coulombic efficiency of stripping and plating have been achieved in both organic and RTIL electrolytes, with >99% achievable in both AlCl 3 -[EMIm]Cl [109] and some organic solvents [110]. A significant factor when dealing with aluminium is the oxide layer that forms upon exposure to air, which reaches a thickness of around 32 Å after just 10 min of exposure [111]. This is an important consideration for manufacturing aluminium batteries, as the oxide layer is insulating, requires additional processing steps to remove, and requires that oxygen must be excluded during battery assembly to avoid its formation [16,56]. This oxide does not build up if the aluminium is deposited in an inert atmosphere such as inside a glove box. Researchers have also tried to remove the oxide layer from an aluminium negative electrode before experimentation, which is very difficult outside of an inert atmosphere because of the speed with which aluminium re-oxidises in the presence of any oxygen [56,97]. This is an important issue to bear in mind when analysing or carrying out research into aluminium deposition/dissolution.
The cyclic voltammogram of aluminium deposition is shown in Fig. 5. The C (cathodic) area corresponds to the deposition of aluminium and the A area (anodic) to the dissolution; the subscript 1 refers to deposition and dissolution of nanocrystalline aluminium, while 2 refers to deposition and dissolution of microcrystalline aluminium [112]. Above 0.8 V, passivation of the electrode causes the current density to drop to zero [56].

Positive electrode materials for non-aqueous aluminium batteries
Positive electrodes for non-aqueous aluminium batteries can be divided into two categories: those that store chloroaluminate species, which have been shown to have high performance, and those that store aluminium, which have not. However, research continues into positive electrodes that store aluminium because if good performance could be obtained, the quantity of electrolyte required would be dramatically reduced, as it would no longer have to buffer a compositional change, and so the specific energy of the battery would be greatly increased. Of the materials discussed, graphite, conducting polymers and iron chloride all store chloroaluminate species. Sulphur, vanadium-based positive electrodes, and transition metal sulphides, and anthraquinone electrodes all store aluminium. Proposed positive electrodes that demonstrated no actual storage behaviour in experiments, such as MnO 2 , have been omitted [113].
The highest observed specific energy of an aluminium battery with a graphite negative electrode is about 68 Wh kg − 1 . This takes into account the required quantity of electrolyte mass (AlCl 3 -[EMIm]Cl, r = 2 when fully discharged), as calculated via Equation (2) [114]. Applying the same analysis to PEDOT gives a specific energy of 50 Wh kg − 1 [46]. In comparison, lithium-ion, based on active components only, can achieve specific energies of over 500 Wh kg − 1 [10], while commercial lithium-ion cells (including the mass of the battery pouch or case) achieve around 250 Wh kg − 1 [22]. Fig. 6 demonstrates full cell specific energies vs. demonstrated cycle life for some of the discussed positive electrode materials, with lithium-ion included (active materials only) for comparison.

