Physical artificial neuron and method of manufacture thereof
The use of polycrystalline memristive layers in artificial neurons addresses the energy consumption issue of CMOS configurations by integrating and leaking in response to energy, facilitating efficient neuromorphic computing through simplified manufacturing of neurons and synapses.
Patent Information
- Application Number
- PCT/FI2026/050230
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-07
- Filing Date
- 2026-05-05
- Publication Date
- 2026-11-12
AI Technical Summary
Current artificial physical neurons require external power sources due to complex CMOS configurations, leading to unsatisfactory energy consumption and hindering the development of neuromorphic computing.
A physical artificial neuron comprising a polycrystalline memristive layer that integrates and leaks in response to applied energy, allowing for the simultaneous manufacture of neurons and synapses using polycrystalline and monocrystalline memristive materials, respectively, without the need for external power sources.
The solution simplifies the structure and manufacturing process of artificial neurons and synapses, reducing energy consumption and enabling efficient neuromorphic computing by leveraging the integrate-and-leak functionality of polycrystalline memristive materials.
Smart Images

Figure FI2026050230_12112026_PF_FP_ABST
Abstract
Description
PHYSICAL ARTIFICIAL NEURON AND METHOD OF MANUFACTURE THEREOFFIELD
[0001] The present disclosure relates to the field of memristors and memristive materials. At least some embodiments relate to neuromorphic chips and / or neuromorphic computing.BACKGROUND
[0002] Memristors are electronic components with current-conducting properties dependent on previous electric conditions they have been subjected to. Properties of memristors include properties such as adjustable resistance and low energy consumption between states of different resistivity that make memristors useful for various different applications. Neuromorphic computing is an approach to computing that is inspired by the structure and function of the brain. Memristive materials and memristors are typically used for their ability to retain a (resistance) state without the need for an external energy source. Therefore, the “memory” of such materials may be used for artificial synapses that may retain their prescribed state, and thus retain, store or remember information. Currently, artificial physical neurons implement, for example, complex complementary metal-oxide semiconductor, CMOS, configurations that require external power sources in order to function. Such unsatisfactory energy consumption requirements therefore hinder the state-of-the art, and more convenient solutions are needed.SUMMARY OF THE INVENTION
[0003] The invention is defined by the features of the independent claims. Some specific embodiments are defined in the dependent claims.
[0004] According to a first aspect of the present disclosure, there is provided a physical artificial neuron comprising a polycrystalline memristive layer configured to, in response to applied energy, integrate and leak.
[0005] According to a second aspect of the present disclosure, there is provided a physical neuromorphic chip comprising at least one physical artificial neuron comprising apolycrystalline memristive layer; and at least one physical artificial synapse comprising a memristive layer.
[0006] According to a third aspect of the present disclosure, there is provided a physical neuromorphic chip comprising at least one physical artificial neuron comprising a polycrystalline memristive layer configured to, in response to applied energy, integrate and leak; and at least one physical artificial synapse comprising a memristive layer.
[0007] According to a fourth aspect of the present disclosure, there is provided a method of manufacturing a neuromorphic chip, the method comprising: obtaining a substrate comprising a first region supporting polycrystalline growth of a first memristive material; and growing on said first region of the substrate a polycrystalline memristive layer of the first memristive material.
[0008] Embodiments of the present disclosure provide significant advantages. For example, a physical artificial neuron according to at least some embodiments may be utilized in, or be part of, a neuromorphic chip, neuromorphic computer, artificial intelligence (Al) chip, or Al accelerator, or part thereof, for example. At least some embodiments provide the benefit of providing a physical artificial neuron that is easily manufacturable, for example. As the (polycrystalline) memristive material may form a physical artificial neuron, the structure as well as manufacture of such a neuron is simplified. In at least some embodiments, at least one physical artificial synapse comprises memristive material comprising the same chemical formula as memristive material comprised in at least one physical artificial neuron. Such may be beneficial in that same material formulation may be used in manufacture of synapse(s) and neuron(s), for example. Such may provide simultaneous manufacture of synapses and neurons. Another benefit of at least some embodiments, is that different neuronal and / or synaptic properties and characteristics of a memristive layer may be obtained by providing different crystallinity (i.e., polycrystallinity or monocrystallinity, and / or degree thereof), for the memristive layer, for example, for a GCMO layer.BRIEF DESCRIPTION OF THE DRAWINGS
[0009] FIG 1 A, IB and 1C illustrate at least some example method steps according to at least some embodiments;
[0010] FIG. 2A and 2B illustrate example experimental data according to at least some embodiments;
[0011] FIG 3A, 3B, 3C and 3D illustrate at least some example method steps according to at least some embodiments;
[0012] FIG. 4A, 4B and 4C illustrate some example embodiments;
[0013] FIG. 5 illustrates an apparatus according to at least some embodiments; and
[0014] FIG. 6, 7A, 7B, 8A, 8B, 9A and 9B illustrate example measurement data according to at least some embodiments.EMBODIMENTS
[0015] In accordance with the present disclosure, there is provided a physical artificial neuron comprising a polycrystalline memristive layer. A polycrystalline memristive material may act as an integrate-and-leak component of a physical artificial neuron. One benefit of at least some embodiments may be that the structure of a physical artificial neuron is simplified as the polycrystalline memristive material (or layer) is configured to integrate-and-leak. In at least some embodiments, the polycrystalline memristive material is Gdi-xCaxMnO3 (GCMO), wherein x is from 0.6 to 0.99. In at least some embodiments, the physical artificial neuron is made of polycrystalline memristive material, such as GCMO. A physical artificial neuron according to at least some embodiments may be part of a neuromorphic chip. A neuromorphic chip according to an embodiment may comprise a layer of polycrystalline memristive material as a physical artificial neuron. Additionally, the neuromorphic chip may comprise a layer of monocrystalline memristive material as a physical artificial synapse. The polycrystalline and monocrystalline memristive material may comprise the same formula, such as both be GCMO. For example, both the polycrystalline memristive material and the monocrystalline memristive material may be GCMO.
[0016] Memristor (short from “memory resistor”) refers to a type of electrical component that regulates the flow of current in an electronic circuit and “remembers” the amount of charge that has passed through it. Memristive refers to a property capable of being a memristor. The expression "memristive material" and the like, is used here for the purpose of unambiguously referring to the material that is comprised in a neuron and / or a synapse, although it is the current understanding that the memristor-type effects (includinghigh and low resistance states) may actually arise in the junction or interface between a memristor material and another material (also referred to as a contact), and not in the bulk of the memristive material itself. Electrically conductive connections (i.e., contacts) may be used to subject the piece of memristive material to desired voltages and to, for example, measure the resulting electric characteristics of the piece of memristive material or a memristive layer. A memristive material may thus be in contact with another material, said another material comprising an oxidizable metal, such as aluminium. A neuron according to an embodiment may comprise a polycrystalline memristive material in contact with an oxidizable metal. A synapse according to an embodiment may comprise a monocrystalline memristive material in contact with an oxidizable metal.
