Thin film structure, electronic components and method of fabrication

Abstract

A method of fabricating a thin film structure comprises the steps of: providing a layer ofmetal nitride MN; in an atmosphere with a first oxygen partial pressure, co-depositing HfO2 and a perovskite with formula ABO3 on the metal nitride layer to form a first layer of multi-component metal oxide with formula Hf(A, B)OZ, wherein 1.7 < z < 2; and, in an atmosphere with a second oxygen partial pressure different to the first oxygen partial pressure, co-depositing HfO2 and the perovskite on the first layer of multi-component metal oxide. The resulting thin film structure comprises: a layer of metal nitride MN; a layer of metal oxynitride MOxNy on the layer of metal nitride, the layer of metal oxynitride having a thickness of at least 10 nm; and a layer of multi-component metal oxide with formula Hf(A,B)O2 on the metal oxynitride layer. The oxygen content x of the metal oxynitride MOxNy layer increases non-uniformly across the thickness of the metal oxynitride layer and is highest at a boundary between the metal oxynitride layer and the oxide layer, and the nitrogen content y decreases non-uniformly across the thickness of the metal oxynitride layer and is lowest at the boundary between the metal oxynitride layer and the oxide layer. Electronic components exhibiting memristive properties are also provided.

Description

[0001] Thin Film Structure, Electronic Components and Method of Fabrication

[0002] The invention relates to a thin film structure, a method of fabricating the thin film structure, and to electronic components comprising the thin film structure. In particular, the invention relates to a switching oxide thin film structure which demonstrates memristive properties suitable for electronic components such as memristors and neuromorphic memory devices.

[0003] Background of the Invention

[0004] Worldwide demand for computer processing power is growing rapidly, and this demand is accelerating due to the ever-increasing adoption of artificial intelligence (Al). This is driving a need for more compact and efficient computer hardware.

[0005] Current Al hardware relies mainly on transistor-based integrated chips, and a much larger number of smaller transistors are needed as Al grows. However, these 3-terminal semiconductor-based devices have already reached their quantum-mechanical and fabrication limitations. In addition, training Al models also creates a need for data centres with sufficiently high capacities to store enormous quantities of digital information, which cannot be achieved by common memories such as widely-used energy-intensive dynamic random-access memories. In this context, neuromorphic compute-in-memory architectures, in which the computing unit is designed within the memory block, are at the core of attention as the next generation of modern, sustainable and high-precision Al hardware.

[0006] The human brain efficiently processes and stores complex signals in a dense network of neurons and synapses. Storing information takes place in the form of synaptic weights, or synaptic plasticity, which is the underlying mechanism for knowledge-based learning. Neuromorphic electronic devices emulate the brain’s multilevel (analogue) pre- and postacquisitions through a network of small, programmable processing units that behave like the neurons. These bio-inspired technologies eliminate the energy-intensive, time-consuming data movement between the computing and storing units, resulting in saving > 70% of current computing power consumption, and can be trained to learn and effectively solve problems.

[0007] In contrast to transistors, 2-terminal insulating(oxide)-based non-volatile memory devices, with simpler capacitor-like structures and fundamentally different materials and device physics than transistors, are serious contenders for neuromorphic Al hardware. These devices, which inherently retain stored information without requiring external power, show key synaptic features: low programming energy, minimal standby leakage current, highswitching speed, good retention and endurance, multilevel synaptic weights, and timingdependent plasticity. Among various emerging 2-terminal non-volatile memories, resistive random-access memory devices (also referred to as memristive devices or memristors) are gaining considerable attention for neuromorphic technologies.

[0008] However, despite their great promises for a broad range of high-throughput memory and neuromorphic applications, memristive devices still have major challenges. In particular, those based on resistively switchable binary oxides, like widely-used HfC>2, typically rely on filamentary resistive switching and usually need initial electroforming processes and current compliances to avoid device hard breakdown. Such devices suffer from device-to-device and cycle-to-cycle variations, which are mainly due to filaments’ stochastic nature. This critical issue significantly targets the computational accuracy and device performance. The other major problems are their restricted number of distinguishable conductance levels and difficulties to control them. In filamentary devices, the conductance levels are directly tuned by varying the current compliance, which is inherently limited, costs further energy, and needs extra special devices (like selectors) that adds more design complexities. Further, these devices usually require high electroforming voltages to create the conductive filaments, and the low resistance states (ON states) typically occur at high currents, leading to high power usage, which is unfavourable for green neuromorphic hardware.

[0009] of the Invention

[0010] The invention provides a method of fabricating a thin film structure, a method of fabricating an electronic component, a thin film structure, and electronic components comprising the thin film structure, as defined in the appended independent claims, to which reference should now be made. Preferred or advantageous features of the invention are set out in dependent subclaims.

[0011] Using an innovative materials design and fabrication method, the present disclosure provides a thin film structure which addresses the memristor challenges mentioned above, enabling the development of advanced energy-efficient neuromorphic Al hardware. The present application relates to Hf(A,B)C>2 as a new class of multi-component HfC>2-based switching systems, and to a new two-step, oxide-thin-film growth method that creates thin-film electronic devices with excellent ultra-low-current, non-volatile, uniform, interfacial resistive-switching and neuromorphic functionalities.Method of a Thin Film Structure

[0012] According to a first aspect of the disclosure, there is provided a method of fabricating a thin film structure comprising the steps of:

[0013] providing a layer of metal nitride MN;

[0014] in an atmosphere with a first oxygen partial pressure, co-depositing HfC>2 and a perovskite with formula ABO3 on the metal nitride layer to form a first layer of multi-component metal oxide with formula Hf(A, B)OZ, wherein 1.7 < z < 2; and,

[0015] in an atmosphere with a second oxygen partial pressure different to the first oxygen partial pressure, co-depositing HfC>2 and the perovskite on the first layer of multi-component metal oxide.

[0016] The method may be a method of fabricating an oxide thin film structure or a switching oxide thin film structure.

[0017] The method preferably comprises two discrete deposition steps:

[0018] - co-depositing HfC>2 and the perovskite on the metal nitride layer from a HfC>2 target and a perovskite target, in an atmosphere with a first oxygen partial pressure; and then

[0019] - co-depositing HfC>2 and the perovskite on the first layer from a HfC>2 target and a perovskite target, in an atmosphere with a second oxygen partial pressure different to the first oxygen partial pressure

[0020] By carrying out the co-deposition of the HfC>2 and perovskite in two discrete steps, under different oxygen concentrations, there is a difference in the initial oxygen content of the multi-component metal oxide Hf(A, B)OZdeposited by the two steps.

[0021] The step of co-depositing HfC>2 and the perovskite on the first layer may initially form a second layer of multi-component metal oxide over the first layer. The second layer of multicomponent metal oxide may have an oxygen concentration which is different from the oxygen concentration (z) of the first layer.

[0022] During the deposition and after the layers are formed, oxygen and nitrogen diffusion naturally occurs between the layers that are deposited. This means that although the method co-deposits HfC>2 and the perovskite in two separate steps, under different oxygen concentrations, oxygen migration occurs immediately after deposition. This may mean that that rather than forming two discrete layers of Hf(A, B)OZhaving identifiably differentoxygen concentrations, the multi-component metal oxide Hf(A, B)OZbecomes a single layer with a uniform oxygen concentration.

[0023] Preferably the first oxygen partial pressure is lower than the second oxygen partial pressure. The first deposition step may initially create a first layer of under-stoichiometric oxide Hf(A, B)OZin which z is less than 2. For example, the oxygen content z in the first layer may be 1.7 < z < 2, or 1.7 < z < 1.9 such that the first layer of multi-component metal oxide may be Hf(A, B)Oi.7-i.g. The second deposition step may preferably be carried out under a higher second oxygen partial pressure, so that the second deposition step preferably deposits multi-component metal oxide with the formula Hf(A, B)C>2.

[0024] Particularly preferably, during the second deposition step under a higher oxygen partial pressure, oxygen may diffuse into the under-stoichiometric first layer so that following both deposition steps, all of the multi-component metal oxide in the thin film merges to form a continuous layer of stoichiometric Hf(A, B)C>2.

[0025] Oxygen and nitrogen diffusion occur in opposite directions across the boundary between the multi-component metal oxide and the metal nitride layer, so that a layer of metal oxynitride MOxNyforms between the layer of metal nitride and the layer of multi-component metal oxide.

[0026] As a result of the oxygen and nitrogen diffusion, the oxygen content x of the metal oxynitride MOxNylayer increases non-uniformly across the thickness of the metal oxynitride layer and is highest at a boundary between the metal oxynitride layer and the multicomponent metal oxide layer, and the nitrogen content y decreases non-uniformly across the thickness of the metal oxynitride layer and is lowest at the boundary between the metal oxynitride layer and the multi-component metal oxide layer. These characteristic nitrogen and oxygen gradients can be achieved only by the two-step deposition of the multicomponent metal oxide material.

[0027] The different oxygen partial pressures applied during the two fabrication steps, and the resulting oxygen and nitrogen diffusion, creates an asymmetric space-charge region in the thin film structure. This asymmetric space-charge region is much wider inside the Hf(A,B)C>2 layer than the MOxNy, originating from the unequal defect and carrier concentrations in Hf(A,B)C>2 and MOxNy.The two-step deposition method, and the resulting oxygen and nitrogen diffusion, creates a p-n heterointerface that forms between p-type Hf(A,B)C>2 and n-type, oxygen-rich MOxNy. This results in a thick depletion region that is asymmetrically extended into the Hf(A,B)C>2 thin film.

[0028] As discussed further below, these electronic properties mean that the thin film structure formed by the present method exhibits excellent characteristics for electronic devices such as memristors.

[0029] The metal M of the metal nitride layer may comprise one or more of Ti, Ta, Zr, Mo, and W.

[0030] A variety of perovskite materials may be used by selecting a suitable perovskite target for the co-deposition step. The perovskite preferably has a formula ABO3, in which A and B are metals. The addition of a co-deposited perovskite ABO3 advantageously converts the codeposited HfC>2 into a p-type multicomponent metal oxide Hf(A, B)C>2.

[0031] The multicomponent metal oxide Hf(A, B)OZ, wherein 1.7 < z < 2, may optionally be termed multicomponent hafnium oxide, or perovskite-doped HfC>2.