. Graphite
The first attempt at a rechargeable aluminium battery with a graphite positive electrode was in 1988 [72]. There were no further publications until 2013 [118], but then took off in earnest from 2015 [43], when Lin et al. produced a battery with a graphitic foam positive electrode that had a cycle life of 7500 cycles, a discharge potential near 2 V vs. Al|Al(III), and a coulombic efficiency of 98%. Since then several papers, both experimental and theoretical, have been published investigating various graphite positive electrodes (see Table 1 for the characteristics of many of the reported aluminium-graphite batteries). With a flat discharge potential profile of 1.8-1.9 V vs. Al|Al(III) [43,114,119,120], >65 C rate capability, coulombic efficiency > 98% [43], round trip efficiency > 80% [42,114] and achievable specific energy of around 65 Wh kg − 1 [43,114], aluminium-graphite batteries already compare favourably to lead acid and vanadium redox flow batteries, indicating their potential for large scale grid storage [121]. The cell equation for the Al-graphite battery is shown by Equation (8) [43]: Graphite is used ubiquitously in lithium-ion batteries as an intercalation negative electrode [37]. In aluminium batteries, however, it is used as an intercalation positive electrode with an aluminium negative electrode. While aluminium cannot intercalate into graphite, chloroaluminate anions can [122,123]. The charge/discharge of a battery with r = 1.5 AlCl 3 -[EMIm]Cl electrolyte and a pyrolytic graphite positive electrode undergoing galvanostatic cycling at 25 • C is shown in Fig. 7, which shows the initial cycle development (a), very flat discharge potential (a), good life cycle stability (b) and high coulombic efficiency (c) of the aluminium-graphite battery.
While compact lithium ions intercalate easily into round, smooth graphite particles, the much larger chloroaluminate species work far better with a low-defect flaky or foamy morphology (see Table 1). This is because the positive electrode must easily expand to accommodate bulky anion intercalation between the layers of graphite. Pristine (very low defect) graphite flakes allow up to 80% geometric volumetric expansion and fast rates (notably, this is an 80% expansion of the entire positive electrode, which would be difficult to accommodate in a commercialised cell) [42]. Meanwhile, pyrolytic graphite has a very low rate capability and around half the capacity due to covalent bonding between the layers constraining expansion [43]. The high rate capabilities and flat discharge potential profiles observed with graphitic foam [43] and graphite flakes [114] are explained by the fact that all of the possible intercalation positions have very similar and very low energy, so the barrier to AlCl 4 − diffusion in graphite is very low (10 × lower than for lithium) [124]. Graphite has a planar structure, and is described in terms of gallery height, which is the distance (d) between the planar graphite layers (3.34 Å for un-intercalated graphite), and in terms of intercalation stage. This is the period of the repeating sequence of single occupied layers separated by a number of unoccupied layers. Fig. 8 shows a stage 1 single layer intercalation in A, a stage 4 single layer intercalation in B, and a stage 4 double layer intercalation in C.
Researchers have used density functional theory (DFT) and experimental analysis to understand the performance of chloroaluminate intercalation in graphite and to establish the preferred intercalation. One publication suggests that graphite exists in a planar quadrangle structure in graphite [124], but most papers conclude that tetrahedral AlCl 4 − intercalation occurs [120,125,126].
When intercalation occurs, the gallery height of graphite increases. Both modelling and experimentation suggest an increase from a welldefined initial value of ~3.34 Å (all graphite types, experiment and modelling), to values including 5.7 Å (experimental result, pristine graphite foil) [43], 6.025 Å (DFT model, bulk graphite) [124], 8.81 Å [125] (DFT models, bulk graphite), and 8.9-9.3 Å (DFT model, bulk graphite) [122]. These papers all presented DFT modelling results showing that Stage 4 intercalation is the preferred state, although one paper [126] further suggested that Stage 4 would be preferred initially, but as the positive electrode reached full capacity it would shift to Stage 1 intercalation. For bulk graphite, capacities were estimated at ~70 mAh g − 1 [125,126], close to the theoretically observed values from Ref. [43] for pyrolytic and natural graphite. In contrast, another paper presented strikingly different results, achieved by modelling the graphitic foam from Ref. [43] as a few-layered structure. The diffusion rate was very sensitive to the number of adjacent layers modelled: the rate for 4-layer foam was 48 × faster, and two layers 225 × faster than for bulk graphite. This paper also proposed stage 3 and 4 double stacking even in bulk graphite, with gallery height up to 12.4 Å for bulk graphite and 13.3 Å for graphitic foam, and capacity up to 117 mAh g − 1 [127]. These results are closer to those obtained for types of graphite flakes (124-142 mAh g − 1 ) [42,114], implying that the DFT modelling approach must fully account for the positive electrode microstructure, because gallery height expansions of this kind cannot be expected to be accommodated by the bulk graphite microstructure. This wide range of results suggests that a consensus has not yet been reached on the mechanism of AlCl 4 − storage in graphite; this is partly because of comparison of models and experimental results carried out on different forms of graphite. Modelling has also not yet accounted for the cycling effects on graphite: initial cycles demonstrate poor reversibility and capacity [72,119], with initially large and then gradual improvements over subsequent cycles, as chloroaluminate anions and EMIm + cations forcibly alter the microstructure, with some remaining in the positive electrode permanently. The scale of the improvement can be as large as 45 mAh g − 1 on the first cycle to 65 mAh g − 1 on the 500th cycle [119]. While positive electrode capacity and discharge potential are highest at around r = 1.3-1.5 [42,43,114], this is mooted by the requirement to cycle between r = 2 and r = 1.1 to maximise specific energy. The use of r = 2 electrolyte is required to achieve 65 Wh kg − 1 , which is close to the theoretical maximum possible using aluminium-graphite batteries with RTIL electrolyte [42,114]. The future of graphite intercalation looks to more exotic architectures of carbon to achieve greater specific energy. One potentially promising area of research is single walled carbon nanotubes, which have the advantage of potentially being able to be sized optimally for AlCl 4 − intercalation. Modelling has indicated the potential for a 275 mAh g − 1 specific capacity [128]. However, the improvements achievable from novel carbon architectures are marginal due to the dependency of the specific energy on the amount of electrolyte present. Currently, flaky graphite positive electrodes offer the best performance overall in non-aqueous aluminium batteries due to their high average discharge potential, long cycle life, and reasonable specific capacity.