[0017] The (polycrystalline and / or monocrystalline) memristive material with a formula Gdi-xCaxMnO3 may be produced on a substrate surface as a film or a layer using, for example, chemical solution deposition CSD, metalorganic chemical vapor deposition MOCVD, metalorganic deposition MOD or pulsed laser deposition, PLD.
[0018] The term “polycrystalline” refers to a property of a solid structure having a plurality of grains. Polycrystalline materials may have high-angle grain boundaries. In an embodiment, a high-angle of a grain boundary of a polycrystalline memristive layer material is above 5 degrees, such as 10 degrees, for example 15 degrees. A poly crystalline structure may also be referred to as a non-epitaxial structure. The material may be grown on a lattice structure having a lattice mismatch from 5% to 20% thereby providing a polycrystalline material, for example. The polycrystallinity may be deduced from X-ray diffraction, XRD, measurements, wherein peaks corresponding to several lattice directions are observable in theta-2theta measurement. In addition, in a texture measurement (chi-phi) at single 2theta value distinct peaks may be absent. Alternatively, several peaks corresponding to several crystal orientations may be observable. Alternatively, continuous or near-continuous peaks may be observed (e.g., fiber texture). The peaks in a theta-2theta measurement may be minimal, reduced or non-existent, therefore indicating lack of a single dominant grain orientation.
[0019] Monocrystalline (also known as single crystalline) refers to a property of a solid structure having an ordered structure and minimal or non-existent number of grain boundaries, therefore forming a substantially singular continuous and ordered structure. The monocrystallinity may be deduced from XRD measurements, wherein pronounced peak(s)of single lattice direction may be observable in a theta-2theta measurement. In addition, in a texture measurement at single 2theta value, peaks may be observable corresponding to crystal symmetry.
[0020] Monocrystalline structures may have low-angle grain boundaries (i.e., substantially monocrystalline structure). Low-angle grain boundaries may refer to angles below 15 degrees, such as angles below 10 degrees, for example, angles below 5 degrees, or below 3 degrees. A substantially monocrystalline structure comprises a single grain although in at least some embodiments a substantially monocrystalline structure may have more than one grain which, however, differ from polycrystalline grains in terms of size and / or number in the material.
[0021] A physical artificial synapse (or herein simply “synapse”) refers to an electronic component. Synapse may be configured to remember its state such that it is readable later on. Especially beneficial are synapses that do not require and external power source to maintain the state (and thus information). Thus, memristive materials and / or layers may be suitable for such applications. A synapse may be obtained from a memristive material having a (substantially) monocrystalline structure, for example. In an embodiment the physical artificial synapse comprises monocrystalline memristive layer and an oxidizable metal, wherein the monocrystalline memristive layer is in contact with the oxidizable metal.
[0022] A physical artificial neuron (or herein simply “neuron”) refers to an electronic component. A neuron may be described with an “integrate-and-fire” model. A neuron in accordance with the integrate and fire model is configured to integrate input signals in a time window, and if the accumulation of input signals exceeds a threshold, the neuron according to the integrate-and-fire model is configured to be capable of providing an output. For example, after integration of signals is above a threshold, the neuron may pass through signals (“fire”), for example. The time window may be 1 second, or 500 milliseconds, for example. In an embodiment, the physical artificial neuron comprises polycrystalline memristive layer and an oxidizable metal, wherein the polycrystalline memristive layer is in contact with the oxidizable metal.
[0023] A physical artificial neuron and / or a physical artificial synapse may have contact terminals such that said physical artificial neuron and / or a physical artificial synapse may be electrically connected to additional and / or external electronics, such as a device capable of producing a voltage, such as a voltage source. One contact terminal of said contactterminals may comprise an oxidizable metal, such as aluminium, and another contact terminal of said contact terminals may comprise non-oxidizable metal, such as gold.
[0024] For a neuron to be functional, the neuron should return, or leak, to or towards a base level (i.e., a high resistance state) it had prior to the input signals (before the time window). In other words, the previous input signal(s) and integration thereof, should not interfere with the upcoming input signal(s) of another time window. A neuron may also be configured to leak towards, the HRS. Therefore, integrate and leak refers to a property of a material in which said material is configured to integrate input signal(s) at a time window, and gradually leak thereby reverting back to the original (resistance) state in the absence of such input signal(s). A leak, that is memristive layer leaking towards the HRS, may be deemed to have substantially occurred if more than 60%, or more than 65%, more than 70%, or more than 75%, such as more than 80% or more than 85%, for example more than 90% or more than 95% of the resistance of the HRS is retained.
[0025] A leaky-integrate function is beneficial at least in that signal (inputs) are not propagated in insufficient amounts but only above a certain threshold or intensity. Moreover, due to leak, another set of signals may be obtained without being affected by a previous set of input signals. Therefore signal(s) of certain intensity or frequency pass through such a material. Such material may be termed leaky integrator.
[0026] In an embodiment, the physical artificial neuron is configured to leak to the HRS in less than a second, or less than 500 milliseconds in the absence of an input. In an embodiment, the polycrystalline memristive layer of a physical artificial neuron is configured to leak towards the HRS such that a threshold is reached in less than a second, or less than 500 milliseconds in the absence of an input. The threshold may be 80% or 90% of the HRS, for example. The duration for leaking (i.e., reaching substantially to the base state) may be less than 500 milliseconds, for example.
[0027] In an embodiment, the polycrystalline memristive layer has a first lower resistance state, and the polycrystalline memristive layer is configured to integrate and leak by: reaching the first lower resistance state from the HRS in response to the applied energy; and leaking towards (or to) the HRS from the first lower resistance state in absence of the applied energy. The first lower resistance state may be lower than the HRS and separated from the HRS.
[0028] In an embodiment, the polycrystalline memristive layer has a plurality of lower resistance states comprising the first lower resistance state and a second lower resistance state having a lower electrical resistance than the first lower resistance state. In a further embodiment, the polycrystalline memristive layer is further configured to integrate and leak by: reaching the second lower resistance state from the first lower resistance state in response to the applied energy; and leaking to the HRS from the second resistance state in absence of the applied energy.
[0029] Lattice mismatch (f) refers to a condition where two lattice structures do not maintain the same or similar lattice structure or parameters across an interface between the two structures. As materials have intrinsic lattice parameters and dimensions, growing a first material on a second material having a lattice mismatch may support polycrystalline growth. Similar or same lattice structures (i.e., a lattice match) may support (substantially) monocrystalline growth, and / or epitaxial growth. It is noted, that if the “lattice mismatch” is a multiple of the lattice parameters between to unit cells, monocrystalline growth may be supported between such structures.
