Multifunctional memristors with BIO-realistic neuromorphic functions

The memristor with a transition metal nitride halide resistance switching layer addresses the limitations of traditional memristors by enabling both volatile and non-volatile operations, achieving fast switching and low power consumption, and simulating synaptic plasticity, thus improving neuromorphic computing capabilities.

WO2024248730A1PCT designated stage expired Publication Date: 2024-12-05NANYANG TECH UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
PCT/SG2024/050340
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-26
Filing Date
2024-05-23
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Traditional memristors are limited by their single-functionality, high power consumption, and inability to simultaneously achieve volatile and non-volatile regimes, which hinders their application in neuromorphic computing and neural networks due to lack of intrinsic plasticity and efficient synaptic behavior.

Method used

A memristor with a transition metal nitride halide resistance switching layer that operates in both volatile and non-volatile regimes, enabling fast switching and low power consumption, and simulating biological synaptic plasticity by controlling the limiting current or voltage.

Benefits of technology

The memristor achieves ultra-fast switching, low power consumption, and stable operation, allowing for the simulation of biological synaptic plasticity and neuromorphic functions, enhancing its applicability in neuromorphic computing and neural networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 000038
    Figure 000038
  • Figure 000039
    Figure 000039
  • Figure 000040
    Figure 000040
Patent Text Reader

Abstract

Herein disclosed is a memristor comprising: a substrate; a first electrode disposed proximal to the substrate; a second electrode disposed distal from the substrate; a resistance switching layer in contact with the first electrode and the second electrode, wherein the resistance switching layer renders the memristor operable between (i) a volatile working regime, (ii) a volatile working regime and a non-volatile working regime, and (iii) a non-volatile working regime, in response to an electrical stimulus, and wherein the resistance switching layer comprises a transition metal nitride halide. A device comprising the memristor and a method for forming the memristor are also described herein.
Need to check novelty before this filing date? Find Prior Art

Description

MULTIFUNCTIONAL MEMR1STORS WITH BIO-REALISTIC NEUROMORPHIC FUNCTIONSCross-Reference to Related Application

[0001] This application claims the benefit of priority of Singapore Patent Application No. 10202301497S, filed 26 May 2023, the content of it being hereby incorporated by reference in its entirety for all purposes.Technical Field

[0002] The present disclosure relates to a memristor and a device that includes the memristor. The preesnt disclosure also relates to a method of forming the memristor.Background

[0003] Artificial systems with brain-like functions appear to have developed rapidly for their promising applications in neuromorphic computing, brain-machine interfaces, and bio-inspired sensorimotor implementation. Neuroplasticity, the biological basis of learning and memory in the brain, relates to short-term plasticity (STP) and long-term plasticity (LTP). STP is considered a short-term dynamic process that may occur in cortical synapse, cause transient enhancement of synaptic connection and quickly decay to its initial state on a millisecond to minute scale. A long or a permanent change in the connection may occur through repeated stimulation, namely, LTP, which may depend on the timing of pre- and post-synaptic spike events, and spike-timing-dependent plasticity (STDP) and spike-rate dependent plasticity (SRDP), may be two important synaptic modification of LTP. It may be worth noting that multistore model: sensory memory (SM), short-term memory (STM), long-term memory (LTM) of human memory in psychology, proposed in certain developments, relates to STP and LTP of synapses, the memorization level may be homologous to connection strength (weight) of synapse, and it may be assumed that SM may be transferred to STM, and to LTM through repeated learning (rehearsals). All of these learning rules about neuroplasticity and multistore model of human memory may serve as basis of developing artificial systems.

[0004] Neuromorphic computing may be considered as the hardware realization of neural network, especially, the implementation of single neuron and synaptic devices. With the advent of a memristor, it seems possible to simulate synaptic plasticity and realize neuromorphic computing since its memory function, namely, the resistance of a memristor may be changcg continuously with the amount of electrical charge flowing through it, modifying its conductance, appearing analogous to biological synapses.

[0005] So far, diverse neuromorphic functions achieved based on traditional memristors, such as two-terminal memristors and three-terminal transistors, suffer limitations. For instance, most of these memristors tend to be single-function, i.e., tend to only work in volatile or non-volatile regime, furthermore, because lack of intrinsic plasticity of neurons or Ca2+dynamics of biological synapse, few electronic devices may demonstrate the coexistence of reliable STP and LTP behavior of biological synapses in a same single device, which limits the implementation of multiple neural networks, such as dynamical virtual reservoir networks, spiking neural network (SNN) and artificial neural networks (ANN) in a same device, thereby in turn limiting the applications of traditional memristors in internet of things (loT), artificial intelligence (Al), consumer electronics, 5G communications, machine learning (ML) and other fields. Several approaches, such as a combination of multiple metal / insulator / metal (MIM)-based non-volatile devices, may be used to improve neuromorphic functions, but the area, complexity and power consumption limits of such alternative devices hinder the integration of high-density circuits for complex applications, therefore, an analog to the biological synapse - especially the emulation of intrinsic plasticity, STP and LTP neuromorphic functions in a single device, namely, multistore model - remains very challenging to achieve with traditional solid-state devices. Also, although traditional memristors, especially resistive random access memory (RRAM) types may be considered the most standout memory technology, there appears no technology that may simultaneously realize versatility, fast switching, low power, and stable operation within a single device since various challenges, such as nonlinearity, variability, unicity and poor endurance, seem difficult to address all at same time or even alone.

[0006] There is thus a need to provide for a solution that addresses one or more of the limitations mentioned above. The solution should at least provide for a device and / or a memristor which resolves one or more of aforesaid limitations. The device and / ormemristor should be a multifunctional, ultra-fast, ultra-low power consumption technology based on a single solid-state memristor for multiple neuromorphic computing. Summary

[0007] In a first aspect, there is provided a memristor comprising: a substrate; a first electrode disposed proximal to the substrate; a second electrode disposed distal from the substrate; a resistance switching layer in contact with the first electrode and the second electrode, wherein the resistance switching layer renders the memristor operable between (i) a volatile working regime, (ii) a volatile working regime and a non-volatile working regime, and (iii) a non-volatile working regime, in response to an electrical stimulus, and wherein the resistance switching layer comprises a transition metal nitride halide.

[0008] In another aspect, there is provided a device comprising the memristor described in various embodiments of the first aspect, wherein the device may be a neural bionic device or a memory storage device or a logic computing device.

[0009] In another aspect, there is provided a method for forming the memristor described in various embodiments of the first aspect, the method comprising: forming the first electrode proximal to the substrate; forming the resistance switching material; having the resistance switching material formed in contact with the first electrode; and depositing the second electrode to be in contact with the resistance switching material.Brief Description of the Drawings

[0010] The drawings are not necessarily to scale, emphasis instead generally being placed upon illustrating the principles of the present disclosure. In the following description, various embodiments of the present disclosure are described with reference to the following drawings, in which:

[0011] FIG. 1 is a cross-section scanning electron microscopy (SEM) image of a nonlimiting example (also see example 2) of a memristor of the present disclosure. The memristor is an Au / β-ZrNBr / Cu memristor, wherein electrode layer 1 is Au, the resistance switching (RS) material in the RS material layer is β-ZrNBr, electrode layer 2 is Cu.

[0012] FIG. 2A shows a volatile I-V curve of the memristor of FIG. 1.

[0013] FIG. 2B shows a non-volatile I-V curve of the memristor of FIG. 1.

[0014] FIG. 2C shows the transition region between volatile and non-volatile resistance switching of the memristor of FIG. 1.

[0015] FIG. 3A shows a volatile I-V curve of the memristor of FIG. 1 except that electrode 2 in this instance is Ag and the RS material layer is β-ZrNCl, such that the memristor is an Au / β-ZrNCl / Ag memristor.

[0016] FIG. 3B shows a non-volatile I-V curve of the memristor of FIG. 3 A.