[0032] In a particularly preferred embodiment, the perovskite is SrTiOs, such that A is Sr and B is Ti. When the perovskite is SrTiOs, the multi-component metal oxide formed by the codeposition is Hf(Sr,Ti)Oz. After both deposition steps are complete, and inter-layer oxygen diffusion has occurred, the multi-component metal oxide layer is Hf(Sr,Ti)C>2.

[0033] The first layer of multi-component metal oxide may preferably have a thickness of 10 nm to 20 nm, or 12.5 nm to 17.5 nm.

[0034] The step of co-depositing HfC>2 and the perovskite on the first layer may initially form a second layer of multi-component metal oxide over the first layer, before oxygen diffuses between the second layers to make them indistinguishable from one another. The second layer may preferably have a thickness of 0.5 nm to 5 nm, or 0.75 nm to 4 nm, or 1 nm to 2 nm. The second co-depositing step may thus deposit a second layer of multi-component metal oxide which is thinner than the first layer.

[0035] In preferred embodiments, the first oxygen partial pressure is lower than the second oxygen partial pressure. This means that the first layer of multi-component metal oxide, which is in contact with the metal nitride layer, may initially be non-stoichiometric with an atomic oxygen content z which is less than 2. This may advantageously create a p-typeregion of Hf(A, B)OZat an interface between the metal nitride layer and the multicomponent metal oxide, and encourage the formation of a thick layer of metal oxynitride MOxNyat this interface.

[0036] In a particularly preferred embodiment, the first oxygen partial pressure may be 0 Pa, such that the step of co-depositing HfC>2 and a perovskite on the metal nitride layer takes place in an atmosphere that is free of oxygen.

[0037] The steps of co-depositing HfC>2 and a perovskite preferably take place in a reaction chamber with a controllable atmosphere. The oxygen partial pressure in the reaction chamber may be controlled by controlling the oxygen flow rate through the reaction chamber.

[0038] SCCM is a unit which is widely used to quantify the flow rate of a fluid in the context of sputtering, semiconductor fabrication and thin film fabrication, where 1 SCCM is identical to 1 cm3sTp / min.

[0039] The first oxygen partial pressure may be achieved with an oxygen flow rate through the reaction chamber of less than 10 standard cubic centimetres per minute (seem), preferably 5 seem or less, particularly preferably 0 seem. The total deposition pressure in the reaction chamber may be between 10 - 30 mTorr.

[0040] The second oxygen partial pressure is preferably greater than 0 Pa. Thus the step of codepositing HfC>2 and the perovskite on the first layer takes place in an atmosphere which contains oxygen.

[0041] The second oxygen partial pressure may be achieved with an oxygen flow rate through the reaction chamber of at least 10 standard cubic centimetres per minute (seem), preferably at least 15 seem, particularly preferably at least 20 seem. The total deposition pressure in the reaction chamber may be between 10 - 30 mTorr.

[0042] The step of co-depositing HfC>2 and a perovskite on the first layer of multi-component metal oxide may be carried out in an atmosphere with a second oxygen concentration that is greater than 0 %. The second oxygen concentration may preferably be greater than 10 %, or greater than 20 %, or greater than 35 %.

[0043] In a preferred embodiment, for example, the second oxygen partial pressure may be achieved with an oxygen flow rate through the reaction chamber of 20 seem O2 in additionto 30 seem Ar. In this embodiment the step of co-depositing HfC>2 and a perovskite on the first layer of multi-component metal oxide is carried out in at atmosphere with a second oxygen concentration of 40% O2 and 60% Ar.

[0044] The steps of co-depositing HfC>2 and perovskite ABO3 are preferably carried out by sputterdepositing using HfC>2 and perovskite targets. In a preferred embodiment, radio frequency magnetron sputtering may be used. The inventor has found that co-sputtering from a HfC>2 target and a perovskite target advantageously forms a multi-component metal oxide with formula Hf(A, B)OZ, wherein 1.7 < z < 2. This multi-component metal oxide has particularly advantageous structural and electronic characteristics which are discussed further below.

[0045] Other than the oxygen partial pressure, which differs between the two deposition steps, the first layer and second layer are preferably deposited under the same conditions. For example the same co-sputtering parameters are preferably used, so that the only difference between these two steps is the concentration of oxygen in the atmosphere in which the deposition is carried out. The first and second co-deposition steps may be carried out for different durations, so that the first layer and the second layer of multicomponent metal oxide have different thicknesses.

[0046] A particular benefit of the present method is that the metal nitride layer may be provided on any substrate suitable for electronic device overgrowth. Unlike many prior art approaches to memristor fabrication, the present method is substrate-independent, and is not limited to requiring a particular substrate type.

[0047] Thin Film Structure

[0048] According to a second aspect of the present disclosure, there is provided a thin film structure comprising:

[0049] a layer of metal nitride MN;

[0050] a layer of metal oxynitride with formula MOxNyon the metal nitride layer, the layer of metal oxynitride having a thickness of at least 10 nm; and

[0051] a layer of multi-component metal oxide with formula Hf(A,B)C>2on the metal oxynitride layer. The oxygen content x of the metal oxynitride MOxNylayer increases non-uniformly across the thickness of the metal oxynitride layer and is highest at a boundary between the metal oxynitride layer and the oxide layer. The nitrogen content y decreases non-uniformly across the thickness of the metal oxynitride layer and is lowest at the boundary between the metal oxynitride layer and the oxide layer.The thin film structure may be termed an oxide thin film structure, or a switching oxide thin film structure.

[0052] The thin film structure of the second aspect is preferably a thin film structure formed by the method of the first aspect, as described above. As described above, a multi-component metal oxide Hf(A,B)C>2is formed by two co-deposition steps onto a metal nitride layer, and oxygen and nitrogen diffusion which takes place between the layers of the thin film structure creates a distinct metal oxynitride MOxNylayer between the metal nitride and metal oxide layers. Distinctive oxygen and nitrogen concentration gradients are formed in this layer, and in an interface region of the Hf(A,B)C>2 adjacent to the metal oxynitride layer, as a result of the different oxygen partial pressures used during the two deposition steps.

[0053] The thin film structure preferably comprises a layer of metal nitride with formula MN, and the layer of metal oxynitride with formula MOxNymay be on the metal nitride layer.

[0054] The layer of metal oxynitride with formula MOxNymay be an n-type layer of metal oxynitride, and the layer of multi-component metal oxide with formula Hf(A,B)C>2may be a p-type layer of Hf(A,B)C>2, such that the thin film structure comprises a p-n heterointerface between the p-type Hf(A,B)C>2 and the n-type MOxNy.

[0055] The thin film structure may contain a depletion region which extends outwards from the p-n heterointerface into the p-type Hf(A,B)C>2 layer and the n-type MOxNylayer. Thus the depletion region may consist of: an oxide depletion region, which is the portion of the depletion region that extends into the oxide layer; and an oxynitride depletion region, which is the portion of the depletion region which extends into the oxynitride layer. In equilibrium conditions when no electrical bias is applied to the thin film structure, the oxide depletion region may be thicker than the oxynitride depletion region. The depletion region may thus extend asymmetrically around the p-n hetero interface, so that it extends further into the p-type Hf(A,B)C>2 layer than it extends into the n-type MOxNylayer. The reason for this may be that the oxynitride MOxNylayer has a higher defect concentration and a significantly higher carrier concentration than the oxide layer of stochiometric Hf(A,B)C>2. A large built-in potential that is generated at the n-MOxNy / p-Hf(A,B)C>2 heterointerface may therefore form mainly inside the Hf(A,B)C>2 film, creating a depletion region which extends asymmetrically into this layer.

[0056] As discussed below in relation to the Figures, thin film structures based on this novel structure of multi-component Hf(A,B)C>2 switching oxide films demonstrate memristivebehaviour that is highly desirable for energy-efficient, resistive-switching-based synaptic Al hardware. Interfacial non-volatile memristive devices formed from these thin films have been found to exhibit exceptional cycle-to-cycle and device-to-device uniformities with ultra-low switching currents (< 10'8A), and 6000 distinct conductance levels (ranging from ~2.5 x io-9to ~1.4 x io-7S) achieved by implementing identical 1.5-V presynaptic spikes.

[0057] The term “thin film” is widely used and well understood in the art, and generally accepted to relate to structures with thicknesses of up to a few microns (pm). The thin film structures formed in the present disclosure may preferably have thicknesses of up to 1 pm, or up to 2 pm, or up to 3 pm. The thin film structure may alternatively be termed a multilayer structure having a thickness of up to 1 pm, or up to 2 pm, or up to 3 pm.

[0058] As discussed in relation to the first aspect, A and B may be components of a perovskite material, preferably in which ABO3 is a perovskite. A and B are preferably metals. In a particularly preferred embodiment, A is Sr and B is Ti.

[0059] In a particularly preferred embodiment, the multi-component metal oxide is Hf(Sr,Ti)C>2. As described above in relation to the first aspect, this can be formed by co-depositing HfC>2 and the perovskite SrTiCh.

[0060] The layer of multi-component metal oxide Hf(A,B)C>2may preferably have a thickness of 10.5 nm to 25 nm, or 12.5 nm to 20 nm, or 14 nm to 18 nm.

[0061] The thin film structure may comprise a substrate. Any substrate suitable for thin film deposition may be used, for example an amorphous silicon substrate.

[0062] In a preferred embodiment, the thin film structure comprises a substrate, a layer of metal nitride MN on the substrate, the layer of metal oxynitride MOxNyon the metal nitride layer, and the layer of multi-component metal oxide with formula Hf(A,B)C>2on the metal oxynitride layer.

[0063] The thin film structure preferably does not contain a distinct layer of TiC>2 between the metal nitride layer and the multi-component metal oxide layer.

[0064] As discussed further below, the compositions of the metal nitride and metal oxynitride layers change on application of electrical bias across the thin film structure, as the electrically-driven diffusion of O and N anions through the structure alters the layercomposition. All definitions of layer thicknesses herein therefore relate to the structure under equilibrium conditions unless otherwise specified.

[0065] The layer of metal oxynitride preferably has a thickness of at least 15 nm, or at least 25 nm, or at least 50 nm. In some preferred embodiments, the layer of metal oxynitride has a thickness of between 10 nm and 100 nm, or between 20 nm and 80 nm.

[0066] The metal M may be Ti, or Ta, Zr, Mo, or W, such that the metal oxynitride may be TiOxNy, or TaOxNy, or ZrOxNy, or MoOxNy, or WOxNy.

[0067] In embodiments containing a metal nitride MN layer against the metal oxynitride layer, the metal nitride MN may be TiN, or TaN, ZrN, MoN, or WN. The metal nitride layer may have a columnar microstructure.