[0030] Therefore, in at least some embodiments, polycrystalline memristive layer / material may be obtained via such lattice mismatch of a substrate surface, and the memristive material in question.
[0031] In a lattice mismatch, the two structures have a substantial difference in lattice dimensions at the interface of the two structures. In other words, the lattice mismatch is between the surface of the substrate (or a region at the substrate) and the memristive layer. Such a lattice mismatch may support polycrystalline (or non-monocrystalline or non-epitaxial) growth on the interface of the materials.
[0032] Lattice mismatch is a ratio which may be obtained by f =as aBaB, wherein as is the substrate in plane lattice parameter and an the lattice dimension of a memristive material. The lattice mismatch equation may be modified to accommodate, for example, diagonal crystal growth.
[0033] In an embodiment, a surface supports polycrystalline growth if there is a lattice mismatch of more than 5% with respect to the memristive layer. In an embodiment a surface supports polycrystalline growth if there is a lattice mismatch of more than 6%, more than 7%, more than 8%, more than 9% or more than 10% with respect to the memristive layer. Inan embodiment, a surface supports polycrystalline growth if there is a lattice mismatch of more than 8% or more than 12% with respect to the memristive layer. The lattice mismatch may be less than 20% to support polycrystalline growth. Lattice mismatch from 20% to 80% may induce amorphous growth, or lack of memristive layer growth. A substantially polycrystalline layer of memristive material may be obtained on a surface having suitable lattice parameters thereon. In an embodiment, a surface (of the substrate) may have a lattice mismatch from 5% to 20%, such as 5% to 15%, for example, from 5% to 12% or from 6% to 11% with respect to the memristive material, in order to support polycrystalline growth of said memristive material. A polycrystalline memristive layer may therefore be obtained on such a surface. A substrate or a region supporting polycrystalline growth may comprise magnesium oxide (MgO), at least for a GCMO memristive layer. A substrate or a region thereon, supporting polycrystalline growth may be an electrical insulator. A substrate or a region thereon, supporting polycrystalline growth may be non-reactive with respect to the memristive material.
[0034] A surface supports monocrystalline growth if there is a lattice mismatch of less than 5%, such as less than 4% or less than 3%, for example less than 2%. In an embodiment, surface supports monocrystalline growth if there is a lattice mismatch of less than 3%. Thus, a substantially monocrystalline layer of memristive material may be obtained on such a surface. A lattice mismatch may be supported on surfaces being (integer) multiples of one another. For example, a lattice mismatch of 100% may refer to a multiple (i.e., two-fold difference) of the lattice dimensions, and thus, monocrystalline growth may be supported between such surfaces. A substrate or a region therein supporting monocrystalline growth may comprise strontium titanate (SrTiCL), lanthanum aluminium oxide (LaAlOs, LAO), strontium lanthanum aluminium oxide (SrLaAlO4, SLAO), lanthanum aluminate - strontium aluminium tantalate (LSAT), at least for a GCMO memristive layer, for example. For example, a substrate may be silicon (Si) coated with strontium titanate, STO. A substrate or a region thereon, supporting polycrystalline or monocrystalline growth may be an electrical insulator. A substrate or a region thereon, supporting polycrystalline or monocrystalline growth may be non-reactive with respect to the memristive material.
[0035] The substrate may be further provided with region providing a suitable support for monocrystalline growth (i.e., a synapse) or polycrystalline growth (i.e. a neuron). For example, such a region may be obtained by depositing. The region may have a suitable lattice mismatch in order to support polycrystalline growth or monocrystalline growth.
[0036] In accordance with at least some embodiments, there is provided a physical artificial neuron comprising a polycrystalline memristive material. The polycrystalline memristive material may be configured to integrate and leak. In an embodiment, the polycrystalline memristive material is, or comprises, GCMO.
[0037] A polycrystalline material (and neuron comprising said material) may be configured to integrate and leak. In other words, the polycrystalline memristive material may have a high resistance state, HRS, and at least one lower resistance state(s) (LRSs) having lower resistance(s) than the resistance of the HRS. There may be a plurality of lower resistance states (LRSs) having a lower resistance than the HRS. The plurality of LRSs are different from one another in terms of resistance. The HRS and LRSs may be separated by a (resistance) gap. The resistance(s) at the LRS(s) may be less than 50% of the HRS, such as less than 20% or 10% of the HRS. The resistance(s) at the LRS(s) may be from 1% to 60%, such as from 5% to 50% or from 6% to 45%, for example, from 10% to 40%, from 11% to 35%, or from 15% to 30% of the HRS.
[0038] Resistance states, such as a high resistance state, HRS, or a low resistance state, LRS, refer to resistance values relative to one another. Such resistance states may have different numeric values, depending, for example, the size and dimensions of the object comprising the memristive material. Moreover, a memristive material may comprise an HRS and a plurality of LRSs. Such may be beneficial in that integration of energy, such as voltage pulse(s), may provide for a better neuron if the neuron comprises more than one LRSs and the HRS. The HRS and LRS(s) may be separated by a (resistance) gap, that is, a resistance value range. The plurality of LRS(s) may be separated by (resistance) gap(s) from one another.
[0039] In an embodiment, the memristive material, such as a polycrystalline memristive material / layer or a (substantially) monocrystalline memristive material / layer is GCMO. GCMO refers memristive material(s) having a formula Gdi-xCaxMnO3 The x in the formula may be from 0.55 to 0.99, such as from 0.65 to 0.95, for example, from 0.7 to 0.9. In accordance with the present disclosure, there is provided a physical neuromorphic chip comprising a physical artificial neuron. The neuron comprises a polycrystalline memristive layer. The physical neuromorphic chip may comprise one or more neurons. The physical neuromorphic chip may further comprise one or more synapses. Such a synapse may be a memristive layer, or a memristor. The synapse may be a filamentary memristor, an interface-type memristor, a phase change memory PCM type memristor, a spin torque transfer magnetic random access memory STT-MRAM type memristor, a resistive switching random access memory ReRAM / RRAM type memristor, a CMOS implemented memristor, a hybrid CMOS-based memristor, a metal-oxide memristor, a semiconductor-based memristor or a 2D-material-based, like graphene-oxide, memristor, for example. In an embodiment, at least one synapse comprises a substantially monocrystalline memristive material. Such a substantially monocrystalline memristive material may be GCMO. The monocrystallinity of the memristive material may provide memristive properties to the material. Polycrystallinity of the memristive material may provide integrate-and-leak properties to the material.
[0040] In an embodiment, the physical neuromorphic chip comprises at least one physical artificial neuron comprising a polycrystalline memristive layer; and at least one physical artificial synapse comprising a memristive layer, wherein the memristive layer of the at least one physical artificial synapse and / or the memristive layer of the at least one physical artificial neuron comprises memristive material having a formula Gdi-xCaxMnO3 wherein x is from 0.55 to 0.99. In a further embodiment, the at least one physical artificial neuron and the at least one physical artificial synapse comprise the same material, for example GCMO. The memristive layer of the at least one physical artificial synapse may be a substantially monocrystalline memristive layer, such as a substantially monocrystalline GCMO layer. The memristive layer of the at least one physical artificial neuron may be a substantially polycrystalline memristive layer, such as a polycrystalline GCMO layer.