[0017] FIG. 3C shows the retention of FIG. 3B. HRS denotes high resistance state and LRS denotes low resistance state.

[0018] FIG. 4 A is a plot of the response time of Au / β-ZrNCl / Cu memristor at 100 ns, read at 0.2 V, indicating a fast- switching speed of the present memristor.

[0019] FIG. 4B is a plot of the response time of Au / β-ZrNCl / Cu memristor at 200 ns, read at 0.2 V, indicating a fast- switching speed of the present memristor.

[0020] FIG. 4C is a plot of the response time of Au / β-ZrNCl / Cu memristor at 16 ns read at 0.05 V, indicating a fast- switching speed of the present memristor.

[0021] FIG. 4D is a plot of the response time of Au / (β-ZrNCl / Cu memristor at 10 ns by PGU (pulse generation unit), read at 0.01 V, indicating a fast-switching speed of the present memristor.

[0022] FIG. 5A demonstrates for the emulation of Au / β-ZrNCl / Cu memristor for ncuromorphic short-term synaptic plasticity. When simulating synaptic plasticity, as FIG. 5 A shows, a small pulse 1.1 V with a width of 2 ms is applied, the conductance ofβ-ZrNCl RS layer firstly increases from 5.2 pS to 91 pS. When pulse ended, the conductance of (β-ZrNCl RS layer relaxed back to its original state (5.2 pS) over a characteristic time of 1.6 ms, this relates to ionic diffusion or thermal dissipation, which is similar to Ca2+diffusion process in biological synapses, thus successfully simulating the STP behavior of synapses.

[0023] FIG. 5B demonstrates for the emulation of Au / β-ZrNCl / Cu memristor for neuromorphic short-term synaptic plasticity. FIG. 5B shows a shorter pulse voltage (2.4 V, width 200 ns) applied, the characteristic relaxation time is 8 ps.

[0024] FIG. 5C demonstrates for the emulation of Au / β-ZrNCl / Cu memristor for neuromorphic short-term synaptic plasticity. FIG. 5C shows simulated spike -rate- dependent plasticity (SRDP) function by Au / β-ZrNCl / Cu memristor. When the interval of pulse is large, namely with a low frequency stimulus (1.2 V, width 100 ms, interval 2s), the conductance of β-ZrNCl firstly jumped to a high-conductance state, but relaxed back to the insulating state within characteristic relaxation time after the pulse ended.

[0025] FIG. 5D demonstrates for the emulation of Au / β-ZrNCl / Cu memristor for neuromorphic short-term and long-term synaptic plasticity and their transformation process. When the frequency of stimulus is increased (1.2 V, width 0.5 ms, interval 5 ms), the conductance gradually increased firstly and then jumped to a high-conductance state, and stayed there for a long time even though pulse ended, this simulates a permanent change occurring in the synaptic connection through repeated stimulation, namely, STP transform to LTP by repeated stimulation.

[0026] FIG. 6 A shows the emulation of Au / β-ZrNCl / Cu memristor for a neuromorphic function of At-dependent paired-pulse facilitation (PPF).

[0027] FIG. 6B shows the emulation of Au / β-ZrNCl / Cu memristor for a neuromorphic function of At-dependent paired-pulse depression (PPD).

[0028] FIG. 6C shows the emulation of Au / β-ZrNCl / Cu memristor for a neuromorphic function of spike-timing-dependent plasticity (STDP).

[0029] FIG. 6D shows the emulation of Au / β-ZrNCl / Cu memristor for a neuromorphic function of linear and symmetric conductance modulation.

[0030] FIG. 6E shows the emulation of Au / β-ZrNCl / Cu memristor for a neuromorphic function of spikc-amplitudc-dcpcndcnt plasticity (SADP).

[0031] FIG. 6F shows the emulation of Au / β-ZrNCl / Cu memristor for a neuromorphic function of multistore model of human memory in psychology.

[0032] FIG. 7A demonstrates a resistance switching (RS) curve of the memristor with different top electrode.

[0033] FIG. 7B demonstrates the high thermal stability of the memristor.

[0034] FIG. 7C demonstrates the long retention at different temperature.

[0035] FIG. 7D demonstrates the robust endurance of the memristor in non-volatile regime.

[0036] FIG. 7E demonstrates the robust endurance of the memristor in volatile regime.

[0037] FIG. 7F shows the RS curve of a lateral Au / β-HfNCl / Au memristor and its device diagram (inset).

[0038] FIG. 8 shows versatility on same device (containing a memristor of the present disclosure) makes multiple neural networks possible.Detailed Description

[0039] The following detailed description refers to the accompanying drawings that show, by way of illustration, specific details and embodiments in which the present disclosure may be practiced.

[0040] Features that are described in the context of an embodiment may correspondingly be applicable to the same or similar features in the other embodiments. Features that are described in the context of an embodiment may correspondingly be applicable to the other embodiments, even if not explicitly described in these other embodiments. Furthermore, additions and / or combinations and / or alternatives as described for a feature in the context of an embodiment may correspondingly be applicable to the same or similar feature in the other embodiments.

[0041] The present disclosure relates to a unique multifunctional, ultra-fast and ultralow power consumption device, wherein the device is operable to provide bio-realistic neurons and synaptic behavior that can meet the needs of multiple neural networks.

[0042] The device may include at least one transition metal nitrogen halide (TMN). The device may include a memristor that has the at least one transition metal nitrogen halide. Said differently, the memristor is capable of providing bio-realistic neurons and synaptic behavior that can meet the needs of multiple neural networks.

[0043] Particularly, the present disclosure relates to microelectronic devices and memory technology, and more particularly to a unique class of materials, a unique class of memristor that has the unique class of materials, devices that include such memristor, and having the function of simulating biological neurons and neural synapses, and having applications in storage and computing.

[0044] Devices and memristors of the present disclosure include a unique class of resistance switching (RS) materials, namely, a “band insulator” which includes a transition metal nitride halide (TMN). To provide a better understanding, a “band insulator” in the present disclosure refers to a material having a large band gap between its valence band and conduction band, such that the material docs not conduct electricity efficiently. The energy band may be derived from a quadratic Hamiltonian. If the highest energy of one band is lower than the lowest energy of the remaining bands, there may be a gap in terms of state density. If the electron Fermi (i.e., Fermi level) can fall in this gap, then the bands below the Fermi energy may all be full, and the above bands may all be empty, rendering a “band insulator”. As such, electrical conductivity of the intrinsic material may be deemed an insulator, which is one of the important characteristics of the present mcmristor resistance switching (RS) material. Distinctly, a traditional memristor has a metal-insulator-metal structure, for example the TiCh-x insulator, or h-BN insulator as RS material work under metal electrodes. Under external electrical stimulation, it is the distinct ion movement in these traditional insulators that causes a large conductance change range from insulation to metal.

[0045] The transition metal nitride halide may be denoted by a chemical formula of MNX, wherein M may be titanium (Ti), zirconium (Zr), hafnium (Hf), wherein X may be a halide, such as chlorine (Cl), bromine (Br), iodine (I), and N may be nitrogen. The resistance switching material may have a layered structure and a two dimensional electronic state, as such, the resistance switching material may be structurally configured as a resistance switching layer in a memristor and in a device of the present disclosure. The resistance switching material may change from an insulator state to a metal state and / or even to a superconductor state by electron or carrier doping, exhibiting superior conductance modulation characteristic. Advantageously, based on such resistance switching materials, memristors of the present disclosure and devices having a memristor of the present disclosure, are multifunctional, ultra-fast and have ultra-low power consumption that can work in both volatile and non-volatile regime (even data and analog regime) within a device (just based on the resistance switching material), simultaneously meeting the needs of multiple neural networks and rendering diverse computational primitives as well as multiple time scales signals analyzation in real time possible, which seems to be an unsolved challenge today given that the unicityof a traditional memristor’s function prevents monolithic integration on the same substrate. Compare to traditional memristors that have a single-function or need to be combined with other materials or devices to achieve multiple functions, the present memristor and method of fabricating the present memristor offer a simpler process, lower material cost and arc suitable for large-scale monolithic integration.