[0068] In some preferred embodiments, both the metal oxynitride and the multi-component metal oxide have a columnar microstructure. This columnar microstructure may be inherited from the crystal structure of the metal nitride MN layer on which the thin film structure is grown. The multi-component metal oxide layer may be described as a multi-component nanocomposite, and / or as a self-organised columnar shell-core nanostructure.

[0069] The metal oxynitride layer may comprise, or consist of, a plurality of equiaxed crystallites.

[0070] The multi-component metal oxide layer may have a nanocolumnar microstructure, preferably in which nanocolumns in the microstructure have diameters of between 5 nm and 10 nm. The multi-component metal oxide layer may have column boundaries with elevated concentrations of A and / or B, such that the concentrations of A and / or B at the column boundaries is higher than the average concentration of A and / or B inside the columns.

[0071] The multi-component metal oxide layer may be crystalline. The multi-component metal oxide layer may have a polar orthorhombic or rhombohedral crystal structure. This crystal structure may advantageously be substrate-independent, so that the multi-component metal oxide layer forms in this crystal structure regardless of the substrate on which the thin film structure is formed.

[0072] As the multi-component metal oxide layer is formed in two stages, the multi-component metal oxide layer may at least initially comprise a first oxide layer portion adjacent the metal oxynitride layer, and a second oxide layer portion separated from the metal oxynitridelayer by the first oxide layer portion. The first oxide layer portion is preferably formed in an oxygen-depleted atmosphere, and may have formula Hf(A, B)OZ, wherein 1.7 < z < 2. Prior to formation of the second oxide layer portion, the first oxide layer portion may have an O / Hf ratio z of less than 1.9, preferably 1.8 or less. The second oxide layer portion is preferably formed in an oxygen-rich atmosphere, and may have formula Hf(A, B)OZ, wherein 1.7 < z < 2. The second oxide layer portion preferably has an O / Hf ratio of greater than 1.95, particularly preferably in which the O / Hf ratio is 2.

[0073] Following inter-layer oxygen diffusion, the first oxide layer portion and the second oxide layer portion typically equilibrate to form a uniform layer of multi-component metal oxide with formula Hf(A,B)O2.

[0074] At the boundary between the metal oxynitride layer MOxNyand the multi-component metal oxide layer, the nitrogen content y of the MOxNymay be less than 0.1, preferably less than 0.05.

[0075] At a boundary between the metal oxynitride MOxNylayer and the metal nitride layer MN, the nitrogen content y of the MOxNymay be at least 0.4, preferably at least 0.45.

[0076] At the boundary between the metal oxynitride MOxNylayer and the oxide layer, the oxygen content x may be at least 0.5, preferably at least 0.55 or at least 0.6.

[0077] At the boundary between the metal oxynitride MOxNylayer and the metal nitride layer, the oxygen content x may be less than 0.15, preferably x may be less than 0.1, particularly preferably x may be less than 0.08.

[0078] Method of

[0079]

[0080] an Electronic

[0081]

[0082] According to a third aspect of the present disclosure, there is provided a method of manufacturing an electronic component.

[0083] The method of manufacturing an electronic component may comprise the step of providing the thin film structure of the second aspect of the disclosure, and forming a conductive electrode on the layer of multi-component metal oxide with formula Hf(A,B)C>2.

[0084] Alternatively, the method of manufacturing an electronic component may comprise a first step of fabricating a thin film structure according to the method of the first aspect of the disclosure. The method of the first aspect may result in a thin film structure having a layerof multi-component metal oxide with formula Hf(A,B)C>2. The method may then comprise the additional step of forming a conductive electrode on a surface of the layer of multicomponent metal oxide with formula Hf(A,B)C>2.

[0085] The conductive electrode may be formed by conventional techniques that are known to the skilled person in the art of electronic device fabrication. For example a metal electrode such as a Mo electrode may be deposited on the top surface of the thin film structure, so that the electrode is electrically connected to the multi-component metal oxide layer.

[0086] Depending on the component being made, further component processing steps may be carried out in order to form the standard parts of the component. Such processes are well-known in the art.

[0087] Electronic

[0088]

[0089] The thin film structure according to the second aspect of the disclosure exhibits electrical characteristics which are sought-after for a variety of electronic components. According to further aspects of the present disclosure, there are provided electronic components and electronic devices comprising the thin film structure described above.

[0090] According to a fourth aspect of the present disclosure, there is provided an electronic component comprising the oxide thin film structure as defined above, and an electrode on the oxide layer. The electrode may be a molybdenum (Mo) electrode, or the electrode may be formed from another suitable conductive material.

[0091] According to a fifth aspect of the present disclosure, there is provided a non-volatile memory device comprising the thin film structure according to the second aspect. The device may be a 2-terminal insulating(oxide)-based non-volatile memory device comprising the thin film structure according to the second aspect.

[0092] According to a sixth aspect of the present disclosure, there is provided a memristor comprising the thin film structure according to the second aspect. The memristor may comprise a top electrode connected to the multi-component metal oxide layer, and a bottom electrode electrically-connected to the other side of the boundary between the metal oxynitride layer and the multi-component metal oxide layer. As discussed above, the thin film structure preferably contains a metal nitride MN layer below the metal oxynitride layer - the metal nitride layer may act as the bottom electrode. Alternatively, a lower portion of the metal oxynitride layer may act as the bottom electrode, for example inembodiments where oxygen is diffused throughout the structure so that there is no metal nitride layer distinguishable from the metal oxynitride layer.

[0093] The memristor may have a typical "top electrode / resistive switching layer / bottom electrode" sandwich structure, in which the thin film structure of the second aspect is arranged as the resistive switching layer located between the two terminal electrodes.

[0094] When an electrical bias is applied between the top and bottom electrodes of the memristor, a resistance shear process occurs in which the material of the thin film structure (the resistive switching layer) undergoes physical changes under the action of the electric field, which ultimately leads to a change in resistance. The characteristics of the thin film structure determine the electrical characteristics, stability and application performance of the memristor, as described in detail below.

[0095] According to a seventh aspect of the present disclosure, there is provided a resistive random-access memory device comprising the thin film structure according to the second aspect.

[0096] According to an eighth aspect of the present disclosure, there is provided a neuromorphic memory device comprising the thin film structure according to the second aspect of the disclosure.

[0097] Features described above in relation to any one aspect of the invention are equally applicable to every other aspect of the invention.

[0098] Brief Description of the Drawings

[0099] The invention will now be described, by way of example only, by reference to the following figures, in which:

[0100] Figure 1 is a schematic illustration of a thin film structure according to the present disclosure;

[0101] Figures 2(a) and 2(b) illustrate RBS experimental spectra and SIMNRA simulations of Hf(Sr,Ti)Ozthin films grown on Si / a-SiO2 / TiN according to an aspect of the present disclosure;

[0102] Figures 3(a)-3(d) set out transmission electron microscopy (TEM) results obtained on thin films according to an aspect of the present disclosure;Figure 4 shows XRD 0-20 scans of as-deposited (a) TiN, (b) TiN / Hf(Sr,Ti)0i.s, and (c) TiN / Hf(Sr,Ti)C>2 grown on Si(001) substrates;

[0103] Figure 5 sets out experimental results demonstrating the memristive characteristics of electronic devices according to an aspect of the present disclosure;

[0104] Figure 6 sets out experimental results demonstrating the neuromorphic characteristics of electronic devices according to an aspect of the present disclosure;

[0105] Figure 7 sets out additional experimental results demonstrating the neuromorphic characteristics of electronic devices according to an aspect of the present disclosure;

[0106] Figure 8 illustrates electrical characteristics of thin film electrical devices according to an aspect of the disclosure, and the conductance modulation mechanism model;

[0107] Figure 9 shows Hall effect voltage as a function of magnetic field for an electronic device according to the present disclosure;

[0108] Figure 10 illustrates (a) UV-vis transmittance spectra, and (b) Tauc plots of reference HfC>2 and Hf(Sr,Ti)C>2 layers fabricated according to the present disclosure; and

[0109] Figure 11 shows isolated energy band diagrams of component layers of thin film structures according to the present disclosure.

[0110] Detailed Description

[0111] As described below in relation to the Figures, the present disclosure provides new multicomponent Hf(A,B)C>2 thin films which exhibit excellent performance as switching oxides for resistive-switching-based synaptic Al hardware. The present disclosure provides a novel two-step, thin-film co-deposition strategy that results in the formation of nanocolumnar, stoichiometric Hf(A,B)C>2 on a metal nitride MN, with M-rich column boundaries and an asymmetrically-defective thick MOxNylayer formed between the metal nitride and the Hf(A,B)C>2 due to the gradual oxidation of the MN bottom electrode.

[0112] Figure 1 is a schematic illustration of a thin film structure 100 according to the present disclosure, formed by the two-step fabrication method described below. The thin film structure 100 consists of a layer of metal nitride MN on a substrate 200, a layer of metal oxynitride MOxNyon the metal nitride layer, and a layer of multi-component metal oxideHf(A,B)C>2 on the metal oxynitride layer. The preferred embodiment described and illustrated below has layers of TiN, TiOxNyand Hf(Sr,Ti)C>2 formed over a SiC>2 substrate.

[0113] The thin-film material engineering and growth design provided by the present disclosure advantageously eliminates the need for any initial electroforming.

[0114] The proposed thin-film material and design lead to analogue, interfacial non-volatile memristive devices that exhibit exceptional cycle-to-cycle and device-to-device uniformities with ultra-low switching currents (< 10'8A). These memristors mimic various key synaptic functionalities. They disclose up to 6000 distinct conductance levels, ranging from ~2.5 * 10'9to ~1.4 x io-7S, by implementing identical 1.5-V presynaptic spikes. Proper linearity and symmetry in synaptic-weight modulation, with outstanding operational stability and replicability, are achieved by optimising training schemes. Short-term synaptic plasticity, consisting of rapid and slow phases, is also demonstrated. Furthermore, these synaptic electronic devices reproducibly satisfy unsupervised learning rules, essential for timedependent Al functionalities, through spike-timing-dependent plasticity. The conductance modulation relies on tuning their interfacial p-n junction energy-barrier heights through controlled N- and O-ionic migration induced by built-in electric field. This material and growth design address the energy-consumption and non-reliability challenges of existing memristors and pave the way for ultra-low-power, brain-inspired computing technologies, which both are essential for energy-efficient Al hardware.