Conductive polymers
The field of conductive polymer research has been active for more than 40 years, since it was discovered that conjugated polyacetylene chains had semiconductor behaviour which could be enhanced via exposure to chlorine or bromine gas [129]. While polyacetylene was the first to be discovered, it was unstable in oxygen, and experienced decreasing conductivity with time. Further development led to (amongst others) the more stable polypyrrole and polythiophene, and then to poly-3,4-ethylenedioxythiophene (PEDOT). Conductive polymers have the advantage of low density compared to metals (1.46 g cm − 3 for PEDOT relative to 8.96 g cm − 3 for copper), offering the opportunity for high specific energy electrodes [130]. So far, conducting polymers have found applications in organic solar cells, organic light emitting diodes, and various kinds of sensors and processors, due to the combination of optical and electrical characteristics [131]. PEDOT is also showing promise in supercapacitors [132].
Conductive polymers store charge via oxidation of the pi-conjugated carbon backbone. As electrons are removed, the charges localise at specific points along the polymer chain due to energetically favourable relaxation of the molecular structure; these 'defects' cause geometry deviations from the equilibrium neutral geometry that each affect several chain units. However, they remain mobile. When anions are introduced however, they coordinate near the polymer (Fig. 8 indicates the typical arrangement of an AlCl 4 − anion near a PEDOT chain) to counterbalance the missing electrons and balance the polymer charge. This effectively pins the localised charges in place, helping to spread the charges throughout the polymer bulk and increasing conductivity, which occurs due to electron transport between defects [133]. The exact mechanism of electrical conduction in polymers is not entirely defined, partly because it actually involves several mechanisms [134][135][136]. Fig. 9 illustrates the storage of AlCl 4 − in PEDOT, while the reaction is given in Equation (9) [95]:

3[MonomerUnit]
The possible range of concentration of inserted anions in a conducting polymer is large (possibly up to one anion per chain unit) [46,138], which allows a conducting polymer to store anions in a cathodic capacity [2]. Anion insertion into polymers is analogous to anion intercalation into graphite. Intercalation involves storage of an ion in specific sites in the crystal lattice or between layers of the material, whereas insertion involves storage near preferential locations on a polymer which may have an amorphous or partially crystalline structure. Polypyrrole, polythiophene and PEDOT have all been trialled in non-aqueous aluminium batteries and are discussed here [13,46]. PEDOT is synthesized from polythiophene. Thiophene is sourced from fossil fuels, from which it is removed by necessity to reduce sulphur emissions; therefore it is abundant, maintaining the benefits of aluminium batteries [139,140].