[0046] Memristors of the present disclosure, being multifunctional, may simultaneously simulate excited state and resting state of biological neurons and the STP and LTP behavior of biological synapses, realize implementation of multistore model, with clear experimental evidence of time-dependent STP and LTP demonstrated in a single solid-state memristors device, and because of resemblance to Ca2+dynamics of biological synapses memristors of the present disclosure offer considerably bio- rcalistic neurons and synaptic behavior with simultaneous application of a psychological memory model.

[0047] In addition, memristors of the present disclosure may simultaneously realize ultra-fast switching, ultra-low power, and stable operation in a single device, as compared to traditional memristors which face various challenges such as nonlinearity, variability, unicity, and poor endurance.

[0048] Advantageously, memristors of the present disclosure may constitute an improvement over traditional memristors, due to the resistance switching material involved (expanding the range of resistance switching materials). Traditionally, most resistance switching materials of memristor may include metal oxides (TiOx, HfOx, ZrOx, A1OX, TaOx), chalcogenides (AgS2, GeSx, GeSe, AgGeSe), certain two dimensional materials (graphene, M0S2, black phosphorus, hexagonal boron nitride (h- BN)), or certain organics and perovskites. However, the resistance switching material in the resistance switching layer of present memristors distinctly includes a transition metal nitride halides (TMNs, denoted by the chemical formula: MNX as mentioned above, wherein M may be Ti, Zr, or Hf, X may be a halide of Cl, Br, or I, and N is nitrogen). Such materials have a layered structures bound by a weak Van der Waals force and are two dimensional. Such materials have good insulation performance and stability, especially, transition metal nitride halides of 0-MNX materials (which crystallizes in R-3m (No. 166) space group with SmSI type structure), which are comparable to h-BN insulator as one example, but advantageously have lowerfabrication cost and higher quality than h-BN, the latter which tends to involve use of Cu, CuNi or other substrate for its fabrication, in turn resulting in many undesirable pores between the material and the substrate.

[0049] Another advantage is the improved polyfunctionality and straightforward fabrication process. Most traditional mcmristors reported arc single-function devices, that is, either volatile or non-volatile, such that the realization of multiple function requires combination different materials or a combination of volatile and non-volatile devices, in which the area, complexity and power consumption then poses limitation in terms of the monolithic integration of high-density circuits for any application, let alone complex applications. In contrast, the present memristors can have both volatile and non-volatile resistive behaviors within the same memristor, e.g., by controlling the limiting current or voltage, even without changing any device structure and composition. Memristors and devices including memristors of the present disclosure show excellent consistency and repeatability, and can realize large-scale selector arrays to avoid sneak path currents. With this, it is possible to simultaneously simulate neurons, variety of time-dependent STP, LTP synaptic behaviors and realize multiple neuromorphic computing.

[0050] A further advantage is the “all in one” superior performance of the present memristors. For instance, switching time of the present memristor is less than 10 ns (it is possible that the switching time is faster given the limitation of measurement instrument used but this value is already 104times faster than biological synapses), ultra-low power consumption smaller than 7.2 fj (femtojoule), a large read / write ratio greater than 103, high memory retention (at least 18.2 years), robust endurance (larger than 1.5xl06for volatile and 1.5xl05for non-volatile), excellent thermal stability (higher than 613 K), and low cycle-to-cycle and device-to-device variability. The realization of versatility, fast switching, low power consumption, and stable operation in a single memristor device demonstrate the potential applications of the present memristors in energy-efficient brain-like computing.

[0051] Details of various embodiments of memristors of the present disclosure, devices containing such memristors, and a method for forming the memristors, and advantages associated with the various embodiments are described below. Where the embodimentsand advantages are described in the examples section further hereinbelow, they shall not be iterated for brevity.

[0052] The present disclosure describes a memristor. The memristor may comprise a substrate, a first electrode disposed proximal to the substrate, a second electrode disposed distal from the substrate, and a resistance switching layer in contact with the first electrode and the second electrode. The first electrode and the second electrode may be referred to as electrode layer 1 and electrode layer 2 in various examples described further hereinbelow. The resistance switching layer (RS layer), resistance switching material layer, transition metal nitride halide layer (TMN layer), and transition metal nitride halide resistance switching layer (TMN RS layer), are phrases that are used exchangeably in the present disclosure.

[0053] In various embodiments, the resistance switching layer renders the memristor operable between (i) a volatile working regime, (ii) a volatile working regime and a non-volatile working regime, and (iii) a non-volatile working regime, in response to an electrical stimulus, and wherein the resistance switching layer may comprise a transition metal nitride halide. The term “volatile working regime” refers to an operation mode or a condition in which a device's memory requires power to maintain the stored information. In other words, when power is removed or the device is turned off, the data stored in a “volatile” memory is lost. In contrast, “non-volatile working regime” refers to an operation mode or a condition in which data stored in a device's memory is retained even when power is removed from the device, which is in contrast to volatile memory, which loses its stored data when power is removed.

[0054] In various embodiments, the transition metal nitride halide may be represented by a formula:MNX

[0055] wherein M is a transition metal comprising titanium, zirconium, or hafnium, wherein N is nitrogen, and wherein X is a halogen (also termed as ‘halide’) comprising chlorine, bromine, or iodine.

[0056] In various embodiments, the substrate may be an electrically insulating substrate, an electrically conductive substrate, or a semiconducting substrate. In various embodiments, the electrically insulating substrate may comprise silicon dioxide, glass, ceramic, or plastic, wherein the electrically conductive substrate may comprise a metalor graphite, or wherein the semiconducting substrate may comprise silicon, oxide, or nitride.

[0057] In various embodiments, the first electrode may comprise an inert and electrically conductive material, wherein the inert and electrically conductive material may comprise a metal, an oxide, a nitride, and / or a carbon material.

[0058] hi various embodiments, the first electrode may comprise gold, palladium, or platinum.

[0059] Tn various embodiments, the second electrode may comprise an active material or an inert and electrically conductive material, wherein the active material may comprise a transition metal, an alkali metal, an alkaline earth metal, a rare-earth metal, and / or aluminum, wherein the inert and electrically conductive material may comprise a metal, an oxide, a nitride, and / or a carbon material.

[0060] In various embodiments, the second electrode may comprise gold, palladium, copper, silver, titanium, nickel, and / or chromium.

[0061] In various embodiments, the first electrode and / or the second electrode and / or the resistance switching layer may each have a thickness in a range of 1 nm to 500 nm.

[0062] In various embodiments, the memristor may further comprise an adhesive layer on the first electrode, wherein the adhesive layer is between the first electrode and the substrate and is distal from the resistance switching layer, wherein the adhesive layer may comprise chromium or titanium.

[0063] In various embodiments, the memristor may further comprise a protective layer on the second electrode and distal from the resistance switching layer, wherein the protective layer may comprise an inert and electrically conductive material, wherein the inert and electrically conductive material may comprise a metal, an oxide, a nitride, and / or a carbon material.

[0064] In various embodiments, the inert and electrically conductive material may comprise gold, palladium, or platinum.

[0065] In various embodiments, the first electrode, the second electrode, and the resistance switching layer may be structurally configured to have a sandwich structure, a T-type structure, an I-type structure, a pyramid-type structure, a lateral structure, a through hole structure, a cross structure, a combination thereof, or a modification thereof.

[0066] The present disclosure also relates to a device comprising the memristor described in various embodiments of the first aspect of the present disclosure. Embodiments and advantages described for the memristor of the first aspect can be analogously valid for the present device described herein, and vice versa. Where the various embodiments and advantages have already been described above and in the examples section further hereinbelow, they shall not be iterated for brevity.

[0067] In various embodiments, the device may be a neural bionic device or a memory storage device or a logic computing device.