[0041] In accordance with the present disclosure, there is provided a method of manufacturing a neuromorphic chip, the method comprising: obtaining a substrate comprising a first region supporting polycrystalline growth of a first memristive material; and growing on said first region of the substrate a polycrystalline memristive layer of the first memristive material. The polycrystalline growth may be obtained via a deposition technique for example. Surface of the first region may have a lattice mismatch from 5 % to 20% with respect to the lattice parameters of the memristive layer. Such a region may support polycrystalline growth of the memristive material, such as GCMO. A physical artificial synapse may be obtained on a second region of the substrate. Such a synapse may be a memristive layer. The memristive layer of the synapse may be a substantially monocrystalline memristive layer. The monocrystalline memristive layer may be a monocrystalline GCMO layer. For at least GCMO, the second region may be, for example, LSAT, LAO, SLAO and / or STO. At least one benefit of manufacturing a neuromorphic chipaccording to at least some embodiments, is that the physical artificial neuron(s) and synapse(s) may be manufactured simultaneously. For example, such neuron(s) and synapse(s) may be grown simultaneously on different regions of a same substrate. For example, GCMO may be grown as a polycrystalline and monocrystalline on a substrate such that at least one physical artificial neuron and at least one physical artificial synapse are formed. The regions may be such that they respectively support polycrystalline and monocrystalline growth, for example by having a suitable lattice mismatch.
[0042] FIG. 1A, IB, and 1C illustrate example method steps for obtaining a physical artificial neuron and a physical artificial synapse in accordance with at least some embodiments of the present disclosure. The physical artificial neuron and the physical artificial synapse may be comprised in a physical neuromorphic chip, for example, an Al chip, an Al accelerator, an integrated chip, an embedded chip, or a neuromorphic computer or part thereof. A neuron according to at least some embodiments, may be a separate component or device.
[0043] FIG. 1A illustrates a substrate 150. The substrate 150 or a region therein supporting monocrystalline growth may comprise STO, LAO, LSAT and / or SLAO. The substrate 150 or a region supporting poly crystalline growth may comprise MgO. For example, MgO may be suitable for supporting polycrystalline growth of a GCMO formulation, as MgO may have a lattice mismatch at its surface (or interface) of approximately from 8% to 10 % with respect to the GCMO formulation. A substrate (or a region thereof) on to which a (polycrystalline or monocrystalline) memristive layer or material is grown, should be an electrical insulator and non-reactive with the memristive material. In other words, band gap of such a substrate may be more than 2 eV, electron volts, for example.
[0044] FIG. IB illustrates the substrate 150 of FIG. 1A, having thereon regions 111 and 121 shown as dashed lines. The substrate 150 may thus, be further provided with a region 121 providing a suitable to support monocrystalline growth (i.e., a synapse) and / or a region 111 suitable to support polycrystalline growth (i.e., a neuron). For example, such a region 111 and / or 121 may be obtained by depositing. The region 111 is a region capable of supporting polycrystalline growth of a memristive material. For example, the region may have (at the surface) a lattice mismatch with respect to the memristive material. The region 121 is a region capable of supporting monocrystalline growth of a memristive material. Theregions may be obtained on the surface of the substrate 150, via different techniques, such as via depositing. For example, region 111 or 121 may be deposited on the substrate while the other region 121 or 111 is the surface of the substrate. The regions 111 and 121 may both be deposited on the substrate 150.
[0045] A memristive layer, such as a polycrystalline memristive layer or a (substantially) monocrystalline memristive layer, may be provided on to a substrate or region thereon, for example, by depositing a layer of memristive material, such as a layer of GCMO, onto the substrate using deposition, such as chemical vapor deposition, CVD, atomic layer deposition, ALD, molecular beam epitaxy, MBE, physical vapor deposition, PVD, chemical solution deposition, CSD, metalorganic chemical vapor deposition, MOCVD, metalorganic deposition, MOD or pulsed laser deposition, PLD, and the like. Ultra-high vacuum, UHV, conditions may be used in at least some embodiments.
[0046] In order to obtain a polycrystalline memristive layer, the substrate (or a region thereon) may have a lattice mismatch from 4% to 30%, such as from 5% to 20%, for example, from 8% to 15% or from 9% to 13%, with the unit cell of the memristive material in question. In order to obtain a monocrystalline memristive layer, the substrate (or a region thereon) may have a lattice mismatch of less than 10%, such as less than 5%, for example less than 3% with the unit cell of the memristive material in question. For a monocrystalline (or substantially monocrystalline) memristive layer, such a layer may be epitaxially grown on a surface.
[0047] The layer of memristive material, such as a polycrystalline memristive layer and / or a substantially monocrystalline memristive layer, may be patterned. Patterning may be done using, for example, wet chemical etching or dry etching, such as reactive ion etching, ion beam etching or argon gas etching. The patterning may be done before depositing, for example, using a mask.
[0048] FIG. 1C illustrates two memristive materials or layers 110 and 120 on the substrate 150. In FIG. 1C, the memristive layers 110 and 120 are in contact with one another. Alternatively, the memristive layers may be connected to one another via a non-oxidizing electrical conductor, such as a noble metal, for example gold. The formula of the memristive materials 110 and 120 may be different from one another. The formula of the memristive materials 110 and 120 may be same. In an embodiment, the memristive materials 110 and 120 is GCMO such that memristive material 110 is polycrystalline GCMO and memristivematerial 120 is (substantially) monocrystalline GCMO. A benefit of having same formula for both neuron(s) and synapse(s), is at least that material needs are reduced, and said neuron(s) and synapse(s) may be obtained, such as deposited or grown, simultaneously.
[0049] One important characteristic for a memristor, and a physical artificial synapse is the ability to retain its resistance state (i.e., retain information). Such may be achieved by a substantially monocrystalline order of memristive material. Surprisingly, it has been observed that a memristive material may also be useable as a physical artificial neuron comprising a polycrystalline memristive material therein instead. Such structure and order may be beneficial for a neuron. Such a neuron may integrate and leak. In other words, a polycrystalline memristive layer may be configured to alter its resistance state (from a high resistance state, HRS, to a lower resistance state, LRS) in a response to applied energy but also return to, or leak towards, the original resistance state (HRS) in absence of such applied energy. Such a polycrystalline memristive layer may be a polycrystalline GCMO layer.