Polypyrrole and polythiophene.
Hudak [13] investigated polypyrrole and polythiophene electrodes with an aluminium negative electrode and AlCl 3 -[EMIm]Cl r = 1.5 electrolyte. Polypyrrole was electropolymerized and the counterions inserted in the same electrolyte using galvanostatic polarization. Elemental analysis was used to determine the molar ratio of chlorine to aluminium, and the ratio of the two chloroaluminate species present was calculated to be 77.7 mol % AlCl 4 − and 22.3 mol % Al 2 Cl 7 − . Furthermore, the percentage of anions relative to polymer units was found to be 24.6%. The polymers stored charge via a faradaic redox reaction until the redox potential of this process was exceeded. Above this, non-faradaic charge storage occurred, likely via a double-layer capacitor effectin its neutrally charged form, the polymer is non-conductive, so non-faradaic charge storage can only take place once it has been oxidised. In addition to having better coulombic efficiency (>98% vs. > 96%), polypyrrole withstood over-oxidation better than polythiophene, however, polythiophene had a slightly higher average discharge potential (1.47 V vs. 1.42 V). The charge/discharge is broadly reversible (>96% coulombic efficiency) so long as the discharge potential is kept between 0.1 and 2.0 V vs. Al|Al(III). Above 2.6 V, the authors proposed that the films reacted with the chloroaluminate species. Even above 1.5 V vs. Al|Al(III) however, some irreversibility was seen. In conclusion, maximum specific energies of 44.4 Wh kg − 1 for polythiophene and 46.4 Wh kg − 1 for polypyrrole were calculated based on the electrolyte and the active electrode mass. The specific capacity of the polymer films was estimated to be 30-100 mAh g − 1 , meaning that polypyrrole and polythiophene fall behind graphite flakes in terms of both average discharge potential and specific positive electrode capacity [13].
PEDOT is extremely stable when charged and discharged, and has a moderate band gap [142,143]. It can be synthesized to have higher potentials in an aluminium battery than polypyrrole or polythiophene [144]. PEDOT has found application in anti-static coatings, supercapacitor electrodes, hole injection layers in OLEDs and organic solar cells [145]. PEDOT can either be made into a free-standing film [77,146,147], or it can be combined with other materials such as non-conductive polymers to improve its mechanical properties [130]. All these properties point to PEDOT as a candidate positive electrode for aluminium batteries.
A PEDOT positive electrode can be electropolymerized at up to 2.3 V vs. Al|Al(III) in AlCl 3 -[EMIm]Cl before reaching the electrolytic breakdown of the ionic liquid, showing that it may have the potential to withstand even higher potentials if a suitable electrolyte is found. The charge/discharge potential of the synthesized electrode in a battery is limited by the maximum potential reached during electropolymerization. Higher potential electropolymerization is theorized to lead to a denser, cross-linked network of longer PEDOT chains which have higher potential anion insertion sites. The mechanism is not yet confirmed [80]; it is anticipated that computational techniques may reveal the mechanisms behind this phenomenon.
The discharged PEDOT film has a granular structure. As it is charged, it undergoes morphological changes, where the grains first expand, then merge. When the electrode is discharged, a new granular structure is formed. This mechanism is illustrated in Fig. 10 [148].
Improved structural stability and surface area of a PEDOT electrode can be achieved by using a three-dimensional reticulated vitreous carbon (RVC) substrate instead of planar vitreous carbon. The PEDOT forms a uniform film over the surface of the RVC without blocking the holes in the macrostructure, allowing for much greater surface area and improved mechanical properties relative to a planar film. The electropolymerization process was successfully reproduced on 3D RVC. The optimum method is cyclic voltammetry, from − 0.5 V to 2.6 V vs. Al|Al (III). This is hypothesised to be due to the formation of anion paths in the polymer during repeated cycling, whereas at constant current or potential, there is no anion removal during electropolymerization. With this method, the PEDOT electrode has around 45% higher capacity than the planar carbon electrode, and has demonstrated structural stability over 250 cycles, with a coulombic efficiency of 94%. The use of a porous carbon current substrate is convenient for battery packaging, because it provides a fixed macroscopic geometry that does not change during chargingthe PEDOT can expand within the pores inside the carbon. However, the latest work still demonstrates large pores of unused space in the RVC substrate, and the opportunity for significant further optimisation [144].
The charge and discharge curve of an aluminium-PEDOT battery is  shown in Fig. 11. The discharged PEDOT film is compact and cannot accept the insertion of anions; this is section I and corresponds to a surface charging mechanism. The PEDOT oxidises and swells during charging, and once it reaches section III the swelling means it can accept large-scale insertion of anions. Section II corresponds to a transition between these two stages [46].
Overall, PEDOT appears to offer an improvement over polypyrrole or polythiophene due to increased chemical stability (it is less vulnerable to reacting at high potentials), and higher specific capacity of the positive electrode (192 mAh g − 1 relative to 30-100 mAh g − 1 ). Recent work indicates that the specific capacity of PEDOT is higher than graphite (191 mAh g − 1 vs. 142 mAh g − 1 ), but the reduced average discharge potential of 1.3 V means the specific energy of the whole battery is lower (50 Wh kg − 1 vs. 65 Wh kg − 1 ) [13,43,46,114].