[0068] The present disclosure also relates to a method of forming the memristor described in various embodiments of the first aspect of the present disclosure. Embodiments and advantages described for the memristor of the first aspect can be analogously valid for the present device described herein, and vice versa. Where the various embodiments and advantages have already been described above and in the examples section further hereinbelow, they shall not be iterated for brevity.

[0069] Tn various embodiments, the method may comprise forming the first electrode proximal to the substrate, forming the resistance switching material, having the resistance switching material formed in contact with the first electrode, and depositing the second electrode to be in contact with the resistance switching material.

[0070] In various embodiments, forming the resistance switching material may comprise having an ammonium halide disposed in a reactor (e.g., a tube having one open end) disposed proximal to an air inlet, and a transition metal disposed proximal to an opening of the reactor and proximal to an air outlet, heating the reactor in the presence of ammonia or an inert environment (e.g., argon) to render a reaction between the transition metal and the ammonium halide so as to form amorphous transition metal nitride halide, transferring the amorphous transition metal nitride halide to a sealed reactor (e.g., a furnace configured horizontal), and heating the scaled reactor with a temperature gradient to form crystals of the transition metal nitride halide. The “temperature gradient” in the context of the present disclosure means one end of the horizontal furnace is at a temperature higher than the other. The end of the reactor that contains a sample is placed proximal to the lower temperature zone, while another end that has no sample is placed proximal to the higher temperature zone. A non-limiting example is described for the purpose of providing a better understanding. Synthesis ofthe transition metal nitride halide may involve two steps. The first step may involve placing the transition metal at the end of an air outlet, while the reactor (e.g., a tube) that contains ammonium halide is disposed at the end of an air inlet, and the opening of the reactor containing the ammonium halide faces the transition metal, then heated with inert gas flow. When the temperature is higher than 340°C as a non-limiting example, the ammonium halide sublimates and is driven (e.g., blown) towards the transition metal to react with the transition metal at an evaluated temperature (e.g., 500°C to 900°C), forming amorphous transition metal nitride halide. The next step may involve the amorphoous trantision metal nitride halide synthesized in the prior step (optionally mixed with some ammonium halide), and then have the reactor (e.g., tube) sealed and have the sealed tube (length 8 cm to 40 cm) placed in a furnace heated with a temperature gradient, in which the end of reactor (e.g., tube) that contains the amorphous transition metal nitride halide is placed in the lower temperature zone (e.g., 700°C or 750°C) while the other end that has no sample is placed in the the higher temperature zone (e.g., 85O°C). When heated with such temperature gradient, high quality crystals of transition metal nitride halide is formed.

[0071] In certain non-limiting embodiments, forming the resistance switching material may comprise heating the transition metal and ammonium halide in a sealed reactor to render a reaction between the transition metal and the ammonium halide so as to form amorphous transition metal nitride halide, and heating the sealed reactor with a temperature gradient to form crystals of the transition metal nitride halide. This differs from the abovementioned step in that, for example, the above involves a gas (e.g., the inert gas or ammonia).

[0072] In various embodiments, heating the reactor in an inert environment, and heating the transition metal with the ammonium halide in a sealed reactor, may be carried out at a temperature in a range of 500°C to 900°C, 600°C to 900°C, etc., and wherein the temperature gradient may comprise a temperature of 700°C to 850°C, 750°C to 85O°C, etc. In various embodiments, the gas environment (either containing an inert gas or ammonia) may comprise a gas (e.g., inert gas or ammonia) introduced at a flow rate of 2 seem to 200 seem.

[0073] In various embodiments, the ammonium halide may comprise ammonium chloride, ammonium bromide, or ammonium iodide.

[0074] In various embodiments, the method may further comprise forming the adhesive layer on the first electrode in a manner where the adhesive layer is distal from the resistance switching layer in the memristor as described in various embodiments of the first aspect of the present disclosure, and / or forming the protective layer on the second electrode in a manner where the protective layer is distal from the resistance switching layer in the memristor as described in various embodiments of the first aspect of the present disclosure.

[0075] The word “substantially” does not exclude “completely” e.g. a composition which is “substantially free” from Y may be completely free from Y. Where necessary, the word “substantially” may be omitted from the definition of the present disclosure.

[0076] In the context of various embodiments, the articles “a”, “an” and “the” as used with regard to a feature or clement include a reference to one or more of the features or elements.

[0077] In the context of various embodiments, the term “about” or “approximately” as applied to a numeric value encompasses the exact value and a reasonable variance.

[0078] As used herein, the term “and / or” includes any and all combinations of one or more of the associated listed items.

[0079] Unless specified otherwise, the terms "comprising" and "comprise", and grammatical variants thereof, are intended to represent "open" or "inclusive" language such that they include recited elements but also permit inclusion of additional, unrecited elements.Examples

[0080] The present disclosure relates a memristor, a device comprising the memristor, and a method for forming the memristor.

[0081] In general, varieties of memristors were prepared based on the formula MNX (wherein M is a transition metal, such as Ti, Zr, Hf; wherein X is a halide, such as Cl, Br, I; wherein N is nitrogen) as the resistance switching material for the resistance switching layer. The transition metal nitride halide can be any structure, such as a sandwich structure, a T-type structure, an I-type structure, a pyramid-type structure, a lateral structure, a through-hole (open cell) structure, a cross structure, or any structure modified or derived from aforesaid structure. The memristor is operable in both volatileand non-volatile regimes in a same device, able to exhibit nanosecond-order switching times (<10 ns), large memory window (>103), a long-term memory retention (18.2 years), ultralow power consumption (7.2 fj). The present memristors can simultaneously simulate various neuromorphic functions, such as resting and excited states of neuron, paired-pulse facilitation / dcprcssion (PPF / PPD), long-term potentiation / depression (LTP / LTD), STDP, SRDP, SADP of synapse. Most importantly, the present memristors can realize multistate resistance storage, thus able to simulate psychological multistore memory model, that is to say, simulating the transition among SM, STM and LTM memory, including forgetting.

[0082] Taking the sandwich structure as an example, the memristor can be a structure composed of electrode layer 1 (bottom electrode), MNX RS layer, and electrode layer 2 (top electrode) from bottom to top, conversely, it can also be composed of electrode layer 2 (bottom electrode), MNX RS layer, and electrode layer 1 (top electrode) from bottom to top. The material for electrode layer 1, electrode layer 2, and MNX RS layer can be select from any of the materials described above or in the following examples demonstrated below.

[0083] The present memristor, device, and method are described in further details, by way of non-limiting examples, as set forth below.

[0084] Example 1: General description of the present memristor

[0085] The present memristor (and hence the device) has improved multifunctionality and provides similarities of bio-realistic neuromorphic functions, which is an advantageous class of multifunctional memristors that can work both in volatile and non-volatile regimes over traditional memristors. The present memristors has intrinsic plasticity of ions, which can simultaneously simulate the neuromorphic functions of biological neurons and synapses, such as resting state and excited state of neurons, paired-pulse facilitation / dcprcssion (PPF / PPD), spikc-timing-dcpcndcnt plasticity (STDP) and long-term potentiation / depression (LTP / LTD) of synapses, demonstrating the coexistence of STP and LTP behavior. The term “plasticity of ions” refers to a phenomenon where ions within the memristive material (i.e., resistive switching material) move in response to an applied electric field (e.g., application of an external electric field or by the flow of current through a device containing the memristor), leading to changes in the resistance state of the device. The movement of ions, e.g..oxygen vacancies or metal ions or proton or other ions, within the memristive material may render a change in the conductivity or resistance of a device. The resulting change in resistance can be reversible due to the transition metal nitride halide as the resistance switching material. The plasticity of ions in memristors helps enable devices to exhibit non-linear and adaptive behavior, resembling aspects of synaptic plasticity in biological systems. This makes the present memristors suitable for use in neuromorphic computing applications, where the present memristors can be used to emulate the synaptic connections and plasticity found in biological neural networks.