[0050] FIG. 2A illustrates example resistance measurements as a function of number of voltage pulses and as a function of time after application of voltage pulses. The measurements are obtained from a physical artificial neuron having a polycrystalline GCMO memristive layer comprised therein. In FIG. 2A, two subplots are shown with logarithmic scales for pulse duration (left subplot) and time (right subplot) with respect to measured resistance values. The subplots therefore represent integrate-and-leak functionality of a memristive material.
[0051] Referring to FIG. 2A, a polycrystalline memristive material may be driven to different resistance states via applied energy. Such energy may be applied via a voltage pulse or plurality thereof, such as a sequence of voltage pulses. In FIG. 2A, different number of pulses are shown together with the corresponding resistance value. An individual pulse of FIG. 2A has a duration of 2 ms, milliseconds, and an amplitude of 4 V, volts. As can be appreciated, the memristive material is driven from a high resistance state, HRS, to a lower resistance state by applying voltage pulses. Moreover, a lower resistance state is obtained as a function of the number of pulses (for example, from 10 pulses to 100 pulses in left subplot), i.e., applied energy to the system. Therefore, a memristive material having a polycrystalline structure, is suitable to integrate energy applied thereto, such as via voltage pulses.
[0052] Moreover in FIG. 2A, a polycrystalline material is also configured to return or leak back to (or at least towards) the high resistance state, HRS, from the lower resistancestate(s). This is exemplified in the right subplot of FIG. 2A as resistance values increase as a function of time. For example, resistance is 60 MOhms at start (i.e., at t=Os, LRS) while the resistance is of the order of 110 MOhms at the end of the measurement (i.e., at t=ls, HRS) Therefore, a polycrystalline memristive material having a polycrystalline structure, is suitable to leak back to, or towards, a higher resistance state, HRS of the neuron.
[0053] FIG. 2B is provided depicting experimental data of integrate-and-leak, in order to further highlight the integrate-and-leak functionality of at least some embodiments of the present disclosure. The example experimental data is obtained with a GCMO memristive material having a polycrystalline structure. The polycrystalline structure was grown on a region of MgO. In FIG. 2B, a high resistance state, HRS, is observable between time points 0 and 2 seconds (near 1 x 109Ohms). When energy is applied to the poly crystalline memristive material, said material will transition to a lower resistance state (near 1 x 107Ohms), illustrated with LRS between time points 1 and 3 seconds. Such transition may be known as “integration” in terms of an artificial neuron model. As can further be seen from FIG. 2B, the resistance state of the polycrystalline material may begin to return to, revert to, or move towards, the HRS once the application of energy is absent. Such may be termed as “leaking” and is shown from 3 seconds to 14 seconds in FIG. 2B. The memristive material may be considered to have returned to, or (substantially) leaked back to the HRS if it has a resistance value of 75 % or more, such as 85% or more, of the HRS, for example.
[0054] An example integration may be done as follows. Energy is applied to a polycrystalline memristive layer via a sequence of voltage pulses. To reach a lower resistance state, the number of voltage pulses in a sequence of voltage pulses may be more than 10, such as more than 20, for example. The sequence of voltage pulses may comprise a voltage pulse, or a plurality thereof. Each pulse may have a duration from 1 ps to 5 ms. In an embodiment, a pulse duration of at least one pulse may be from 1 ps to 5ms, such as, from 2 ps to 4 ms, from 500 ps to 3 ms, for example, from 600 ps to 2 ms. Each pulse may have a voltage amplitude from 1 V to 5 V. The inter-pulse spacing between two consecutive voltage pulses of the sequence of voltage pulses may be from 1 ms to 4 ms.
[0055] Typically, memristive materials (and memristors) may be viewed as rather ill-fitted to artificial neuronal function. The present disclosure provides embodiments wherein such materials may be utilized as physical artificial neurons. Because a memristive material may be used as a physical artificial neuron, in at least some embodiments the neuron(s) andthe synapse(s) may be manufactured from the same memristive material albeit having different crystal structures. In other words, the memristive material may have the same chemical formula (or substantially the same chemical formula), but the size and number of grains of the crystal structure of the neuron(s) and the synapse(s) may be different. That is, a memristive layer of a neuron may be polycrystalline and a memristive layer of a synapse may be (substantially) monocrystalline. Such designs provide simplified manufacturing schemes as well as overall reduced material needs, for example. Moreover, in at least such embodiments, the memristive materials may be grown on the corresponding regions simultaneously. For example, GCMO may be grown simultaneously thereby providing at least one neuron and at least one synapse comprising a polycrystalline GCMO structure and a (substantially) monocrystalline GCMO structure, respectively.
[0056] At least one neuron and at least one synapse may form a physical neuromorphic chip. The physical neuromorphic chip may be a physical neuromorphic computer or part thereof.
[0057] FIG. 3A, 3B, 3C and 3D illustrate example steps for obtaining a physical neuromorphic chip in accordance with at least some embodiments of the present disclosure.
[0058] In FIG. 3 A a substrate 350 is obtained. The substrate 350 may have different regions having different lattice structures, for example. For example, certain regions may support polycrystalline growth of a memristive material. Certain regions may support monocrystalline growth of a memristive material.
[0059] In example of FIG. 3B, two regions 311 and 321 are provided on the surface of the substrate 350. The first region 311 may be configured to support poly crystalline growth. The second region 321 may be configured to support monocrystalline growth. The polycrystalline growth and the monocrystalline growth may be supported via a suitable lattice mismatch with respect to the memristive material or layer in question, for example.
[0060] One or more memristive layers may be obtained on the substrate. In FIG. 3C, memristive layers 310 and 320 are obtained on regions 311 and 321, respectively. The memristive layers 310 and 320 may be same material, such as GCMO. The memristive layer 310 comprises a poly crystalline structure of a memristive material. The memristive layer 320 comprises a substantially monocrystalline structure of a memristive material. Thememristive layer 310 is thus, comprised in a neuron, and the memristive layer 320 is thus, comprised a synapse.
[0061] A neuron and synapse, and / or the memristive layers therein, may be attached, that is, be in contact, with one another. However, the neuron and synapse may be connected via an electrical conductor. In FIG. 3D, a physical neuromorphic chip 300 is obtained by connecting at least one neuron and at least one synapse to one another. In FIG. 3D an electrical conductor 330 is obtained between the polycrystalline memristive layer 310 (i.e., a neuron) and the substantially monocrystalline memristive layer 320 (i.e., a synapse). The electrical conductor may be a non-oxidizing metal, such as a noble metal, for example gold.
[0062] FIG. 4A, 4B and 4C illustrate example parts of a neuromorphic chip 400A, 400B and 400C, respectively. FIG. 4A to 4C illustrate physical artificial neuron 410 comprising a poly crystalline memristive material 415 and physical artificial synapse(s) 420 comprising a (substantially) monocrystalline memristive material 425 according to at least some embodiments. Dashed regions 415,425 illustrating the crystallinity of the memristive layers of the neuron 410 and synapses 420 are not necessarily to scale. For simplicity, substrate, region(s), neuron and synapses are presented as rectangles although it may be appreciated that other geometries may be applicable.