[0115] In the following description, a preferred embodiment is described in which SrTiCh is selected as perovskite ABO3, and TiN is selected as the metal nitride MN. SrTiCh is co-deposited with HfC>2 onto a TiN layer in two deposition steps with different oxygen atmospheres, to form a thin film structure in which layers of TiN and Hf(Sr,Ti)C>2 are separated by an intermediate layer of TiOxNy.

[0116]

[0117] Thin film growth

[0118] Multi-component Hf(Sr,Ti)C>2 thin films were sputter-deposited on Si / a-SiO2 / TiN in a reaction chamber, following a two-step growth approach: ~15-nm-thick layers were firstly co-sputtered from HfC>2 and SrTiCh targets at powers of PHf02= 50 W and PST0= 35 W, respectively, in a non-reactive atmosphere (P02= 0 seem), and then oxygen with flow ratePo2= 20 seem was immediately added to the chamber for 200 s, resulting in ~1-nm-thick layers, while keeping other deposition parameters constant.

[0119] TiN and Hf(Sr,Ti)C>2 thin films were sputter-deposited in a chamber equipped with 5.08-cm (2-inch) Ti, HfC>2, and SrTiCh targets on commercial Si(001) with a 200-pm-thick amorphous SiC>2 top layer as substrates. The Si / a-SiC>2 substrates were mounted on a substrate holder at the top of the growth chamber that was rotated at a speed of 20 rpm during the thin-film growth to provide uniform compositions and thicknesses. The system base pressure was < 5.0 x 10'8Torr (0.67 x 10'5Pa). Prior to each deposition, targets were pre-sputtered in a pure Ar atmosphere under closed shutters for 1 hour. Thereafter, the growth chamber was baked for 12 hours and then coated with pure Ti for 1 hour.

[0120] In the beginning, TiN was deposited on the Si / a-SiC>2 substrates using a pure Ti target by ion-assisted DC magnetron sputtering in a reactive nitride atmosphere. TiN layers have an electrical resistivity of -3x1 O'7Q.m, a N / Ti ratio of -1.05, and density and surface roughness of -5.6 g / cm3and -0.3 nm, respectively. Thereafter, the multi-component oxide thin films were grown on Si / a-SiO2 / TiN in the same chamber at 700 °C by radio frequency magnetron sputtering, following two steps: first, they were sputter-deposited in a non-reactive atmosphere (P02= 0 seem) for 600 s, and then oxygen with P02= 20 seem was immediately added to the growth chamber for 200 s, without changing any other deposition parameters. During both growth steps, HfC>2 and SrTiCh target powers were PHfo2= 50 W and PST0= 35 W, respectively.

[0121] Excluding the oxygen diffusion possibility, this growth design could theoretically result in a double-layer formation: (i) an ~15-nm-thick under-stoichiometric Hf(Sr,Ti)0i.s bottom layer and (ii) an ~1-nm-thick stoichiometric Hf(Sr,Ti)C>2 top layer. As discussed below, however, the two-step fabrication method was found to result in a single layer of stoichiometric Hf(Sr,Ti)C>2 separated from the TiN layer by a layer of TiOxNycontaining asymmetric oxygen and nitrogen gradients.

[0122] Device fabrication

[0123] Thin film structures fabricated using this two-step method were converted into electrical devices for characterisation of their electrical properties.

[0124] The device fabrication was carried out firstly by spin-coating a layer of positive UV photoresist (AZ 4533) on the surface of Si / a-SiO2 / TiN / Hf(Sr,Ti)C>2. The samples were then baked at 100°C for 2 min and exposed to UV light for 10 s through a photolithography mask. The resist was thereafter developed in an AZ351B developer. Then, the samples were coated with Mo by DC magnetron sputtering. Thereafter, the samples were immersed in ethanol without any sonication to remove the unexposed UV resist and lift off the metal on its top. Finally, they were baked for 10 min at 60 °C to evaporate ethanol residuals. This process resulted in fabricating circular-shape Mo top electrodes with different diameters on Si / a-SiO2 / TiN / Hf(Sr,Ti)O2.

[0125] Materials characterisations

[0126] Figure 2 illustrates Rutherford Backscattering Spectrometry (RBS) experimental spectra and SIMNRA simulations of Hf(Sr,Ti)Ozthin films grown on Si / a-SiO2 / TiN (a) in the non-reactive atmosphere and (b) following the two-step growth strategy. The RBS depth scale (x1015atoms / cm2) was converted to depth (nm) by using the average thickness measured by STEM. The ideal thin film structure for each growth design is schematically illustrated on top of each panel.

[0127] RBS analyses in Figure 2 indicate that the oxides grown in the non-reactive atmosphere (P02= 0 seem) are under-stoichiometric (O deficient) with an O / metal ratio of ~1.8, while those sputter-deposited following the two-step co-sputtering growth strategy of the present disclosure are stoichiometric (O / metal ratio = -2.0). In both cases, the multi-component oxides contain -4.0 at.% Ti and -3.6 at.% Sr. The slightly higher Ti concentration can be mainly due to their different sputtering yields. Opposite to the Hf(Sr,Ti)0i.s device for which TiN was not oxidised (Fig. 2(a)), the two-step Hf(Sr,Ti)O2growth leads to oxidation of the TiN bottom layer and the formation of a TiOxNylayer on TiN, under Hf(Sr,Ti)O2, see Fig. 2(b). The corresponding depth profile reveals that N in TiOxNygradually increases from -2.0 at.% at close to the TiOxNy / Hf(Sr,Ti)O2interface to -46.5 at.% at the TiN / TiOxNyinterface, while O decreases from -65.0 at.% to -5.0 at.%. Ti concentrations inside the TiN and TiOxNylayers remain almost constant at -33.0 at.% and -48.5 at.%, respectively. This implies the formation of O- and N-gradient TiOxNylayer sandwiched between TiN and Hf(Sr,Ti)O2.

[0128] Figure 3 summarises the transmission electron microscopy (TEM) results obtained from the Si / a-SiO2 / TiN / Hf(Sr,Ti)O2 / Mo devices. Shown in Figure 3 are: (a) Cross-sectional STEM image acquired from the pristine Si / a-SiO2 / TiN / Hf(Sr,Ti)O2 / Mo device; (b) HAADF-STEMimage from Hf(Sr,Ti)C>2; (c) High-resolution HAADF-STEM image from inside a Hf(Sr,Ti)C>2 column; (d) its corresponding fast Fourier transform micrograph.

[0129] Figure 4 compares the XRD 0-20 scans of as-deposited (a) TiN, (b) TiN / Hf(Sr,Ti)Oi.s, and (c) TiN / Hf(Sr,Ti)C>2 grown on the Si(001) / a-SiC>2 substrates. The peaks at 32.8° arise from the forbidden 002 Si(001 ) reflection that appears due to multiple scattering events. The highly dense and smooth TiN layers crystallise in a cubic crystal structure with a mixture of

[0111] and

[0200] orientations. The XRD pattern of TiN / Hf(Sr,Ti)Oi.s consists of the TiN peaks and an extra peak appearing at 20 = -30.3°, which can be assigned to the orthorhombic HfC>2-based phase with

[0111] orientation. This data indicates the formation of Hf(Sr,Ti)0i.s solid solution with no XRD-detectable phase separation. The 0-20 scan of TiN / Hf(Sr,Ti)C>2 preserves the TiN and HfC>2-based components, but also has a set of extra peaks that can be assigned to the formation of tetragonal TiC>2(N) phase.

[0130] The cross-sectional STEM image in Figure 3(a) confirms the RBS data showing the presence of TiOxNy. It also exhibits a ~16-nm-thick Hf(Sr,Ti)C>2 layer formed on top of TiOxNy. While TiN preserves its dense columnar microstructure, the TiOxNylayer consists of enlarged equiaxed crystallites. The Z-contrast (HAADF-STEM) image acquired from the Hf(Sr,Ti)C>2 thin film in Figure 3(b) reveals the formation of a columnar microstructure, comprising ~7-nm-wide nanocolumns that are aligned along the vertical growth direction. The high-resolution Z-contrast image from inside a column, Figure 3(c), indicates that Hf(Sr,Ti)C>2 is crystalline with an -3.5-A out-of-plane lattice distance. Figure 3(d) exhibits the corresponding fast Fourier transform micrograph obtained from the same region in Figure 3(c) that has diffraction spots with a hexagon-shape pattern, proving the Hf(Sr,Ti)C>2 crystalline structure that also agrees with the corresponding XRD data in Figure 4. Moreover, Figure 3(b) shows a contrast difference between columns and column boundaries. While the columns are bright, the column boundaries appear dark. These dark atomic-scale-thin areas can be related to the formation of low-Z (here, Ti and O) rich regions. The dark atomic-scale-thin column boundaries are Ti rich compared to the bright columns.

[0131] Memristive and neuromorphic characteristics

[0132] Figure 5 illustrates the memristive characteristics of the devices containing the thin film structure obtained by the two-step fabrication method of the present disclosure. Shown in Figure 5 are: (a) Resistive switching characteristics (first 30 l-V sweeping cycles) of a typical SiO2 / TiN / Hf(Sr,Ti)C>2 / Mo device; (b) LRS and HRS distribution data of 50 devices (obtained at -1.0 V); (c) Endurance data determined by collecting 1100 consecutive l-V sweepingcycles; (d) LRS and HRS values obtained from the endurance data in (c) at -1.0 V; (e) LRS and HRS data determined from pulsed endurance measurements by applying ±3.5 V pulses with read pulses of -0.2 V; (f) LRS and HRS retention data after applying preset pulses of ±3.5 V with read voltages of -0.2 V; (g) l-V curves for peak voltages ranging from ±0.75 to ±3.5 V with increments of ±0.25 V; (h) LRS and HRS values obtained from the l-V curves in (g) at -0.5 V; and (i) LRS and HRS values obtained from devices with dTE = 25, 50, and 100 pm at -0.2 V. The data in panels (a) to (h) are determined from devices with the Mo topelectrode diameter dTE = 50 pm.