Iron chloride
There are potential benefits to storing Cl − instead of AlCl 4 − , chiefly that Cl − is smaller [2,72], assuming that Cl − can be extracted from Al 2 Cl 7 − species in the acidic electrolyte. This would have to be by preferential reaction as free Cl − is not present in acidic chloroaluminate electrolytes; researchers who intended to intercalate Cl − ions into graphite or conductive polymers either did not carry out analysis to prove Cl − had been stored [72] or later found that it was chloroaluminate anions that actually provided the charge storage [2]. Perhaps the most persuasive example of Cl − storage used an FeCl 3 positive electrode that was reported to convert to FeCl 2 on discharge. The FeCl 2 was impregnated in reticulated vitreous carbon in AlCl 3 -[E-MIm]Cl at a 1.38 M ratio. A stainless steel current collector was used. A full cell discharge potential of 1.85 V vs. Al|Al(III) was found, and the researchers proposed Equation (10) as the cell reaction [45]: However, the positive electrode was not analysed post-discharge to assess if FeCl 2 to FeCl 3 had actually occurred [45]. Stainless steel is also known to be electroactive in AlCl 3 -[EMIm]Cl, which may also have contributed [149]. The achieved energy density was not stated, and because the positive electrode was soluble in the electrolyte, the battery was prone to self-discharge, making it unsuitable for further investigation [45].

Positive electrodes storing Al(III)
The materials discussed in this section have been claimed to intercalate Al(III) directly. While some researchers reviewing the literature up until very recently (2017) did not consider Al(III) to ever have been proved to be successfully intercalated into any positive electrode material [75,149], we believe there is some evidence that Al(III) intercalation has taken place despite the difficulties of intercalating polyvalent ions [13]. Because the electrolyte in these systems is not depleted during charge or discharge, these papers tend to use RTIL electrolytes with r of 1.1-1.3, close to minimum Lewis acidity required aluminium deposition. In this section, sulphur, vanadium-based positive electrodes, and transition metal sulphides are discussed. On a side note, Al(III) intercalation has been successfully shown in aqueous aluminium batteries [150], however, this material has not been demonstrated in ionic liquid electrolytes [151].

Sulphur
As a positive electrode material operating by a conversion reaction (stored ions react to form a new chemical phase with the positive electrode), sulphur is now being considered for use in aluminium batteries. As expected, the latest research on aluminium-sulphur makes use of chloroaluminate ionic liquids [116,[152][153][154]. The reaction is as shown in Equation (11): For lithium-ion and sodium-ion batteries, sulphur is one of the most promising materials for moving forwards in terms of specific energy from the current generation of metal oxide positive electrodes. Although it offers increased specific energy and is highly abundant, it has poorer energy density than metal oxide positive electrodes. Sulphur has also been attempted as a positive electrode for aluminium batteries due to its theoretical specific energy (based on total electrode mass) of 1340 Wh kg − 1 based on a discharge potential of 1.1-1.25 V vs. Al|Al(III)significantly higher than current lithium-ion batteries [116]. A primary aluminium-sulphur battery has been demonstrated attaining very close to this theoretical specific energy [152]. However, as with all other positive electrodes that store aluminium, aluminium-sulphur is characterised by poor cycle life (a few tens of cycles compared to thousands for positive electrodes storing chloroaluminate anions). To improve the kinetics of the AlS x oxidation reaction, various strategies have been attempted. These include encapsulating sulphur into the micropores (2 nm) of activated carbon, achieving a cycle life of 20 with capacity retention of 65% [116], and incorporating Li + ions into the electrolyte, achieving a cycle life of 50 cycles with a capacity retention of 60% [154]. been theorized that Al(III) could be inserted via intercalation according to the following full-cell reaction shown in Equation (12):

Vanadium
An aluminium battery with AlCl 3 -[EMIm]Cl electrolyte and Li 3 VO 4 positive electrode was reported with an initial discharge capacity of 137 mAh g − 1 which dropped to 48 mAh g − 1 after 100 cycles. After the initial loss of capacity was accounted for, the battery was reported to have ~100% coulombic efficiency at an average cell discharge potential of <1 V vs. Al|Al(III) [38]. This study raises several questions. A stainless steel current collector was used, known to exhibit redox behaviour due to Fe and Cr redox reactions, producing similar discharge potentials to those found here [149]. It is suggested that V 4+ is oxidised to V 5+ during the charge process. Additionally, the Li 3 VO 4 was completely coated with approximately 5 nm of carbon, both graphitic and disordered types, strongly indicating the potential for some chloroaluminate redox reaction with graphite. The initial charge behaviour indicates that the structure has been intercalated and altered by the presence of anions, suggesting a mixture of graphitic storage and stainless steel electro-activity. Finally, the use of vanadium and lithium, neither of which are abundant elements, lowers the appeal of this positive electrode material for post-lithium batteries [38]. 5 . V 2 O 5 is one of the more controversial positive electrode materials for aluminium batteries, with researchers arguing both for and against the possible storage of Al(III). In this review, it is argued that there is some evidence for Al(III) intercalation in V 2 O 5 , but a low average discharge potential of less than 0.6 V vs. Al|Al(III) limits its use as a battery system. The cell reaction is as shown in Equation (13) [115]:

V 2 O
The reversible intercalation of Al(III), Mg(II) and Zn(II) into V 2 O 5 was shown in a 1998 study on V 2 O 5 aerogels. Initial intercalation was carried out by immersing aerogel electrodes for one week into dibutylmagnesium, dimethylzinc or trimethylaluminium for Mg 2+ , Zn 2+ or Al 3+ respectively, all in heptane solution. All demonstrated reversible electroactive behaviour [155].
In 2011, in one of the earliest studies on V 2 O 5 in an aluminium battery [156], Jayaprakash et al. found that using V 2 O 5 nanowires and a stainless steel current collector allowed for a battery system that demonstrated a sharp drop in cell discharge potential from 1.8 V vs. Al| Al(III) open-circuit potential to a plateau of about 0.55 V vs. Al|Al(III). The capacity of the positive electrode was around 273 mAh g − 1 after 20 cycles, which is a good capacity, although the poor discharge potential means that the overall specific energy was not high. This paper used an r = 1.1 AlCl 3 -[EMIm]Cl electrolyte [156]. Reed et al. [149] attempted in 2013 to replicate the battery system and found that the low discharge potential (0.6 V vs. Al|Al(III)) achieved with V 2 O 5 in a Swagelok cell was attributable to reactions involving the iron and chromium in the stainless steel current collector, destroying the cell via dendrite growth after just 20 cycles. This was supported by the fact that V 2 O 5 demonstrated no electrochemical activity in a battery when combined with an unreactive platinum current collector.
More recently in 2015, Wang et al. [115] demonstrated that V 2 O 5 deposited on a Ni foam current collector is electroactive, while Ni foam on its own is not. Although the cells were charged to 2.5 V vs. Al|Al(III), the potential dropped rapidly during discharge limiting the average discharge potential, as for Jayaprakash et al., to 0.6 V vs. Al|Al(III). This perhaps indicates that Al(III) or chloroaluminate species are forming stable compounds at a potential of only 0.6 V vs. Al/Al(III). The cell had an initial specific discharge capacity based on positive electrode mass of 239 mAh g − 1 which reduces by around 20% over the first five cycles. The electrolyte was r = 1.1 AlCl 3 -[BMIM]Cl, the minimum value at which Al 2 Cl 7 − is present for aluminium deposition. Though it was shown that the V was oxidised from V 4+ to V 5+ during charging, this could account for chloroaluminate species intercalation during charging, and no conclusive evidence of Al(III) intercalation is presented. Chiku et al. [157] found that using a V 2 O 5 positive electrode with an ionic liquid electrolyte consisting of AlCl 3 , dipropylsulfone and toluene in a 1:10:5 ratio, gave a 1st cycle discharge capacity of 150 mAh g − 1 V 2 O 5 , followed by 100 mAh g − 1 on the 15th cycle with good coulombic efficiency. A closer analysis found that before cycling, only V 5+ was present, but this was reduced to V 3+ and V 4+ during discharging. A molybdenum current collector was used, having been shown to be unreactive in the relevant discharge potential range in the electrolyte. Clearly, V 2 O 5 is electroactive, and the capacity obtained is approximately matched to the theoretical calculation. The authors concluded that V 2 O 5 is electroactive and might able to store Al(III), but in a battery system, it provides a steeply sloping and low average discharge potential profile, limiting its use based on current progress.