[0063] FIG. 4A illustrates physical artificial neuron 410 and two physical artificial synapses 420 according to at least some embodiments. In FIG. 4A, a neuron 410 comprising memristive material 415 is illustrated. The neuron 410 is situated on region 451A of a substrate 450A having a surface capable supporting polycrystalline growth of the memristive material 415. For example, for a GCMO material, such a region may comprise a layer of MgO. In FIG. 4A, the substrate 450A comprises a surface capable of supporting monocrystalline growth of the memristive material. The synapses 420 are situated on the substrate capable supporting monocrystalline growth of the memristive material 425. For example, for a GCMO memristive material, such a region supporting (substantially) monocrystalline growth may comprise strontium titanate, STO. The synapses 420 may thus, be capable of retaining a resistance state in absence of applied energy, such as via an external voltage.
[0064] FIG. 4B illustrates physical artificial neuron 410 and two physical artificial synapses 420 according to at least some embodiments. In FIG. 4B, a neuron 410 comprising memristive material 415 is illustrated. In FIG. 4A, the substrate 450A comprises a surfacecapable of supporting poly crystalline growth of the memristive material 415. The neuron 410 is thus, situated on the substrate capable supporting poly crystalline growth of the memristive material 425. For example, for a GCMO material, such a region may comprise a magnesium oxide, MgO. The synapses 420 are situated on region(s) 452B of the substrate 450B having a surface 452B capable supporting monocrystalline growth of the memristive material 425. For example, for a GCMO material, such a region supporting (substantially) monocrystalline growth may comprise strontium titanate, STO.
[0065] FIG. 4C illustrates a physical artificial neuron 410 and two physical artificial synapses 420 according to at least some embodiments. In FIG. 4C, the neuron 410 is located on a region 451C comprising a surface capable of supporting poly crystalline growth of the memristive material comprised in the neuron 410. The synapses 420 are located on region(s) 452C comprising a surface capable of supporting monocrystalline growth of the memristive material comprised in the synapses 420.
[0066] At least one physical artificial synapse and at least one physical artificial neuron may be connected to one another. For example, the at least one physical artificial synapse and at least one physical artificial neuron may be in contact with one another. Such may be possible, for example, when the regions (capable of supporting monocrystalline and polycrystalline growth of one or more memristive materials) are attached to one another. The at least one physical artificial synapse and at least one physical artificial neuron may be connected to one another via an electrical conductor. For example, the at least one physical artificial synapse 420 and at least one physical artificial neuron 410 of FIG. 4 A, 4B or 4C may be connected via a non-oxidizing metal, such as a noble metal, such as gold. The neuron and synapse may be in contact with one another (not shown in FIG. 4A, 4B or 4C).
[0067] A physical artificial neuron and / or a physical artificial synapse may comprise at least one contact pad, wire or terminal for additional and / or external electronics. Such contact pads, wires or terminals may comprise gold and / or silver. Contact terminal for a neuron may be a contact pad. A contact pad may be deposited using, for example, chemical vapor deposition, CVD, physical vapor deposition, PVD, sputtering, pulsed laser deposition PLD, electron beam deposition, thermal evaporation, molecular beam epitaxy, MBE, or atomic layer deposition, ALD. A deposited contact pad may be patterned using for example, wet chemical etching or dry etching, such as reactive ion etching, ion beam etching or argongas etching. Alternatively, the contact terminals may be bond wire or contact pins, which may be attached to the physical artificial neuron and / or the physical artificial synapse.
[0068] FIG. 5 illustrates a schematic of a physical artificial neuron 500 comprising a polycrystalline memristive layer 510 on a substrate 550. The physical artificial neuron 500 also comprises a first terminal 511 and a second terminal 512 on the polycrystalline memristive layer 510. The first terminal 511 may be a non-oxidizing metal, such as a noble metal, for example gold. The first terminal may comprise at least one of gold or silver. The second terminal 512 may be an oxidizing metal, such as aluminium. The second terminal may be at least one of copper, molybdenum, tungsten, indium, aluminium or titanium. An ohmic contact is formed between the first terminal 511 and the second terminal 512 as well as between the first terminal 511 and the memristive layer 510. For example, an ohmic contact is formed between an aluminium, Al, wire and a gold, Au, pad and between Au and a memristive material, such as GCMO. The example schematic of FIG. 5 is also applicable for embodiments, wherein a monocrystalline memristive layer is used (i.e., a synapse).
[0069] FIG. 6 illustrates an example X-ray diffraction, XRD, measurement of GCMO grown on STO (shown as a dark line) and MgO (shown as a gray line). As can be appreciated, STO provides an ordered, epitaxial structure for the GCMO. In the dark line of FIG. 6, peaks of GCMO near the peaks originating from the STO are observable. In other words, ordered peaks of STO are accompanied by (albeit of smaller intensity) peaks of GCMO approximately at least at points (of 29) 25, 50, and 110 degrees.
[0070] Referring to FIG. 6, for MgO as a substrate more polycrystalline structure is observable for the GCMO. Conversely to the STO substrate, for MgO as a substrate, no clear peaks are observable for GCMO therefore indicating a less-oriented (polycrystalline) structure. Therefore, for a polycrystalline memristive layer, the grains therein are randomly oriented, whereas for a substantially monocrystalline memristive layer orientation is more uniform. XRD may thus be used to differentiate polycrystalline and monocrystalline structures.
[0071] In at least some embodiments, a benefit is that a neuron may be manufactured from a same material as a synapse. Another benefit may be that at least one physical artificial synapse and at least one physical artificial neuron may be grown simultaneously. Moreover, in at least some embodiments, the at least one physical artificial synapse and at least one physical artificial neuron comprise the same memristive material (such as GCMO).Therefore, a physical neuromorphic chip according to at least some embodiments of the present disclosure may be obtained with a single material for neurons and synapses therein. Such provides at least the benefit that manufacture, as well as material needs are reduced.
[0072] FIG. 7A, 7B, 8A and 8B provide texture analyses (2theta-phi and phi -chi plots) for polycrystalline and monocrystalline memristive materials, in this case GCMO. The plots are obtained from XRD measurements. The texture analysis illustrates a (possibly) preferred orientation in crystalline materials and films. “2Theta” illustrates interplanar spacing of atoms in a (possible) lattice structure. “Phi” illustrates in-plane rotational orientation of the sample in question. Intensity is shown as a colorized scale. In FIG. 7B and 8B, the radius of the polar coordinate represents chi-angle and the rotational angle of the polar coordinates represents the Phi-angle.