[0086] As shown in FIG. 1, the memristor can include a substrate, an electrode layer 1 (i.e., first electrode), an electrode layer 2 (i.e., second electrode), and a resistance switching layer that is electrically connected with electrode layers 1 and 2. The substrate can be an insulating substrate (thermal oxide silicon (i.e., silicon dioxide) wafer, glass, ceramic, plastic, etc.), a semiconducting substrate (silicon, oxide, nitride, etc.), or an electrically conductive substrate (metal, graphite, etc.). Considering for compatibility with existing CMOS process and the application in integrated electronics, the substrate can be a silicon-based substrate, e.g., a thermal oxide silicon wafer. The material of electrode layer 1 can be one or more of inert materials, such as electrically conductive metals, electrically conductive oxides, electrically conductive nitrides, electrically conductive carbon materials, etc. The term “inert” in the context of the present disclosure, such as an inert material or inert component, refers to a material or component that does not participate in any reaction or does not get reacted away. A non-limiting example of an inert electrically conductive metal can be Au, Pd, and / or Pt, which are selected only for the non-limiting purpose of demonstration. An adhesive layer of chromium (Cr) or titanium (Ti) can be plated to electrode layer 1 (between the first electrode and substrate), and the thickness of electrode layer 1 can be about 1 nm to about 500 nm. Electrode layer 2 can be made of one or more kinds of active metals, such as a transition metal (copper, silver, nickel, titanium, zinc, iron, tin, manganese, etc.), an alkali metal (lithium, sodium, potassium, rubidium, etc.), an alkaline earth metal (magnesium, calcium, strontium, barium, etc.), a rare-earth metal (europium, ytterbium, etc.), and / or other active metals (aluminum, etc.), with a thickness of about 1 nm to about 500 nm. Copper and silver may be preferred. If electrode layer 2 includes one or more kinds of aforesaid active metals, a protective layer (one or more kinds ofinert electrically conductive metal, inert electrically conductive metal, inert electrically conductive oxides, inert electrically conductive nitride, and / or inert electrically conductive carbon materials), can be added on the electrode layer 2 (away from the resistance switching layer). A non-limiting example of an inert electrically conductive metal is Au, Pd, and / or Pt, which is selected only for the non-limiting purpose of demonstration. Similar to electrode layer 1, electrode layer 2 can also be made of one or more of the inert electrically materials, such as aforesaid metals, aforesaid oxides, aforesaid nitrides, aforesaid carbon materials, etc. Non-limiting examples of inert electrically conductive metals include Au, Pd, and / or Pt, which are selected only for the non-limiting purpose of demonstration. Electrode patterns of electrode layers 1 and 2 can be prepared by photolithography or electron-beam lithography (EBL) technique, and then electrodes arc deposited by electron-beam evaporator and followed by a liftoff process.

[0087] The resistance switching (RS) material in the RS material layer (for brevity referred to as resistance switching layer in the present disclosure) includes at least one transition metal nitrogen halides (TMNs), denoted by the formula MNX (wherein M can be Ti, Zr, or Hf; wherein X can be Cl, Br, or 1; wherein N is nitrogen) - any one or any combination of aforesaid materials (the transition metal nitride halides), including aforesaid materials that are formed by doping or intercalation (chemical doping, electrochemical doping, solution doping, ionic gating, co-intercalation of organic molecules, de-intercalation, etc.), or any modification (e-beam-irradiation, plasma treatment, etc.) thereof.

[0088] The resistance switching layer of TMN can be prepared by any solid-state reaction, chemical vapor transport (CVT), chemical vapor deposition, magnetron sputtering, thermal evaporation, sol-gel coating, or other suitable methods, preferably solid-state reaction and CVT may be preferred. Particularly, transition metal nitride halides of the formula MNX (M = Ti, Zr, Hf; X = Cl, Br, I; N = nitrogen) can be prepared by reacting a metal or a metal hydride (M or MH?. M = Ti, Zr, Hf) with an ammonium halide (NH4X, wherein X can be Cl, Br, or T), and purification via chemical vapor transport in a evacuated quartz tube under a temperature gradient, and adjusting temperature, molar ratio and total amount of reaction, highly crystalline a- and P-MNX (M = Ti, Zr, Hf; X = Cl, Br, I; N = nitrogen) can be collected at higher temperaturezone. For MNX that may not be easily obtained at ambient pressure, crystal samples can be prepared in high pressure. Mechanical exfoliation and transfer technique may be used in preparation of resistance switching layer having a thickness of about 1 nm to about 500 nm, the transfer method can be either dry transferred (using polydimcthylsiloxanc (PDMS), polymethyl methacrylate (PMMA), or polypropylene carbonate (PPC), etc.) or wet transfer (liquid wedging, substrate etching, etc.). Dry transfer is selected here for the non-limiting purpose of demonstration, and the PDMS transfer may be preferred.

[0089] In various non-limiting examples demonstrated in the present disclosure, the memristor unit structure can include electrode layer 1, the RS layer, and electrode layer 2, structurally configured as a sandwiched structure, a T-type structure, an I-type structure, a pyramid-type structure, a lateral structure, a through-hole (open cell) structure, a cross structure, or any structure modified or derived from aforesaid structure.

[0090] When using a current-voltage (I-V) curve to characterize performance of present TMN memristor, applying voltage on electrode layer 2, different state of ions, or metal clusters, or conductive filament can be induced by controlling the limiting current (Ice) or threshold voltage, thus changing the working regime of the TMN memristor. The ions, or metal clusters, or conductive filament that influence the working regime can be formed by non-metal (hydrogen, nitrogen, halogen, etc.), alkali metal (lithium, sodium, potassium, rubidium, etc.), alkaline earth metal (magnesium, calcium, strontium, barium, etc.), transition metal (copper, silver, nickel, titanium, zinc, iron, tin, manganese, etc.), rare-earth metal (europium, ytterbium, etc.), or any other transition metal or active metal (aluminum, etc.). When lccis in a small range, the TMN memristor shows a volatile working regime, when Iccincreases to a certain range, the TMN memristor exhibits both volatile and non-volatile working regimes, and TMN memristor changes to non-volatile working regime when the Iccincreases to a certain value. As a non-limiting example, for Cu electrode, when Icc<30 μA, Au / β-ZrNBr / Cu memristor shows volatile RS. When 30 μ A<Icc<500 μA, the memristor exhibits both volatile and non-volatile RS, this region also called self-recovered region. When Icc>500 μA, the Au / β-ZrNBr / Cu memristor shows typical non-volatile RS. As another non-limiting example. For Ag electrode, Icc<300 μ A, Au / β-ZrNCl / Ag memristor showsvolatile RS, when 300 μ A<1CC<1 mA, it exhibits both volatile and non-volatile RS, i.e., self-recovered region, when ICC>1 mA, Au / β-ZrNCl / Ag memristor shows stable nonvolatile RS.

[0091] The TMN memristor can simulate the function of biological neurons and synapses, electrode layer 2 simulates the prcsynaptic membrane, resistance changes of TMN RS layer (or TMN layer for brevity) induced by electrical or photic stimulus simulates the excited or resting states of biological neurons, and its conductance change simulates the change of synaptic weight, electrode layer 1 simulates the postsynaptic membrane.

[0092] When simulating the resting and excited state of neurons by TMN memristors, when the resistance of the TMN layer changes from a high resistance state (HRS) to a low resistance state (LRS), the threshold excited function of biological neurons is simulated, when the resistance of the TMN layer changes from a LRS to HRS, simulates transformation of biological neurons change from an excited state to resting state.

[0093] When a small threshold voltage is applied to electrode layer 2 or 1 , the conductance of the TMN layer increases first and then quickly relaxes back to the original state due to ionic diffusion or thermal dissipation, which is similar to Ca2+diffusion process in biological synapses, thus STP behavior of synapses is simulated. Through repeated stimulation or increase in the rate (frequency) of external stimulus, the conductance of the TMN layer gradually increases and then jump to a high saturation value that is maintained for a long time, simulating a permanent change in the synaptic connection, i.e., simulates the transformation process from STP into LTP and spike-rate-dependent plasticity (SRDP) function of synapse.