[0133] Figure 5 summarises the room-temperature-obtained memristive characteristics of the Si / a-SiO2 / TiN / Hf(Sr,Ti)C>2 / Mo devices. The memristive properties in Figures 5(a) to 5(h) were determined from the devices with the Mo top-electrode diameter dTE of 50 pm. Figure 5(a) shows the first 30 current-voltage (l-V) curves of a typical Si / a-SiO2 / TiN / Hf(Sr,Ti)C>2 / Mo device. The pristine device is in its high resistance state HRS (-1.0 x 1010Q at +0.2 V). The device resistance exponentially decreases by applying a positive voltage to the Mo top electrode (~1.5 x 106Q at +3.5 V). This results in switching the device to its low resistance state LRS (SET process) without any need for an initial electroforming process or current compliance. Reversing the voltage polarity then switches the device to its HRS (RESET process). After acquiring the first l-V curve, the HRS / LRS ratio (memory window) slightly decreases for both voltage polarities, but it remains highly stable for the other l-V cycles. Figure 5(b) compares the LRS and HRS distributions of 50 different devices. For each device, the LRS and HRS datapoints represent the average values obtained from 50 l-V cycles at -1.0 V, excluding the first 5 cycles. All devices show highly stable hysteretic l-V curves that prove exceptional cycle-to-cycle uniformity, error bars in the inset in Figure 5(b). The resistance state distributions in Figure 5(b) also indicate excellent device-to-device uniformity with memory windows > 10.

[0134] Two different measurement strategies were employed to assess the cyclic endurance of these devices: (i) collecting 1100 consecutive l-V curves and (ii) acquiring LRS and HRS data for 5 x 104pulse-switching cycles. Figure 5(c) exhibits 1100 consecutive l-V sweeping cycles of a typical TiN / Hf(Sr,Ti)C>2 / Mo device. The l-V curves at the positive voltages show remarkable cycle-to-cycle uniformity. Although both LRS and HRS simultaneously undergo slight decreases at the negative voltages, the memory window does not significantly change and remains > 10, see Figure 5(d). The endurance data obtained from the pulse-switching measurement, where every single value of LRS and HRS was reported in Figure 5(e), indicate robust, stable resistance states up to 5 x 104switching cycles. Figure 5(f) shows thebinary-state retention data of a device after applying preset pulses of ±3.5 V and reading the resistance states at -0.2 V with 1-s intervals. Both LRS and HRS remain highly stable up to 5 x io5s with a memory window > 10.

[0135] Achieving multilevel non-volatile resistance states is a key figure of merit for neuromorphic applications. Figure 5(g) exhibits the l-V curves of a typical TiN / Hf(Sr,Ti)C>2 / Mo device, acquired at different SET-RESET voltages ranging from ±0.75 to ±3.5 V with increments of ±0.25 V. The presence of several distinct resistance levels can be confirmed in the negativevoltage side of the l-V curves. The corresponding LRS and HRS values determined at -0.5 V are plotted in Figure 5(h). While there is a slight increase in the HRS value from ~3.0 x 109Q at ±0.75 V to ~5.0 x 109Q at ±3.5 V, the value of LRS significantly decreases from ~2.0 x 109Q to ~8.0 x 107Q by increasing the SET-RESET voltage. This results in a gradual increase in the memory window from ~2.0 for ±0.75 V to -15.0 for ±2.75 V and then a rapid increase to -63.0 for ±3.5 V, as plotted in the inset in Fig. 5(h). The stability of these data was further evaluated by recording 50 l-V cycles for each SET-RESET voltage. The results show highly uniform l-V curves, confirming an excellent cycle-to-cycle uniformity for each sweeping voltage.

[0136] Figure 5(i) compares the average values of LRS and HRS as a function of the Mo topelectrode diameter (dys). These data were collected at the read voltages of -0.2 V from the 10thl-V curves of 50 different devices for each d?E. Both resistance states increase by decreasing dTE, but this upward trend is more significant for HRS compared to LRS. As a result, there is a substantial increase in the memory window from -4.0 for dTE = 100 pm to -32.0 for dTE = 50 pm, to -1900 for dTE = 25 pm. This is typically considered as a reliable indication of interface-dominated resistive switching. Overall, the TiN / Hf(Sr,Ti)O2 / Mo devices with the Hf(Sr,Ti)O2 switching layers fabricated by the two-step, co-sputtering growth approach exhibit robust memristive characteristics required for advanced data storage technologies.

[0137] Figure 6 illustrates neuromorphic characteristics of the devices containing the thin film structure obtained by the two-step fabrication method of the present disclosure. Shown in Figure 6 are long-term potentiation and depression as a function of the number of presynaptic spikes with various programming schemes: (a) identical spike scheme; spikes with amplitudes of 1.5, 2.0, and 2.5 V and time intervals dt = 1 ms; (b) and (c) identical spike scheme; 1 ,5-V spikes with various spike numbers (dt = 1 ms); (d) Conductance numbers and values of this work (pink data) compared to -50 reported HfC>2-based memristors; and (e) their Wmax / Wmin ratios; (f) and (g) multilevel retention data obtained by implementing 1.5-Videntical spikes; Long-term potentiation and depression data obtained by using (h) identical spike scheme, 1.5-V spikes with dt increasing from 1 to 30 ms in 1-ms increments; and (i) non-identical spike scheme, spikes with amplitudes changing from ±0.7 to ±2.5 V in ±0.1-V increments (dt = 1 ms).

[0138] Shown in Figure 7 are: (a) PPF and PPD ratios as a function of inter-spike time intervals Atppbetween two successive presynaptic spikes. / 1 and I2 are the maximum postsynaptic currents of the first and second spikes, respectively. Relative synaptic weight change AW as a function of time difference between presynaptic and postsynaptic spikes At realised by implementing triangle-shaped 1.5-V spikes: paired spikes are shown in Figure 7 with (b) similar and (c) opposite polarities. The solid black curves show exponential fits. All data were obtained at the read voltages of -0.2 V.

[0139] Figures 6 and 7 show the room-temperature-obtained key neuromorphic characteristics of the Si / a-SiO2 / TiN / Hf(Sr,Ti)C>2 / Mo devices. Similar to the biological neurons, it is essential for the synaptic electronic devices to mimic both memory and learning functionalities, with satisfying at least two key requirements: synaptic efficacy and plasticity. Synaptic efficacy refers to the ability of bio-inspired devices to alter stored synaptic weights W (conductance levels) in step-wise fashions upon applying presynaptic spikes, while synaptic plasticity is the device’s ability to modulate W by implementing a particular learning rule. In general, these devices need to have many distinct synaptic weights, ideally varying in linear and symmetrical manners.

[0140] Typical long-term potentiation and depression (synaptic efficacy) of the Hf(Sr,Ti)C>2-based devices, realised by implementing various identical and non-identical spike programming schemes, are plotted in Figure 6 as a function of presynaptic spikes (training voltage pulses). The presynaptic spikes of each scheme are also schematically illustrated above each panel. In all schemes, applying positive spikes increases the device postsynaptic conductance levels (potentiation), while the conductance decreases by negative-spike application (depression). These synaptic characteristics directly depend on the amplitudes and time intervals of the presynaptic spikes. Figure 6(a) exhibits the synaptic weights realised from an identical training scheme with 30 positive and 30 negative spikes that have different amplitudes (1.5, 2.0, and 2.5 V), at time intervals dt = 1 ms. Increasing the spike amplitude increases the weight-modulation dynamic range (Wmax / Wmin ratio) from ~6.0 for 1.5-V spikes to ~6.9 for 2.5-V spikes. Incremental and decremental weight changes are distinct with proper precision, but not a perfect symmetry and linearity due to the complex electronic-ionic nature of these devices, discussed later. The potentiation nonlinearity coefficients (<p)determined by applying identical 1.5-V, 2.0-V, and 2.5-V spikes are (p = 0.11, 0.08, and 0.06, respectively, which are close to the ideal value (p = 0.

[0141] The operational capability of these artificial synapses for various spiking neural network applications was further assessed by implementing an identical programming scheme with 1.5-V spikes for which the total number of positive and negative spikes increases from 100 to 12000 (dt = 1 ms), see Figures 6(b), 6(c). For all pairs of nerve impulses, W steadily changes with proper state separations. Figure 6(c) demonstrates the excellent replicability of the 6000-training scheme, which is also confirmed for several different devices -- in agreement with their memory-based device-to-device uniformity shown in Figure 5.

[0142] Figure 6(d) and 6(e) compare the weight numbers and values of all these works and their Wmax / Wmin ratios. We report hundreds of reproducible levels ranging from ~2.5 x 10'9to ~1.4 x 10'7S with Wmax / Wmin ratio = 56. The non-volatile nature of these conductance levels is demonstrated in Figures 6(f) and 6(g), indicating > 300 stable and distinguishable states. Each level was obtained after implementing 20 identical spikes with 1.5-V amplitudes.

[0143] Training the devices with fewer spikes caused the retention data to overlap, mainly due to noise levels present in ultra-low currents. The achieved number of non-volatile conductance levels is almost three orders of magnitude higher than the ones required for deep neural networks.

[0144] To enhance the symmetry and linearity in the conductance modulation, the present inventor optimised the programming schemes by implementing (i) identical 1.5-V spikes with dt increasing from 1 to 30 ms in 1-ms increments, Figure 6(h), and (ii) non-identical spikes with amplitudes changing from ±0.7 to ±2.5 V in ±0.1 -V increments (dt = 1 ms), Figure 6(i). The results prove the improved symmetric and linear synaptic efficacy with significantly low conductance values, ideal for sustainable Al hardware.

[0145] The outstanding operational synaptic stability of the Hf(Sr,Ti)C>2-based devices was confirmed by utilising identical and non-identical training schemes with 1320 consecutive cycles. Both potentiation and depression are appropriately symmetric and linear for all cycles and have highly stable, distinct weights. T raining with identical 1 ,5-V spikes provides an ultralow conductance change, varying between -9.0 x 10'10S and -1.2 x 10'9S.