[0073] FIG. 7A and 7B illustrate experimental data obtained for a polycrystalline GCMO grown on MgO surface having a lattice mismatch (f) of approximately 9.847% with respect to GCMO. At least in part because of the lattice mismatch, GCMO may be grown as a polycrystalline structure. As can be appreciated, no discernible intensity peaks are observable in FIG. 7A, and thus the texture has no clear preference in orientation (i.e., single crystallinity that would prevail over others). In FIG. 7B, only minute peaks are observable. The observed peaks in FIG. 7B correspond to (202) peaks of MGO and corresponding symmetric rotational orientation. Therefore, it may be deduced that GCMO is in polycrystalline form, lacking clear crystalline preference in terms of orientation, for example.
[0074] FIG. 8A and 8B illustrate experimental data obtained for a monocrystalline GCMO grown on STO surface having a lattice mismatch (f) of approximately 2.350% with respect to GCMO. At least in part because of the lattice match (i.e., a small lattice mismatch), GCMO may be grown as a monocyrstalline structure. As can be appreciated, increased intensity from 2theta peaks is observable in FIG. 8A. Moreover, a symmetric set of peaks is observable in FIG. 8B highlighting rotational (phi) symmetry of the monocrystalline structure. Therefore, it may be deduced that GCMO is in monocrystalline form.
[0075] In the following FIG. 9A and 9B, example experimental observations are provided. Regarding the examples of FIG. 9A and 9B, films (or layers) of GCMO were deposited on SrTiOs and MgO single crystal substrates using pulsed laser deposition, PLD.
[0076] FIG. 9A illustrates resistance - voltage measurement results for GCMO grown on (a) strontium titanate, STO. and (b) GCMO on magnesium oxide, MgO. Retention results immediately after SET / RESET are shown in (c) and (d) for GCMO grown on STO and MgO (shown as dots and crosses). There, a decay of low resistance state, LRS, towards stable high resistance state, HRS, is demonstrated for (d) GCMO on MgO. On the other hand, as may be appreciated in FIG 9A (c), GCMO on STO maintains a distinguishable LRS. The demonstrated switching endurance included 2000 SET / RESET pulses. The retention results indicate a volatile LRS in MgO, and the state does not remain distinct from HRS. Although slight LRS relaxation before resistance stabilization is identified in STO, the distinction between the two states remains clear.
[0077] In addition to the GCMO films grown directly on STO and MgO substrates, GCMO film was also prepared on STO substrates with a MgO buffer layer. An MgO buffer layer may have a thickness of 7 nanometers, for example. FIG. 9B illustrates (a) resistance - voltage results for GCMO grown on MgO 7 nm, (b) endurance results after repetitive SET / RESET pulses for MgO 7 nm, and (c) retention results where the decay of LRS towards stable HRS is demonstrated for GCMO on MgO, where MgO is 7 nm in thickness. The distinguishable LRS in MgO-buffered device is likely affected by the HRS resistance level as the absolute resistance is lower compared to the devices residing on the MgO substrate.
[0078] It is to be understood that the embodiments of the invention disclosed are not limited to the particular structures, process steps, or materials disclosed herein, but are extended to equivalents thereof as would be recognized by those ordinarily skilled in the relevant arts. It should also be understood that terminology employed herein is used for the purpose of describing particular embodiments only and is not intended to be limiting. Reference throughout this specification to “one embodiment” or “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, appearances of the phrases “in one embodiment” or “in an embodiment” in various places throughout this specification are not necessarily all referring to the same embodiment. As used herein, a plurality of items, structural elements, compositional elements, and / or materials may be presented in a common list for convenience. However, these lists should be construed as though each member of the list is individually identified as a separate and unique member. Thus, no individual member of such list should be construed as a de facto equivalent of any other member of the same list solely based on their presentation in acommon group without indications to the contrary. In addition, various embodiments and example of the present invention may be referred to herein along with alternatives for the various components thereof. It is understood that such embodiments, examples, and alternatives are not to be construed as de facto equivalents of one another, but are to be considered as separate and autonomous representations of the present invention. Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided, such as examples of lengths, widths, shapes, etc., to provide a thorough understanding of embodiments of the invention. One skilled in the relevant art will recognize, however, that the invention can be practiced without one or more of the specific details, or with other methods, components, materials, etc. In other instances, well-known structures, materials, or operations are not shown or described in detail to avoid obscuring aspects of the invention. While the forgoing examples are illustrative of the principles of the present invention in one or more particular applications, it will be apparent to those of ordinary skill in the art that numerous modifications in form, usage and details of implementation can be made without the exercise of inventive faculty, and without departing from the principles and concepts of the invention. Accordingly, it is not intended that the invention be limited, except as by the claims set forth below. The verbs “to comprise” and “to include” are used in this document as open limitations that neither exclude nor require the existence of also un-recited features. The features recited in depending claims are mutually freely combinable unless otherwise explicitly stated. Furthermore, it is to be understood that the use of "a" or "an", i.e. a singular form, throughout this document does not exclude a plurality.INDUSTRIAL APPLICABILITY
[0079] At least some embodiments of the present disclosure find industrial applicability in manufacture and use of physical artificial neurons methods of manufacture thereof, and / or neuromorphic computing, for example.ACRONYMS LISTAcronym Definition Acronym DefinitionALD atomic layer deposition MBE Molecular beam epitaxy CMOS complementary metal-oxide MGO magnesium oxidesemiconductor,CSD chemical solution deposition NGO neodymium gallium oxide,NdGaO3,GCMO memristive material having a pulsed laser deposition formula Gdi-xCaxMnO3 PLDHRS high resistance state PVD physical vapor deposition LRS low(er) resistance state STO strontium titanate LSAT lanthanum aluminate - XRD X-ray diffraction strontium aluminium tantalateMOCVD metalorganic chemical vapor UHV Ultra-high vacuum depositionMOD metalorganic depositionREFERENCE SIGNS LISTReference Definition Reference Definition 100, 400 A, 400B, (part of) a 150, 350, 450A, substrate400C neuromorphic chip 450B, 450C, 550110, 310, 415 polycrystalline 330 electrical conductor memristivelayer / material111, 311, 451A, region supporting 511,512 terminal / contact 451C polycrystallinegrowth120, 320, 425 monocrystalline 410, 500 neuron memristivelayer / material121, 321, 452B, region supporting 420 synapse452C (substantially)monocrystallinegrowth
Claims
CLAIMS:
1. A physical artificial neuron comprising a polycrystalline memristive layer configured to, in response to applied energy, integrate and leak.
2. The physical artificial neuron claim 1, wherein the poly crystalline memristive layer has at least a high resistance state, HRS, and a first lower resistance state having a lower electrical resistance than the HRS.
3. The physical artificial neuron according to claim 2, wherein the poly crystalline memristive layer is configured to integrate and leak by:- reaching the first lower resistance state from the HRS in response to the applied energy; and- leaking towards the HRS from the first lower resistance state in absence of the applied energy.