[0094] When simulating spike-timing-dependent plasticity (STDP) function of a synapse, the positive and negative spike difference of electrode layer 1 and electrode layer 2 was used to simulate the synaptic weight regulation function of synapses. When the spike difference of electrode layer 2 and 1 was positive, the conductance of the TMN layer decreases, simulating the decrease of synaptic weight; when the spike difference of electrode layer 2 and 1 is negative, the conductance of the TMN layer increases, simulating the increase of synaptic weight.

[0095] Tn addition, STDP can depend on the activities of pre- and post-synaptic neurons, when simulating STDP function of synapses, the time difference betweenelectrode layer 2 and 1 is controlled to be greater than zero (At > 0), that is, a spike of electrode layer 2 (pre-spike) precedes a spike of electrode layer 1 (post-spike), the conductance of the TMN layer increases, simulating an increase of synaptic weight, if controlling the time difference between electrode layer 2 and 1 to be less than zero (At < 0), that is, a spike of electrode layer 1 (post-spike) precedes a spike of electrode layer 2 (pre-spike), the conductance of the TMN layer decreases, simulating a decrease of synaptic weight. Similarly, various other STDP functions of the synapse can also be simulated by modifying the amplitude, shape, and time difference of spike applied in electrode layers 2 and 1 of the present memristor which traditional memristors are not capable of.

[0096] The following examples help in further explaining, for instance the technical scheme and more non-limiting cmbodimcnts / cxamplcs. The following examples and explanation / demonstration are meant to provide an understanding that the memristors, devices, and method, of the present disclosure can be implemented in many different forms and not limited to the embodiments / examples described herein.

[0097] Example 2: A method for preparing a volatile and non-volatile B-ZrNBr memristor

[0098] FIG. 1 shows a cross-sectional SEM image of Au / β-ZrNBr / Cu memristor, which is prepared by the following process:

[0099] 1) Preparation of bottom electrode:

[0100] First, a Si / SiCh substrate was ultrasonic cleaned with deionized water, ethanol and acetone solutions, respectively, and then dried with nitrogen. Using standard EBL process (including photoresist coating preparation, baking, exposure, development, etc.) to prepare a pattern of the bottom electrode (i.e., first electrode) on the substrate, then the adhesive layer is deposited using electron-beam evaporation, e.g., Cr or Ti, with a thickness of 5 nm, and then the bottom electrode is deposited, for example, Au with a thickness of 15 nm, followed a lift-off process.

[0101] 2) Preparation of RS layer:

[0102] Synthesis and exfoliation of RS material: MNX (wherein M is a transitiona metal, such as Ti, Zr, Hf; wherein X is a halide, such as Cl, Br, I) crystals can be synthesized by solid-state reaction and CVT. Here, β-ZrNBr can be obtained by reacting Zr metal with NH4Br (loaded in one quartz tube with one end closed and put near theair intake end) at 500°C to 900 °C (e.g., 600°C to 900°C) under the argon flow (flow rate: 2 seem to 200 seem, 50 seem is demonstrated in this example), then put it in a sealed quartz tube, and recrystallize and purify in a horizontal furnace with a temperature gradient from 700°C to 850°C (e.g., 750°C to 850°C), with high-quality bulk crystal obtained at higher temperature zone. Instead of argon, ammonia can be used. Alternatively, β-ZrNBr also can be obtained by direct reaction of Zr metal with NtUBr in a seal quartz tube at 500°C to 900 °C (e.g., 600°C to 900°C) which forms amorphous transition metal nitride halide, and then purify and recrystallize in a horizontal furnace with a temperature gradient to form crystals of the transition metal nitride halide. Then the β-ZrNBr nanosheet was obtained through mechanical exfoliation.

[0103] Transfer of the RS material to bottom electrode: the dry transfer method with the aid of polydimethylsiloxane film (PDMS film) is selected here. This involved putting a scotch tape with β-ZrNBr sample to PDMS film and press, leave some (1- ZrNBr sample on PDMS film, then transfer selected β-ZrNBr sample to target area of bottom electrode through 2D directional transfer platform under microscope, then heating, allowing the β-ZrNBr nanosheet to fall on the target area, lifting the PDMS, the heating temperature and time during this process is 30°C to 80°C and 1 mins to 6 mins, respectively.

[0104] 3) Preparation of top electrode (second electrode):

[0105] The top electrode pattern is prepared on the RS material and bottom electrode by EBL process, then the top electrode is deposited with electron-beam evaporation, here the top electrode is Cu with the thickness of 35 nm, then a protective layer is deposited on the top electrode, wherein Au is selected with a thickness of 10 nm, then followed a lift-off process.

[0106] An operation control method for volatile and non-volatile Au / β-ZrNBr / Cu memristor is described as follows: select a memristor unit, apply voltage in top electrode, and the bottom electrode is grounded, volatile and non-volatile working regimes can be adjusted by controlling limiting current Icc-

[0107] FIG. 2A shows the volatile I-V curve of the Au / β-ZrNBr / Cu memi istor of the current example 2. When the voltage increases reach threshold, the conductance increases rapidly and the resistance jumps from the high resistance state (HRS) to thelow resistance state (LRS). When the voltage decreases and turns off, the conductance decays rapidly, and the resistance changes from LRS to HRS, indicating that Au / β~ ZrNBr / Cu memristor can work in a volatile regime, so-called threshold switching (TS) behavior. Similarly, increase or decrease the voltage at the negative end shows similar' TS behavior (as FIG. 8 threshold type shows). rooios] FIG. 2B shows the non-volatile I-V curve of the Au / β-ZrNBr / Cu memristor of current example 2. When the positive voltage increases to near set voltage, the conductance increases rapidly and the resistance jumps from HRS to LRS state, however, when the positive voltage decreases and turns off, the resistance remains at LRS unchanged. When the negative voltage increases near reset voltage, the conductance decreases rapidly, and the resistance jumps from LRS to HRS, and when the negative voltage decreases and turns off, the resistance remains at HRS unchanged, indicates Au / β-ZrNBr / Cu memristor works in a non-volatile regime, and shows good consistency and endurance (In FIG. 8, both data RS and analog RS belong to nonvolatile regime, the resistance of data RS suddenly change while the resistance of analog RS gradually change as voltage).

[0109] FIG. 2C indicates there is a transition region between stable volatile and nonvolatile resistive switching (RS), when Lt<30 μA, Au / β-ZrNBr / Cu memristor shows volatile RS. When 30 μA<Icc<500 μA, the memristor exhibits both volatile and nonvolatile RS, this region also called self-recovered region. When Lc>500 μA, the Au / β- ZrNBr / Cu memristor shows typical non-volatile RS.

[0110] Example 3: A method for preparing volatile and non-volatile B-ZrNCl memristor

[0111] The preparation method and procedure of the current example 3 is similar to that of example 2, the difference being that β-ZrNCl was obtained by reacting Zr metal with NH4C1 (loaded in one quartz tube with one end closed and put near the air intake end) at 500°C to 900°C under the argon flow (which helps form amorphous transition metal nitride halide), then put it in a seal quartz tube and recrystallize and purify in a horizontal furnace with a different temperature gradient from 700°C to 85O°C (e.g., 750°C to 85O°C) to form crystals of the transition metal nitride halide. Instead of argon, ammonia can be used. Alternatively, β-ZrNCl also can be obtained by direct reaction of Zr metal with NH4C1 in a seal quartz tube at 500°C to 900°C (e.g., 600°C to 900°C)which forms amorphous transition metal nitride halide, and then purify and recrystallize in a horizontal furnace with a temperature gradient to form crystals of the transition metal nitride halide. Here, the thickness of β-ZrNCl is 8 nm, and the top electrode is Ag with a thickness of 65 nm, the protective layer is Au with a thickness of 30 nm.