[0146] Figure 7(a) shows paired-pulse facilitation and depression (PPF and PPD) ratios, triggered by applying a pair of spikes, as a function of inter-spike time intervals Atpp. Such short-termsynaptic plasticity is essential for decoding temporal information in visual and auditory signals. The PPF and PPD ratios are defined as / 2 / / 1 x 100%, where / 1 and I2 are the maximum postsynaptic currents of the first and second spikes, respectively. Positive spikes generate PPF, whereas negative spikes produce PPD. While the PPF ratio with I2 > / 1 decreases by -6% with increasing Atpp, the absolute value of the PPD ratio with / 1 < I2, which leads to negative values given in Figure 7(a), follows a decrease of -15%. In the PPF measurements, for Atppshorter than the relaxation time (T) of mobile charge carriers triggered by the first spike, a fraction of these carriers does not gain adequate time to diffuse back to their equilibrium positions. Hence, the postsynaptic current is facilitated upon applying the second spike. An opposite manner can be expected for the PPD responses. The best fits for these experimental data have a double-exponential decay function that perfectly matches the short-term biological synaptic plasticity, which consists of two distinct phases: (i) a rapid phase that lasts for a few milliseconds and (ii) a slow phase with a longer millisecond duration [2]:

[0147]

[0148] where A and r are the initial facilitation(depression) magnitude and relaxation time for each phase, respectively. For PPF, Y = 100%, A1 = 3.8%, A2 = 4.6%, TX= 1.4 ms, and T2=15.4 ms, while for PPD, Y = -100%, A1 = -13.8%, A2 = -6.7%, TX= 2.2 ms, and T2= 14.9 ms. Having TX» T2demonstrates the presence of both rapid and slow phases in the Hf(Sr,Ti)C>2-based artificial synapses, in agreement with the biological synapse responses.

[0149] Spike-timing-dependent plasticity (STDP) is a key unsupervised learning rule in neuromorphic devices for time-dependent functionalities such as speech recognition and image detection, in which the synaptic weights are modulated as a function of time difference between presynaptic and postsynaptic spikes At [3], Figs. 7(b) and 7(c) show relative STDP weight changes AW realised by implementing two different conventional programming schemes with triangle-shaped 1.5-V spikes: paired spikes with similar and opposite polarities. The weight magnitude changes are directly controlled by At time intervals. The Hf(Sr,Ti)C>2-based artificial synapse is initially activated upon firing the first spike. While its conductance gradually decreases after the first stimulation, the second spike is applied at a certain At to strengthen or weaken the device conductance, depending on the training spike state. Thus, the largest relative weight changes, |AW|, occur at the smallest |At| values, and |AW| significantly decreases by increasing |At|. Here, both training schemes result in asymmetric, anti-Hebbian STDP effects. Potentiation is achieved by firing postsynaptic spikes before presynaptic ones (At < 0), while depression is generated by applyingpostsynaptic spikes after presynaptic spikes (At > 0), schematically illustrated in insets in Figs. 7(b) and 7(c). Similar to STDP in biological neural networks, the artificial synaptic weight-change data can be fit well with the following equation [4]:

[0150]

[0151] where A± and

[0152]

[0153] are the exponential-function scaling factors and time constants, respectively. For the paired-spike scheme with similar polarities, Fig. 7(b), A± values are - values are 4.6 and -4.3, while A± values are -141.7 and +134.3,

[0154]

[0155] -4.9 for the paired-spike scheme with opposite polarities, Fig. 7(c). The replicability of these STDP data was also confirmed from different other devices, demonstrating that the Hf(Sr,Ti)C>2-based synaptic devices can effectively emulate the spatiotemporal biological STDP learning rules within millisecond-scale learning windows.

[0156] P-type electronic conductivity in Hf(Sr,Ti)O2

[0157] Figure 8 illustrates the principal characteristics and conductance modulation mechanism model of device operation. Shown in Figure 8 are: defect and dopant formation energies calculated for (a) under-stochiometric (O-deficient) and (b) stochiometric (O-rich) Hf(Sr,Ti)C>2 at equilibrium; (c) schematic illustration summarising the conductance modulation model proposed for the Si / a-SiO2 / TiN / Hf(Sr,Ti)C>2 / Mo devices; and (d) corresponding energy band diagrams changing upon spike applications. Figure 8 illustrates (a), (b) principal characteristics and conductance modulation mechanism mode and, (c),(d), electrical characteristics of thin film electrical devices according to an aspect of the disclosure, and the conductance modulation mechanism model.

[0158] Opposite to most HfO2-based memristors that show dominant filamentary switching, the Si / a- SiO2 / TiN / Hf(Sr,Ti)O2 / Mo devices exhibit bipolar diode-like characteristics with pure interfacial conductance changes. To determine the key modulating interface, the inventors first evaluated the TiN / TiOxNystack with removing Hf(Sr,Ti)O2 and thus studied Si / a- SiO2 / TiN / TiOxNy / Cr / Au devices. The corresponding l-V curves do not exhibit any hysteretic loops, and results confirm that the TiN / TiOxNyand Hf(Sr,Ti)O2 / Mo interfaces do not significantly influence the conductance changes.

[0159] Based on l-V curves such as those in Figure 5(a), the present inventor postulates that the observed ultra-low conductance modulation is primarily attributed to the formation of a p-n like heterointerface between p-type Hf(Sr,Ti)C>2 and n-type, O-rich TiOxNy. The asymmetricl-V curves of the TiN / Hf(Sr,Ti)C>2 / Mo devices consist of a current rectification at positive voltages that follows a significant increase in resistance during the reverse voltage sweeps. The l-V curves of these devices were compared with those of SrTi03 / Lao.7Sro.3MnC>3 / Hf(Sr,Ti)02 / Mo that form a metal / oxide-like bottom interface with Schottky-emission-dominated switching. The rectification ratio is -155 forTiN / Hf(Sr,Ti)C>2 / Mo and -2.6 for Lao.7Sro.3Mn03 / Hf(Sr,Ti)C>2 / Mo. The considerable increase in resistance during reverse voltage sweeps together with the pronounced rectification ratio observed in the TiN / Hf(Sr,Ti)C>2 / Mo devices are typical electrical characteristics of p-n heterointerfaces.

[0160] The n-type semiconductor nature of TiC>2 is already well known [5], However, p-type electronic conductivity in Hf(Sr,Ti)C>2 is expected based on our work’s hypothesis (creation of p-type donors from lower-valent ion doping on the Hf4+sites [6, 7]). Thus, the inventor carried out a series of ab-initio calculations and Hall effect measurements to ascertain p-type electronic conductivity in Hf(Sr,Ti)C>2. The equilibrium transition energy-level diagram of Hf(Sr,Ti)C>2 in Figure 8(a) indicates that inducing p-type conductivity under O-deficient conditions is unlikely due to charge compensation effects [8], However, in O-rich environments, although Ti does not have any p-type doping influences as its defect levels are situated well above the valence band maximum (VBM), we found that Sr dopants can effectively lower the Fermi level below the equilibrium state, as shown in Figure 8(b). Carrier concentration calculations also prove holes as the majority charge carriers in Hf(Sr,Ti)C>2. These findings, for the first time, demonstrate that Sr acts as a p-type dopant in stoichiometric Hf(Sr,Ti)C>2. Consistent with these results, the positive Hall coefficient depicted in Figure 9, which shows the Hall effect voltage as a function of magnetic field measured at 375 K, further corroborates the presence of p-type conduction in the sputter-deposited Hf(Sr,Ti)C>2 thin films. The inventor considers that the key reasons for the formation of p-type Hf(Sr,Ti)C>2 stem from the synergistic effects of (i) engineering its bandgap and (ii) substituting Hfn+in the Hf cation sublattice with Sr2+acceptor dopants. For stoichiometric layers, UV-vis measurements and density-of-state calculations reveal a decrease in the bandgap from -6.0 eV for HfC>2 to -4.5 eV for Hf(Sr,Ti)C>2, which is mainly attributed to adding Ti dopants (see Figure 10). This significant bandgap reduction largely overcomes the doping limitations associated with wide-bandgap HfC>2, enabling Sr in an O-rich environment to increase the concentration of holes and thus, effectively shift the Fermi level closer to VBM.

[0161] Figures 10(a) and 10(b) show the UV-vis transmittance spectra and Tauc plots of the stoichiometric HfO2 and Hf(Sr,Ti)O2 layers grown on quartz substrates. Over a wide range of wavelength, both layers have transmittance > -90 %. Optical bandgap energies estimatedfrom the Tauc plots of HfC>2 and Hf(Sr,Ti)C>2 are Eg= ~5.7 and ~4.5 eV, respectively. These values are in good agreement with those obtained from calculated optical absorption spectra that are -6.1 eV for HfC>2 and ~4.7 eV for Hf(Sr,Ti)C>2.

[0162] The present devices are distinguished from existing memristors by their highest resistance states and largest interfacial-memristive memory window.

[0163] Conductance modulation mechanism model

[0164] Figure 11 illustrates: (a) Isolated energy band diagrams of TiN, TiC>2 (here, O-rich TiOxNy), Hf(Sr,Ti)C>2, and Mo; and (b) Energy band diagram of TiO2(here, O-rich TiOxNy) / Hf(Sr,Ti)O2 heterostructure at the pristine state (thermal equilibrium).

[0165] Figure 11(a) schematically illustrates the isolated energy band diagrams of TiN, TiO2 (here, O-rich TiOxNy), Hf(Sr,Ti)O2, and Mo. Work function (0), electron affinity ( / ), and bandgap (Eg) values are 0 = ~4.5 eV, = ~4.0 eV, and Eg= ~3.2 eV for O-rich TiOxNy, and 0 = ~5.6 eV (UPS determined), = -1.8 eV (since the value for Hf(Sr,Ti)O2 is unknown, we use that of HfO2. Similar approximation was made for O-deficient sputter-deposited HfO2 with Eg= -4.5 eV), and Eg= -4.5 eV (UV-vis determined) for Hf(Sr,Ti)O2. As illustrated in Figure 11(a), the Fermi energy level lies close to the valence band maximum (VBM) for p-type Hf(Sr,Ti)O2, while it is near the conduction band minimum (CBM) for n-type O-rich TiOxNy. The energy band diagram for the O-rich TiOxNy / Hf(Sr,Ti)O2 heterostructure at the pristine state (thermal equilibrium) is schematically shown in Figure 11(b). The work function difference between O-rich TiOxNyand Hf(Sr,Ti)O2 generates a built-in potential Vbj of -1.1 V (theoretical value obtained from the band diagram). This value is in the range of Vw determined from our current-voltage and capacitance-voltage measurements, -1.0 V and -1.2 V, respectively. Since the TiOxNylayer is more defective than stochiometric Hf(Sr,Ti)O2, it has significantly higher carrier concentration. Thus, the built-in potential is expected to largely form in the Hf(Sr,Ti)O2 thin film that causes a wider depletion region in this p-type layer: forming a space-charge region asymmetrically extended inside Hf(Sr,Ti)O2.