4. The physical artificial neuron according to claim 2 or 3, wherein the polycrystalline memristive layer has a plurality of lower resistance states comprising the first lower resistance state and a second lower resistance state having a lower electrical resistance than the first lower resistance state.
5. The physical artificial neuron according to claim 4, wherein the poly crystalline memristive layer is further configured to integrate and leak by:- reaching the second lower resistance state from the first lower resistance state in response to the applied energy; and- leaking towards the HRS from the second resistance state in absence of the applied energy.
6. The physical artificial neuron according to any one of claims 1 to 5, wherein the applied energy is applied via a voltage.
7. The physical artificial neuron according to any one of claims 1 to 6, wherein the polycrystalline memristive layer comprises memristive material having a formula Gdi-xCaxMnCL wherein x is from 0.55 to 0.99.
8. The physical artificial neuron according to any one of claims 1 to 7, comprising oxidizing metal in contact with the polycrystalline memristive layer.
9. A physical neuromorphic chip comprising:- at least one physical artificial neuron comprising a polycrystalline memristive layer;and- at least one physical artificial synapse comprising a memristive layer.
10. The physical neuromorphic chip according to claim 9, wherein the memristive layer of the at least one physical artificial synapse is a substantially monocrystalline memristive layer.
11. The physical neuromorphic chip according to claim 9 or 10, wherein the at least one physical artificial neuron is the artificial neuron according to any one of claims 1 to 8.
12. The physical neuromorphic chip according to any one of claims 9 to 11, wherein the memristive layer of the at least one physical artificial synapse is configured to substantially retain a resistance state in absence of an external voltage.
13. The physical neuromorphic chip according to any one of claims 9 to 12, wherein the memristive layer of the at least one physical artificial synapse comprises a memristive material having a formula Gdi-xCaxMnO3 wherein x is from 0.55 to 0.99.
14. The physical neuromorphic chip according to any one of claims 9 to 13, wherein the at least one artificial neuron and the at least one artificial synapse are in contact with one another.
15. The physical neuromorphic chip according to any one of claims 9 to 14, wherein the at least one artificial neuron and the at least one artificial synapse are connected to one another via a non-oxidizing electrical conductor, for example gold.
16. The physical neuromorphic chip according to any one of claims 9 to 15, wherein the neuromorphic chip comprises a substrate, wherein the at least one physical artificial neuron is on a first region of the substrate and the at least one physical artificial synapse is on a second region of the substrate.
17. The physical neuromorphic chip according to claim 16, wherein surface of the first region has a lattice mismatch from 5% to 20, such as from 6% to 17%, from 7% to 15% or from 9 to 15%, with respect to the polycrystalline memristive layer.
18. The physical neurom orphic chip according to claim 16 or 17, wherein the first region comprises a layer of magnesium oxide, MgO.
19. The physical neurom orphic chip according to any one of claims 16 to 18, wherein the second region has a lattice mismatch of less than 5%, less than 4%, less than 3% or less than 2% with respect to the memristive layer of the at least one physical artificial synapse.
20. The physical neuromorphic chip according to claim 16 to 19, wherein the second region comprises a layer of strontium titanate, SrTiOs, lanthanum aluminium oxide, LaAlOs, LAO, strontium lanthanum aluminium oxide, SrLaA104, SLAO, and / or lanthanum aluminate -strontium aluminium tantalate, LSAT.
21. The physical neuromorphic chip according to any one of claims 9 to 20, wherein the at least one physical artificial synapse comprises a first physical artificial synapse and a second physical artificial synapse, said first physical artificial synapse being connected to the second physical artificial synapse via the at least one physical artificial neuron.
22. A method of manufacturing a neuromorphic chip, the method comprising:- obtaining a substrate comprising a first region supporting polycrystalline growth of a first memristive material; and- growing on said first region of the substrate a polycrystalline memristive layer of the first memristive material.
23. The method according to claim 22, wherein the first region has a lattice mismatch from 5% to 20% such as from 6% to 17%, from 7% to 15% or from 9% to 15%, with respect to the first memristive material.
24. The method according to claim 22 or 23, wherein the first region comprises magnesium oxide, MgO.
25. The method according to any one of claims 22 to 24, wherein the first memristive material has a formula Gdi-xCaxMnO3 wherein x is from 0.6 to 0.99.
26. The method according to any one of claims 22 to 25, wherein the substrate further comprises a second region supporting monocrystalline growth of a second memristive material, and the method further comprises:obtaining, on said second region of the substrate, a substantially monocrystalline memristive layer of the second memristive material.
27. The method according to claim 26, wherein the first memristive material is the same material as the second memristive material.
28. The method according to claim 26 or 27, wherein the second memristive material has a formula Gdi-xCaxMnO3 wherein x is from 0.6 to 0.99.
29. The method according to any one of claims 26 to 28, wherein the polycrystalline memristive layer and the memristive layer of the at least one physical artificial synapse are grown simultaneously.
30. A method of manufacturing a neurom orphic chip, the method comprising:growing a first memristive material on a first region of a substrate, wherein the first region supports polycrystalline growth of the first memristive material; andgrowing a second memristive material on a second region of the substrate, wherein the second region supports monocrystalline growth of the second memristive material.
31. The method according to claim 30, wherein the first and second memristive materials are grown simultaneously on the first and second regions, respectively.
32. The method according to claim 30 or 31, wherein the first memristive material has a polycrystalline structure and a formula Gdi-xCaxMnO3 wherein x is from 0.6 to 0.99.
33. The method according to any one of claims 30 to 32, wherein the second memristive material has a monocrystalline structure and a formula Gdi-xCaxMnO3 wherein x is from 0.6 to 0.99.
34. The method according to any one of claims 30 to 33, comprising: growing the first memristive material and the second memristive material in contact with one another, or providing, a non-oxidizable metal contact between the first and second memristive materials.
35. The method according to any one of claims 30 to 34, wherein the first region has a lattice mismatch from 5% to 20% such as from 6% to 17%, from 7% to 15% or from 9% to 15%, with respect to the first memristive material, and the second region has a lattice mismatch of less than 5% with respect to the second memristive material.
36. A physical artificial neuron comprising a polycrystalline memristive layer, wherein the polycrystalline memristive material is configured to, in response to applied energy, integrate and leak37. A physical artificial neuron comprising a polycrystalline Gdi-xCaxMnO3 memristive layer, wherein x is from 0.6 to 0.99, and wherein said polycrystalline material is configured to, in response to applied energy, integrate and leak.
38. A physical artificial neuron made of a polycrystalline Gdi-xCaxMnO3 memristive material, wherein x is from 0.6 to 0.99.
39. A neuromorphic chip comprising:at least one physical artificial neuron according to claim 38; and at least one physical artificial synapse connected to said at least one neuron, wherein the at least one physical artificial synapse is made of moncrystalline Gdi-xCaxMnO3 memristive material, wherein x is from 0.6 to 0.99.