VO 2 .
In 2013, Jiang et al. [41] investigated a VO 2 material with a stainless steel current collector for the positive electrode. They use computational modelling to demonstrate the potential intercalation behaviour of aluminium in the VO 2 matrix via the following reaction (x is the experimentally found value) shown in Equation (14): They found that all formation energies were negative, indicating that intercalation was possible and that intercalating a small number of aluminium atoms was well tolerated by the matrix, but above that, the bonds started to bend and break. They concluded that aluminium could not replace V in the matrix due to a positive energy of formation. The positive electrode maintains a capacity of over 100 mAh g − 1 in testing based on the mass of the positive electrode alone, but the discharge potential is low at around 0.55 V vs. Al|Al(III). The electrolyte was r = 1 AlCl 3 -[BMIM]Cl and the authors report a 48 Wh kg − 1 specific energy [41]. However, as previously discussed the use of stainless steel as a current collector means the results must be treated with caution [149].
In summary, all attempts to use vanadium-containing positive electrodes have resulted in low discharge potentials of less than 0.6 V, resulting in specific energy values that fall behind aluminium-graphite (48 Wh kg − 1 vs. 65 Wh kg − 1 ) [41,43,114].

Transition metal sulphides
Conversion electrodes use chemical reactions between the positive electrode and the intercalating species to form entirely new products stored together at the positive electrode [158]. Conversion electrodes based on fluorides, oxides and sulphides have been explored for aluminium batteries, but fluorides and oxides form very strong ionic bond with aluminium, causing high binding energies. Sulphides, which form weaker ionic bonds with aluminium, remain as positive electrode materials of interest [39].

TiS 2 .
In 2017, Geng et al. [40] investigated a transition metal sulphide, TiS 2 , which has a more polarizable anionic structure than transition metal oxides like V 2 O 5 , and therefore should have reduced barriers to the transport of multivalent ions. Storage of aluminium was investigated from an r = 1.5 AlCl 3 -1-butyl-3-methylimidazolium chloride electrolyte. Reversibility was demonstrated for both layered TiS 2 and spinel-based cubic Cu 0.31 Ti 2 S 4 , and it was shown that Al(III) resided in the octahedral sites in the layered TiS 2 . While Al(III) intercalation was demonstrated, its passage through the crystal structure was slow, and the binding energy of aluminium in TiS 2 was still high enough to cause low discharge potential. Experimentally, for the better performing layered TiS 2 , a highly sloping discharge potential with an average value of less than 1 V vs. Al|Al(III) was found with a capacity of about 70 mA h g − 1 (based on positive electrode mass) after several cycles, indicating that around 11% of the available octahedral intercalation sites were taken by Al(III). The positive electrode initially demonstrated poor discharge capacity and coulombic efficiency, and both improved on subsequent cycles to a stable value. However, while specific energy is not calculated, both the specific capacity of the TiS 2 electrode and its average discharge potential fall far behind Al-graphite (70 mAh g − 1 vs. 142 mAh g − 1 and 1 V vs. 1.77 V respectively [114].

Ni 3 S 2 on graphene microflakes.
In 2016, Yu et al. [117] investigated Ni 3 S 2 on graphene microflakes in an AlCl 3 -[EMIm]Cl electrolyte. With a discharge potential of around 1 V vs. Al|Al(III), the discharge capacity remained over 60 mAh g − 1 (based on positive electrode mass) after 100 cycles, with an initial charge/discharge capacity of 300 and 235 mAh g − 1 respectively. These results are similar to those reported for TiS 2 , and the same comparisons to aluminium-graphite apply.

FeS 2 .
Also in 2016, Mori et al. [39] investigated the theoretical mechanism for Al(III) intercalation in FeS 2 -a transition metal sulphide that has been used successfully in lithium and sodium battery positive electrodes. They used an r = 2 AlCl 3 -[EMIm]Cl electrolyte. FeS 2 has the advantage of being naturally abundant. For lithium-ions, the FeS 2 goes through an intermediate, layered phase of Li 2 FeS 2 before reaching its fully charged configuration. For aluminium however, the high charge density makes this intermediate phase unstable, and so diffusion into the bulk is required. The cell reaction is shown in Equation (15): FeS 2 becomes FeS and Al 2 S 3 -a conversion reaction. In practise, sulphide dissolution occurs, causing poor cycling stability, and low discharge potential results from the difficulty of diffusing Al(III) through bulk FeS 2 . Nevertheless, the theoretical mechanism for intercalation was successfully demonstrated via XRD and XAS measurements. Future work includes combining with a nanostructure carbon substrate to provide diffusion paths and increased surface area.