[0166] The inventor estimated the width of the depletion region (Wd) inside Hf(Sr,Ti)O2 using the

[0167] equation [1],

[0168]

[0169] where E0is the vacuum permittivity (8.854 x 10'14F / cm), eris the oxide dielectric permittivity (using that of Sr-doped HfC>2, ~30, which is also close to that of Ti-doped HfC>2) ,q is the electron charge (1.6 x 10'19C), and Ndis the carrier concentration. Overall, depending on O deficiency and dopant concentrations, three different regimes of low (~1.0 x 1017cm-3), medium (~1.0 x 1019cm-3), and high (~1.0 x 1021cm-3) carrier concentrations can be considered for HfO2 with close to stoichiometry compositions. Thus, Wdcan range from -190 to -19 to -2 nm. Based on calculations carried out on Hf(Sr,Ti)O2 with an ideal crystal structure at equilibrium conditions, using the py-sc-fermi package, the estimated concentrations of electrons and holes in p-type stoichiometric Hf(Sr,Ti)C>2 are -3.5 x 10'19cm-3and -2.3 x 1018cm-3, respectively. Clearly, having holes as the majority charge carriers further supports the p-type nature of Hf(Sr,Ti)C>2. This concentration of holes, which is comparable with the present Hall measurement results, is almost three orders of magnitude less than the electron concentration in n-type, O-rich TiOxNy(-4.9 x 1021cm-3, determined from capacitance-voltage measurements and is in the range reported for typical n-type TiC>2). This confirms the hypothesis that the space-charge region largely extends into Hf(Sr,Ti)C>2. The calculated hole concentration lies between the low and medium Ndregimes, resulting in Wd= -40 nm that can exceed the Hf(Sr,Ti)C>2 thickness. This shows the formation of a thick depletion region that is asymmetrically extended inside Hf(Sr,Ti)C>2. This unique electronic structure can explain the low conductance values observed in the devices of the present disclosure and their asymmetric trends that vary significantly with the polarity of the presynaptic spikes, see Figures 7(a)-(c).

[0170] The present inventor proposes the analogue conductance modulation observed in these interfacial synaptic devices occurs via two main mechanisms that operate concurrently: (i) the transfer of electronic charge carriers across the TiOxNy / Hf(Sr,Ti)O2 interface and (ii) the electro-migration of negative ionic carriers inside the bulk TiOxNyand Hf(Sr,Ti)C>2 layers, as schematically illustrated in Figure 8(c). Figure 11(a) shows the isolated energy band diagrams of TiN, O-rich TiOxNy, Hf(Sr,Ti)O2, and Mo, where the Fermi energy level lies close to VBM inside p-type Hf(Sr,Ti)O2, while it is found near the conduction band minimum in n-type O-rich TiOxNy. At the pristine state (equilibrium), the devices are originally in their HRS due to a large built-in potential generated at the n-TiOxNy / p-Hf(Sr,Ti)O2 heterointerface, see Figure 11(b). The defective TiOxNylayer has a significantly higher carrier concentration than stochiometric Hf(Sr,Ti)O2. Thus, as schematically shown in Figure 8(d), the built-in potential mainly forms inside the Hf(Sr,Ti)O2 film, creating a depletion region asymmetrically extended in this layer. This calculation also estimates that the thickness of this region can even exceed that of Hf(Sr,Ti)O2, further discussed in Figure 11.By applying positive training spikes, holes and electrons are gradually injected into the depletion region from the Mo top electrode and n-type O-rich TiOxNy, respectively (see Fig.

[0171] 8(d)). Since the devices were programmed by identical spikes with relatively low voltage amplitudes, we expect the density of injected carriers to be higher than thermally generated charges. The charge carriers are increasingly captured into trapping sites such as point defects, mainly vacancies that typically have high concentrations in the lattice of sputter-deposited layers, and cations in the Hf(Sr,Ti)C>2 sublattice. This agrees with l-V curve analyses indicating that the conduction is governed by the space-charge limited current model. The carrier transfer during device programming is also indirectly demonstrated by XPS and EELS data, e.g., Fig. 5(i) shows an increase of ~4% in the Hf4+fraction after applying positive spikes. Thus, the controlled carrier transport steadily decreases the thickness of the depletion region, lowers the barrier height at the p-n heterointerface, and hence, increases the device conductance in an analogue manner. The remarkably low conductance values of the present devices can be mainly attributed to two key factors: (i) the wide depletion region and (ii) the limited defect concentration (mainly Vo) and hole mobility inside the p-type Hf(Sr,Ti)C>2, which the latter is significantly lower than in typical p-type semiconductors.

[0172] In addition to the interfacial electronic carrier transfer, EELS, XPS, and in-situ Raman measurements also revealed bulk ionic drifts upon the positive-spike applications. The results indicate that the negatively-charged N and O ions undergo a long-range diffusion inside TiOxNytowards Hf(Sr,Ti)C>2. This gradually reduces TiOxNyand increases the thickness of the N-rich region and hence, the TiOxNyconductivity. This ionic-drift process can also decrease the width of the depletion region at the p-n heterointerface, contributing to increasing the device conductance upon positive spiking.

[0173] Conversely, inverting the spikes’ polarity generates a reverse current, the depletion region becomes wider, trapped carriers are released from the trapping sites, and negatively-charged ions electro-migrate from the Hf(Sr,Ti)C>2 layer towards the TiN / TiOxNyinterface, which further oxidises TiN and extends the TiOxNythickness, as shown in Figs. 5(l) and 8(c). The asymmetric conductance changes obtained upon applying identical negative spikes, shown in Figs. 7(a)-7(c), can be mainly related to the asymmetric nature of the space-charge region that is much wider inside Hf(Sr,Ti)C>2 than TiOxNy, originating from the unequal defect and carrier concentrations in Hf(Sr,Ti)C>2 and TiOxNy. The inventor also attributes the excellent device-to-device uniformity observed in these devices’ resistive switching andneuromorphic performance to the laterally uniform distribution of point defects in the oxide layers.

[0174] Materials characterisation

[0175] Electrical and neuromorphic measurements

[0176] The Si / a-SiO2 / TiN / Hf(Sr,Ti)C>2 / Mo devices were employed for memory and neuromorphic characterisations. In all measurements, which were carried out at room temperature, the voltage was applied to the Mo top electrodes, while the TiN bottom electrode was grounded. The conductivity of the grounded area was a few ohms. The non-volatile memory and neuromorphic performance of these devices were obtained using a computer-controlled Keysight B2912A connected to a probe station.

[0177] Hall measurements

[0178] Hall measurements were undertaken on Hf(Sr,Ti)C>2 grown on quartz substrates in a DynoCool physical property measurement system, from Quantum Design, connected to the Keysight B2912A system.

[0179] Rutherford backscattering spectrometry (RBS)

[0180] The elemental compositions of the Hf(Sr,Ti)O2 thin films grown on Si(001) / a-SiQ2 / TiN substrates were obtained by RBS in a 5-MV 15SDH-2 tandem accelerator. 2-MeV4He+ions were employed for RBS measurements, and backscattered ions were detected at a scattering angle of 170°. Possible ion-channelling effects were minimised by adjusting the equilibrium incidence angle to 5° with respect to the surface normal and performing multiple-small-random-angular movements within a range of 2° during data acquisition.

[0181] X-ray diffraction and reflectivity (XRD and XRR)

[0182] XRD and XRR scans were carried out in a PANalytical Empyrean high-resolution x-ray diffractometer operated at 45 kV and 40 mA with a Cu Kasource (A = 0.15406 nm) to determine the crystal structure, thickness, roughness, and density of the thin films.

[0183] Scanning transmission electron microscopy (STEM)

[0184] Cross-sectional STEM analyses were carried out in a monochromated probe Csaberration-corrected ThermoFisher Scientific Spectra 300 electron microscope operated at 300 kV monochromated < 100 meV at 100 pA. Images were acquired in both annular-bright-field and high-angle-annular-dark-field (HAADF) modes. Electron energy-loss spectroscopy (EELS) spectra were acquired employing a Gatan Continuum 1066 EELS spectrometer energy resolution of 150 meV per channel integrated over 1 s. Absolute volumetric density spectra for N-K, Ti-L, and O-K edges were calculated using Gatan Microscopy Suite 3.52 with implemented routines for multiple linear least squares (MLLS) fitting models with Hartree-Slater scattering cross-sections and including zero-loss centring, power law background subtraction, plural scattering removal, and excluding energy-loss near edge structure (ELNES) regions of 20 eV. TEM specimens were prepared by focused ion beam (FIB) technique employing a FEI Helios Nanolab DualBeam instrument..

[0185] Raman spectroscopy

[0186] In-situ Raman spectroscopy was performed using Integrated Optics' continuous wave (GW) laser at a single longitudinal mode, wavelength of 633 nm, and power of 1.5 mW. Raman signals were transmitted to an Andor Kymera 328i spectrometer connected to an Oxford Instruments' Newton EMCCD camera. The sample was electrically connected to a 7-nm-thick conductive top electrode (Cr / Au) to enable in-situ Raman measurements. Instead of aprobe tip, potential was applied to the sample with a conducting cantilever to avoid puncturing the top electrode into the sample. Apex Probes SD-qp-CONT-TL cantilever tips were coated with a 3-nm-thick Cr / 6-nm-thick Au layer to apply potential. The Keithley 2634B source meter was utilised for in-situ electrical measurements.

[0187] X-ray photoelectron spectroscopy (XPS)

[0188] Room-temperature, depth-resolved XPS core levels were obtained in a Thermo Scientific Escalab 250Xi instrument with monochromatic Al Karadiation (hv = 1486.6 eV) at -1 x 10'9Torr using 400-eV Ar-ion surface sputter-etching. To prevent the destructive influences of sputter-etching on XPS core levels and fabricate top electrodes, after the Hf(Sr,Ti)C>2 growth, each sample was immediately transferred to an inert-atmosphere glovebox, then a shadow mask was fixed on top of that, and immediately afterwards was loaded into another sputtering chamber, dedicated to metal deposition. After sputter-depositing circular-shape Mo top electrodes (with thicknesses and diameters of -16.0 nm and -1.0 mm, respectively) on the Si / a-SiO2 / TiN / Hf(Sr,Ti)C>2 surface, two devices were positively- and negatively-spiked (4000 pulses of 11.5| V), and finally, the sample was immediately transferred to the XPS instrument. During all transport steps, the sample was kept in a vacuum vessel which could be connected to both glovebox and XPS instrument. This preparation protocol avoided the sample airexposure. The Mo thin films were kept in electrical contact with the XPS sample holder to prevent possible peak shifts resulting from sample charging. Depth-resolved, high-energy-resolution XPS core levels were acquired from 400 x 400 pm2regions located in the centre of 1 x 1 mm2sputter-etched areas. The spectra’s binding energies were calibrated against their Fermi edge cut-offs. Elemental quantification and Hf 4f XPS peak deconvolution were carried out using CasaXPS software after the Shirley-type background subtraction. The Hf 4f core-level spectra were deconvoluted by retaining the same 4f5 / 2-4f? / 2 binding energy separations (1.6 eV), line shapes (Gaussian-Lorentzian), full-width-at-half-maximum values (1.6 eV), and 4f5 / 2:4f? / 2 area ratios (3:4), while peak areas and positions were changed. XPS depth scales (time) were converted to depth (nm) by using the average thicknesses obtained from STEM images.