[0112] One control method for volatile and non-volatile Au / β-ZrNCl / Ag memristor: Similar to example 2, voltage was applied in top electrode, and the bottom electrode is grounded, volatile and non-volatile working regime can be adjusted by controlling limiting current Icc-

[0113] From FIG. 3A and 3B, it can be seen that Au / β-ZrNCl / Ag memristor are operable in volatile and non-volatile regimes. Compare to example 2, the transition range of stable volatile and non-volatile RS in Au / β-ZrNCl / Ag memristor is different, when Icc<300 u A, Au / β-ZrNCl / Ag memristor shows volatile RS, when 300 μA<Icc<l mA, it exhibits both volatile and non-volatile RS, i.e., self-recovered region, when Icc>l mA, Au / β-ZrNCl / Ag memristor shows stable non-volatile RS, and shows good consistency and endurance. FIG. 3C shows the Au / β-ZrNCl / Ag memristor work in nonvolatile regime has a long-term retention at least larger than 5x105s, and it can be extrapolated to 18.2 years as shown in the inset of FIG. 7C.

[0114] Example 4: A method for preparing volatile and non-volatile An / fi- ZrNBr / Ag memristor

[0115] Different from example 2, the top electrode is Ag with a thickness of 65 nm, the protective layer is Au with a thickness of 30 nm. It can work in volatile and nonvolatile regime, and the transition range of volatile and non-volatile RS similar to that of example 3.

[0116] Example 5: A method for preparing volatile and non-volatile Au / B- ZrNCl / Cu memristor

[0117] Different from example 3, the top electrode is Cu with a thickness of 50 nm, the protective layer is Au with a thickness of 15 nm. It can work in volatile and nonvolatile regime, and the transition range of volatile and non-volatile RS similar to that of example 2.

[0118] Example 6: A method for preparing volatile and non-volatile B-HfNCI memristor

[0119] The preparation method and procedure of current example 6 is similar to that of example 2, the difference is that β-HfNC1 was obtained by reacting Hf metal powder with NH4C1 (loaded in one quartz tube with one end closed and put near the air intake end) at 500°C to 900°C (e.g., 600°C to 900°C) under the argon flow (or ammonia flow) to form amorphous transition metal nitride halide, then put it in a seal quartz tube and recrystallize and purify in a horizontal furnace with a different temperature gradient from 700°C to 850°C (e.g., 750°C to 850°C) to form crystals of the transition metal nitride halide. And beyond that, β-HfNC1 also can be obtained by direct reaction of Hf metal with NH4C1 in a seal quartz tube at 500°C to 900°C (e.g., 600°C to 900°C) to form amorphous transition metal nitride halide, and then purify and recrystallize as above method to form crystals of the transition metal nitride halide. Here, the thickness of β-HfNC1 is 12 nm, and the top electrode is Ag with a thickness of 65 nm, the protective layer is Au with a thickness of 30 nm. It can work in volatile and non-volatile regime, and the transition range of volatile and non-volatile RS similar to that of example 3.

[0120] Example 7: A method for preparing Au / B-HfNCl / Au memristor

[0121] Different from example 6, the top electrode is Au with a thickness of 50 nm, it has good RS performance as FIG. 7A shows.

[0122] Example 8: A method for preparing Au / B-HfNCI / Pd memristor

[0123] Different from example 6, the top electrode is Pd with a thickness of 30 nm, the protective layer is Au with a thickness of 30 nm, it has good RS performance as FIG. 7A shows.

[0124] Example 9: A method for preparing Au / B-HfNCI / Ti memristor

[0125] Different from example 6, the top electrode is Ti with a thickness of 30 nm, the protective layer is Au with a thickness of 30 nm, it has good analog RS performance as FIG. 7A shows, the analog RS means the resistance gradually change as voltage rather than change abruptly like data RS.

[0126] Example 10: A method for preparing Au / B-HfNCI / Cr memristor

[0127] Different from example 6, the top electrode is Cr with a thickness of 5 nm, the protective layer is Au with a thickness of 50 nm, it has good RS performance as FIG. 7A shows.

[0128] Example 11: A method for preparing lateral Au / B-HfNCl / Au memristor

[0129] Different from examples 2 to 10, the structure of memristor in this example is lateral, wherein lateral electrodes were prepared by standard EBL process (including photoresist coating preparation, baking, exposure, development, etc.), then the adhesive layer is deposited with electron-beam evaporation, e.g., Cr with a thickness of 5 nm, and then the electrodes arc deposited, e.g., Au with a thickness of 30 nm, followed a lift-off process. Then, β-HfNC1 RS material is transferred on to the lateral electrodes. As FIG. 7F shows, the lateral Au / β-HfNC1 / Au memristor also has RS performance, but it needs large set and reset voltages to work and the ratio is small.

[0130] Example 12: A method for preparing volatile and non-volatile q-HfNBr memristor

[0131] The prepar ation method and procedure of current example is similar to that of example 6, the difference is that α-HfNBr was obtained by reacting Hf metal powder with NI LBr under the argon flow, then purify as described above. Alternatively, a- HfNBr also can be obtained by direct reaction of Hf metal with NH4Br in a seal quartz tube, and then purify and recrystallize in a horizontal furnace with a temperature gradient. It can work in volatile and non-volatile regimes and the transition range of volatile and non-volatile RS is similar to that of example 3.

[0132] Example 13A: Discussion of one example

[0133] FIG. 4A to 4D show the response time of the Au / β-ZrNCI / Cu memristor in example 5, voltage was applied in top electrode, and the bottom electrode is grounded, a current amplifier (FEMTO DLPCA-200) was used to amplify the channel current, and the amplified current signals were collected by an oscilloscope (RIGOL DG1032Z). As FIG. 4A shows, when apply a 2.6 V pulse with a width of 100 ns, the resistance of β-HfNC1 RS layer changes from HRS to LRS (current increases from 100 nA to 32 μA). FIG. 4B shows, when apply a -2 V pulse with a width of 200 ns, the resistance of 0- ZrNCl RS layer changes from LRS to HRS (current decreases from 68 u A to 60 nA), which indicates Au / β-ZrNCl / Cu memristor with a nanosecond-order response time. In order to explore how much faster of our memristor than biological neurons and synapses (the action potential and speed of mobile Na+, K+, Ca2+in neurons is about 100 mV and milliseconds), and explore the speed limit of the present memristors, two different ways were used to test response speed of the memristors. In FIG. 4C, a current amplifier (FEMTO DHPCA-100) was used to amplify the channel current, and theamplified current signals were collected by an oscilloscope (RIGOL DG1032Z), it shows the response speed is less than 16 ns. In FIG. 4D, another method was used to test the response speed, for example, the bottom electrode is connected to the Source Measurement Units (SMUs) of a Keysight B1500A Semiconductor Device Analyzer, while the top electrode is connected to the Pulse Generation Unit (PGU) of the same instrument, the data indicates that the memristors can stably switch between LRS and HRS state (more than 100 cycles) at 10 ns fast-speed pulse, showing the switching time of memristor less than 10 ns (note that 10 ns reach the limits of the Pulse Generation Unit of the Key sight B1500A Semiconductor Device Analyzer, means the speed of our memristor may faster than 10 ns, but this value is already 104times faster than biological synapses). The energy consumption for set transition can be calculated from the formula ESET = ILRSX VSETX tSET= 1.2 μ A x 0.6 V x 10 ns = 7.2 fj. Therefore, the high operation speed and ultra-low energy consumption demonstrated shows the present memristors qualifies as a next-generation memory device.

[0134] Example 13B: A method to simulate neuromorphic functions of neurons and synapses

[0135] In this discussion, example 5 - the Au / β-ZrNCl / Cu memristor is referred as one of the non-limiting examples of a memristor of the preesnt disclosure. Particularly, a method of simulating a variety of neuromorphic functions is introduced, other TMN memristors which can simulate following neuromorphic functions.