Anthraquinone
The aluminium-anthraquinone (AQ) battery uses an anthraquinonebased poly(antraquinoyl sulfide) (PAQS) positive electrode to store charge and a r = 1.5 AlCl 3 -[EMIm]Cl electrolyte [159]. PAQS has previously been used for lithium batteries [160], where a discharge potential of 2.2 V vs. Al|Al(III) has been achieved. For the AQ battery [159], XPS results were used to calculate that the major Al containing species must be AlCl 2+ . However, as previously discussed, this species does not appear in the electrolyte. Instead, AlCl 4 − and Al 2 Cl 7 − are present in r = 1.5 AlCl 3 -[EMIm]Cl. Therefore, questions remain over the operation of the battery, which cast doubt over the quoted energy density of 101 Wh kg − 1 ; it is most likely that the Al-containing species are actually being stored. More than 60% of the initial capacity remains after 500 cycles, but this is not competitive with Al-graphite or Al-PEDOT batteries.

Current collectors
The use of acidic ionic liquid electrolyte in aluminium batteries means that care must be taken when selecting current collectors. Materials that are completely inactive in these systems include molybdenum, tungsten and platinum, but clearly these are not abundant or affordable [42]. Other commonly used equipment also cannot withstand these electrolytes [113]; states an experimental failure due to Celgard Swagelok separator corrosion. Even some binders are activealthough PTFE is stable, PVDF reacts [115]. Carbon-coated aluminium foil [144] and carbon have been used successfully as anodic current collectors [161]. If the aluminium purity is extremely high (so as not to contaminate the electrolyte), it can act as its own current collector at the negative electrode, although it might seem prudent to use an unreactive current collector to minimise the risk of holes forming because of uneven dissolution of the electrode during discharge.

Summary and future perspective
Due to the particularly difficult challenge of finding positive electrodes that have good discharge potential and cycle life, the research focus is likely to remain towards those materials that store chloroaluminate species. No electrolytes have yet been shown to have a wider electrochemical window or better stability than AlCl 3 -[EMIm]Cl. Forming a gel of this electrolyte is a very interesting research avenue, as it is the only obvious route for improving the electrochemical window of RTILs.
Deep eutectic solvents are of interest because some of them are cheap and particularly non-toxic. However, they offer a limited electrochemical stability window and composition range suitable for aluminium deposition, and because specific energy is such a key parameter for new energy storage, it seems unlikely that their cost and toxicity credentials can make up for their lack of performance. Entirely new DES with improved performance need to be discovered for them to gain relevance in non-aqueous aluminium batteries.
Graphite-based positive electrodes offer the best overall performance due to the stability and high average discharge potential. Cheap and relatively abundant graphite flakes seem to offer the best balance of performance and cost and achieve near the theoretical maximum in terms of specific energy and thousands of cycle lives [114]. Conducting polymers offer similar operation but average discharge potential and cycle life are lower than for graphite [13,46]. Elaborate structures such as carbon nanotubes may elicit marginal gains in the specific capacity of graphite electrodes, but these are unlikely to be economically justified considering the cost of developing carbon nanostructures compared to assembling graphite flakes. There may also be ways to improve the capacity of conducting polymer electrodes using refined electrosynthesis techniques, but again these improvements are likely to be marginal. Other positive electrode materials that store chloroaluminate species fall far behind graphite and conducting polymers. Positive electrode materials that store Al(III) offer much better theoretical performance but there is little evidence of this theoretical performance being achieved over a long cycle life; aluminium-sulphur probably appears most promising but still only maintains 60% of its capacity over 50 cycles [154].
Considering the highest achieved specific energy of 68 Wh kg − 1 , and accounting for the extra mass involved in commercial packaging, current aluminium batteries may have sufficient specific energy to meet grid storage requirements. However, the cost of ionic liquids is a barrier to entering the market. Improved performance, cheaper electrolyte, or both, would help to improve prospects. If aluminium batteries could reach higher specific energies, perhaps greater than 200 Wh kg − 1 , they might begin to compete in the electric vehicle market. They would still not compete with lithium ion batteries on specific energy, but might offer better charging time, affordability, environmental credentials and cycle life. Then, the technology might become optimal for applications demanding predictable length journeys with regular charging opportunities. Combined with an electric drive this could fulfil a huge number of mass transit applications globally, most of which are currently serviced by diesel engines. With these potential applications in mind, aluminium battery technology should be relentlessly pursued towards its theoretical performance limits, to ensure its full potential is realised as soon as possible in the quest to develop a sustainable future.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.