[0189] Ultraviolet photoelectron spectroscopy (UPS)

[0190] Work functions of TiN and Mo layers were determined by UPS measurements conducted in Thermo Scientific Escalab 250Xi instrument using He I radiation ( / ico = 21.22 eV). To remove native surface oxides, sample surfaces were gently sputter-etched using 200-eV Ar ions. Ultraviolet visible (UV-vis) spectroscopy

[0191] The optical energy bandgaps (Eg) of stoichiometric HfC>2 and Hf(Sr,Ti)C>2 layers sputter-deposited on quartz substrates were determined from their absorption spectra acquired using a Shimadzu UV-3600i plus spectrophotometer, following the Tauc method: a hv)2= A( / iv -Eg), where a, h, and v are the absorption coefficient, Planck constant, and photon’s frequency, and A is a constant. The Egvalues were obtained by extrapolating the linear parts of the Tauc plots, (a / iv)2vs hv, to x-axis intersection points.

[0192] Ab-initio Calculations

[0193] First-principles calculations were conducted using Vienna ab-initio Simulation Package (VASP) with spin-polarisation. A 550-eV cutoff energy was employed for plane wave basis set. The PAW pseudopotential (dataset version PBE_54) of Sr_sv, Ti_pv, Hf_pv, and O were used, with Heyd-Scuseria-Ernzerhof hybrid functional (HSE06) to obtain bandgaps closer to experimental values. AiiDA framework was used for data provenance and workflow automation. Initial orthorhombic phase HfC>2 (space group of Pca2i) structure was obtained from the Materials Project (mp-685097). Full relaxation was performed using a force threshold of 0.03 eV / A, with a F-centred -point grid density of 0.05X2TT A-1. TO calculate the formation energies of different defects, a 96-atom 2x2x2 supercell was constructed. TheBrillouin zone of the supercell was sampled using a single F point to reduce computational costs. To ensure the reliability of this approach, we also calculated formation energies with a F-centred 2x2x2 -point grid for several cases and found negligible difference. The formation energy of defects D in the charge state q can be expressed as

[0194]

[0195] where Eo,q and EH are total energies of defect and host supercell, respectively. / Ji is chemical potential of type / atom, with n, atoms added or removed to form the defect. EFis the Fermi level, and Eco is the adopted Kumagai-Oba finite-size charge correction. The preparation and post-processing of each charged defect supercell were carried out using the doped package. Moreover, the defect concentration and self-consistent Fermi level were analysed using py-sc-fermi.

[0196] References

[0197] [1] W. Zhang, B. Gao, J. Tang, P. Yao, S. Yu, M.-F. Chang, H.-J. Yoo, H. Qian, H. Wu, Neuroinspired computing chips, Nature Electronics 3(7) (2020) 371-382.

[0198] [2] Y.H. Liu, L.Q. Zhu, P. Feng, Y. Shi, Q. Wan, Freestanding Artificial Synapses Based on Laterally Proton-Coupled Transistors on Chitosan Membranes, Advanced Materials 27(37) (2015) 5599-5604.

[0199] [3] I. Chakraborty, A. Jaiswal, A.K. Saha, S.K. Gupta, K. Roy, Pathways to efficient neuromorphic computing with non-volatile memory technologies, Applied Physics Reviews 7(2) (2020).

[0200] [4] M. Shahsavari, P. Devienne, P. Boulet, Spiking Neural Computing in Memristive Neuromorphic Platforms, in: L. Chua, G.C. Sirakoulis, A. Adamatzky (Eds.), Handbook of Memristor Networks, Springer International Publishing, Cham, 2019, pp. 691-728.

[0201] [5] V.C. Anitha, A.N. Banerjee, S.W. Joo, Recent developments in TiO2 as n- and p-type transparent semiconductors: synthesis, modification, properties, and energy-related applications, Journal of Materials Science 50(23) (2015) 7495-7536.

[0202] [6] T. Zhu, L. Ma, S. Deng, S. Liu, Progress in computational understanding of ferroelectric mechanisms in HfO2, npj Computational Materials 10(1) (2024) 188.

[0203] [7] M.H. Park, T. Schenk, M. Hoffmann, S. Knebel, J. Gartner, T. Mikolajick, U. Schroeder, Effect of acceptor doping on phase transitions of HfO2 thin films for energy-related applications, Nano Energy 36 (2017) 381-389.

[0204] [8] H. Li, Y. Guo, J. Robertson, Dopant compensation in HfO2 and other high K oxides, Applied Physics Letters 104(19) (2014).

[0205] [9] A. Goyal, P. Gorai, H. Peng, S. Lany, V. Stevanovic, A computational framework for automation of point defect calculations, Computational Materials Science 130 (2017) 1-9.

Claims

Claims1. A method of fabricating a thin film structure comprising the steps of:providing a layer of metal nitride MN;in an atmosphere with a first oxygen partial pressure, co-depositing HfC>2 and a perovskite with formula ABO3 on the metal nitride layer to form a first layer of multicomponent metal oxide with formula Hf(A, B)OZ, wherein 1.7 < z < 2; and, in an atmosphere with a second oxygen partial pressure different to the first oxygen partial pressure, co-depositing HfC>2 and the perovskite on the first layer of multicomponent metal oxide.

2. The method of claim 1, comprising the step of co-depositing HfC>2 and the perovskite on the first layer by sputtering from a HfC>2 target and a perovskite target.

3. The method of claim 1 or 2, in which the metal M of the metal nitride layer comprises one or more of Ti, Ta, Zr, Mo, and W.

4. The method of claim 1 , 2 or 3, in which the perovskite dopant is SrTiCh, such that A is Sr and B is Ti.

5. The method of any preceding claim, in which the first oxygen partial pressure is lower than the second oxygen partial pressure, preferably in which the first oxygen partial pressure is zero.

6. The method of any preceding claim, in which the second oxygen partial pressure is achieved with an oxygen flow rate through the reaction chamber of at least 10 standard cubic centimetres per minute (seem), preferably at least 15 seem, particularly preferably at least 20 seem, at a total deposition pressure of between 10 and 30 mTorr.

7. The method of any preceding claim, in which, other than the oxygen partial pressure, the first layer and second layer are deposited under the same conditions.

8. A thin film structure comprising:a layer of metal nitride MN;a layer of metal oxynitride MOxNyon the layer of metal nitride, the layer of metaloxynitride having a thickness of at least 10 nm; anda layer of multi-component metal oxide with formula Hf(A,B)C>2on the metal oxynitride layer,wherein the oxygen content x of the metal oxynitride MOxNylayer increases non- uniformly across the thickness of the metal oxynitride layer and is highest at a boundary between the metal oxynitride layer and the oxide layer, and wherein the nitrogen content y decreases non-uniformly across the thickness of the metal oxynitride layer and is lowest at the boundary between the metal oxynitride layer and the oxide layer.

9. The thin film structure of claim 8, in which the layer of metal oxynitride with formula MOxNy is an n-type layer of metal oxynitride, and the layer of multi-component metal oxide with formula Hf(A,B)O2is a p-type layer of Hf(A,B)O2, such that the thin film structure comprises a p-n heterointerface between p-type Hf(A,B)O2 and n-type MOxNy.

10. The thin film structure of claim 8 or 9, in which A and B are components of a perovskite material, preferably in which ABO3 is a perovskite.

11. The thin film structure of any of claims 8 to 10, in which the metal nitride MN is TiN, orTaN, ZrN, MoN, orWN.

12. The thin film structure of any of claims 8 to 11 , in which A and B are metals, preferably in which A is Sr and B is Ti.

13. The thin film structure of any of claims 8 to 12, in which the multi-component metal oxide is Hf(Sr,Ti)O2.

14. The thin film structure of any of claims 8 to 13, in which the layer of metal oxynitride has a thickness of at least 15 nm, or at least 25 nm, or at least 50 nm.

15. The thin film structure of any of claims 8 to 14, in which the layer of multicomponent metal oxide Hf(A,B)C>2 has a thickness of 10.5 nm to 25 nm, or 12.5 nm to 20 nm, or 14 nm to 18 nm.

16. The thin film structure of any of claims 8 to 15, in which the multi-component metal oxide layer is crystalline, and has an orthorhombic or rhombohedral crystal structure.

17. The thin film structure of any of claims 8 to 16, wherein at the boundary between the metal oxynitride layer and the oxide layer, y is less than 0.1, preferably less than 0.05, and / or at a boundary between the metal oxynitride layer and the metal nitride layer, y is at least 0.4, preferably at least 0.45.

18. The thin film structure of any of claims 8 to 17, wherein at the boundary between the metal oxynitride layer and the oxide layer, x is least 0.5, preferably at least 0.55 or at least 0.6, and / or wherein at a boundary between the metal oxynitride layer and the metal nitride layer, x is less than 0.15, preferably less than 0.1, particularly preferably less than 0.08.

19. The thin film structure of any of claims 8 to 18, wherein the layer of multi-component metal oxide with formula Hf(A,B)C>2 has a transmittance of at least 80% for wavelengths of 300 - 800 nm.

20. A method of manufacturing an electronic component from the thin film structure of any of claims 8 to 19, comprising the step of forming a conductive electrode on the multi-component metal oxide layer.

21. An electronic component comprising the thin film structure of any of claims 8 to 19, and an electrode on the oxide layer.

22. A non-volatile memory device comprising the thin film structure of any of claims 8 to 19.

23. A 2-terminal insulating(oxide)-based non-volatile memory device comprising the thin film structure of any of claims 8 to 19.

24. A memristor comprising the thin film structure of any of claims 8 to 19.

25. A neuromorphic memory device comprising the thin film structure of any of claims 8 to 19.

Images