[0136] When simulating neurons and synapses, the top electrode layer simulates the presynaptic membrane, the resistance changes of β-ZrNCl RS layer induced by electrical stimulus simulates the excited or resting states of biological neurons, the conductance change of β-HfNC1 RS layer simulates synaptic weight, the bottom electrode layer simulates the postsynaptic membrane. In the example, pulse voltage applied on the top electrode and the bottom electrode is grounded.

[0137] As FIG. 4A shows, the resistance of β-ZrNCl RS layer changes from HRS to LRS (the corresponding current increased from 100 nA to 32 μA), simulating the threshold excited state of neurons, when the resistance of β-ZrNCl RS layer changes from LRS to HRS (the corresponding current decreased from 68 μ A to 60 nA), simulates the transformation of neurons from excited state to resting state.

[0138] When simulating synaptic plasticity, as FIG. 5A shows, when applied a small pulse 1.1 V with a width of 2 ms, the conductance of β-ZrNCI RS layer firstly increases from 5.2 pS to 91 pS. and when pulse ended, the conductance of β-HfNC1 RS layer relaxed back to its original state (5.2 pS) over a characteristic time of 1.6 ms, this relates to ionic diffusion or thermal dissipation, which is similar to Ca2+diffusion process in biological synapses, thus successfully simulates the STP behavior of synapses. From FIG. 5B, it can be seen that when apply a shorter pulse voltage (2.4 V, width 200 ns), the characteristic relaxation time is 8 ps.

[0139] FIG. 5C and 5D show simulated spike-rate-dependent plasticity (SRDP) function and transformation process from STP to LTP of synapse by Au / β-ZrNCl / Cu memristor. When the interval of pulse is large, namely with a low frequency stimulus (1.2 V, width 100 ms, interval 2s), the conductance of β-ZrNCl firstly jumped to a high- conductance state, but relaxed back to the insulating state within characteristic relaxation time after the pulse ended. However, when increase the frequency of stimulus (1.2 V, width 0.5 ms, interval 5 ms), the conductance gradually increased firstly and then jumped to a high-conductance state, and stayed there for a long time even though pulse ended, this simulated a permanent change occur in the synaptic connection through repeated stimulation, namely, STP transform to LTP by repeated stimulation.

[0140] FIG. 6A shows simulated the paired-pulse facilitation (PPF) behavior, which is an important STP phenomenon in neuroscience that represents the facilitation effect of excitatory postsynaptic current (EPSC), when the pre-synaptic membrane is stimulated by two consecutive positive spikes, the second spike triggers a reinforced EPSC on the basis of first spike, and the degree of amplitude enhancement is determined by the time interval between the two spikes, the PPF experimental data show that PPF value decreases with the increase of interval time (At). Similarly, the paired-pulse depression (PPD) represents the depression effect of EPSC when the pre- synaptic membrane is stimulated by two consecutive negative spikes, the second spike triggers a weakened EPSC based on the first spike, the PPD value also decreases with the increase of interval time (At). The synaptic plasticity is also related to spike-timing- dependent plasticity (STDP) learning rules, another important LTP phenomenon in neuroscience, FIG. 6C shows one kind of simulated (STDP) function, it depends on the activities of pre- and post-synaptic neurons, when pre-spike precedes post-spike.namely, At (At = tpost-tpre) > 0, the conductance (synaptic weight) increases, on the contrary, post-spike precedes pre-spike (At < 0), the synaptic weight decreases, however, there is no change in conductance (synaptic weight) when the pulse interval between pre-spike and post-spike is too large. It should be noted that various other STDP functions of the synapse can also be simulated by modifying amplitude, shape, and time difference of pre- spike and post-spike.

[0141] FIG. 6D shows the linear and symmetric programmable conductance levels of the memristor towards simulating the long-term synaptic plasticity of long-term potentiation / depression (LTP / LTD), it can be seen the conductance gradually linearly increase by 55 positive 0.5 V / 0.2 ms pulse or decrease by 55 negative -0.5 V / 0.2 ms pulse.

[0142] FIG. 6E simulates spikc-amplitudc-dcpcndcnt plasticity (SADP), namely, large spike amplitude induces an enhanced excitatory postsynaptic current (EPSC) and shortens time to reach saturated postsynaptic current, this can be used to modulated LTP and LTD process.

[0143] These TMN memristors also can simulate psychological multistore memory model. As FIG. 6F and FIG. 5D show, when new information arrives, it stores in sensory register as SM, it transforms to STM and stores for short periods of time after selected for study, when repeated (rehearsal, N=l-5) before information complete decay (forgetting) with a high repetition rate, STM changes to LTM, and memory of this information keep for a long time, however, if rehearsal with a low repetition rate, memory decays in a power function, and STM cannot form LTM as FIG. 5C shows.

[0144] Example 14: Summary description of examples of present memristor and applications

[0145] The present disclosure, including the above examples, describe for an unique class of multifunctional memristors that can function in both volatile and non-volatile regime based on an unique class of resistance switching (RS) materials.

[0146] The present disclosure describes for amemristor unit or array device, comprising: a substrate, an electrode layer 1 (first electrode), a resistance switching (RS) material layer (also termed as resistance switching layer); and an electrode layer 2; wherein the RS material layer is in between the electrode layer 1 and the electrode layer 2; the RS material layer is made of a transition metal nitride halide (TMN).

[0147] The memristor, as described above and in one or more examples, may have a sandwich structure, a T-type structure, an I-type structure, a pyramid-type structure, a lateral structure, a through hole (open cell) structure, a cross structure, or any modified structure of these structures.

[0148] The memristor as described above and in one or more examples, wherein the RS material layer is made of any one of the transition metal nitride halide (TMN), any combination of TMNs, or any materials that formed by doping or intercalation (e. g. chemical doping, electrochemical doping, solution doping, ionic gating doping, cointercalation of organic molecules, de-intercalation, etc.) of TMN, or any modified TMN formed through e-beam-irradiation, plasma treatment or other methods.

[0149] The memristor, as described above and in one or more examples, wherein preferably the RS material layer is made of an MNX, wherein M is a transition metal (e.g. Ti, Zr or Hf), N is a nitrogen element, and X is a halogen (e.g. Cl, Br or I); further, the thickness of RS material layer can range from 1 to 500 nm.

[0150] The memristor, as described above and in one or more examples, wherein the RS material layer is prepared by solid-state reaction, chemical vapor transport (CVT), chemical vapor deposition (CVD), magnetron sputtering, thermal evaporation, sol-gel method, coating, or other methods.

[0151] The memristor, as described above and in one or more examples, wherein the substrate is an insulating substrate (e.g. thermal oxide silicon wafer, glass, ceramic, plastic, etc.), a semiconductor substrate (e.g. silicon, oxide, nitride, etc.), or a conductive substrate (e.g. metal, graphite, etc.).

[0152] The memristor, as described above and in one or more examples, wherein the electrode layer 1 is made of one or more of inert materials, such as conductive metals, conductive oxides, conductive nitrides and conductive carbon materials; further, the thickness of electrode layer 1 can range from 1 nm to 500 nm. The term “conductive” being electrically conductive.

[0153] The memristor, as described above and in one or more examples, wherein the electrode layer 2 can be made of one or more of active metals, such as transition metals (e.g. copper, silver, nickel, titanium, zinc, iron, tin and manganese, etc.), alkali metals (e.g. lithium, sodium, potassium, rubidium, etc.), alkaline earth metals (e.g. magnesium, calcium, strontium, barium, etc.), rare-earth metals (e.g. europium, ytterbium, etc.), andother any active metals (e.g. aluminum, etc.); and also can be made of one or more of inert materials, such as conductive metals, conductive oxides, conductive nitrides and conductive carbon material; further, the thickness of electrode 2 can range from 1 nm to 500 nm. The term “conductive” being electrically conductive.

[0154] The mcmristor, as described above and in one or more examples, wherein volatile and non-volatile working regimes can be adjusted by controlling limiting current (Icc) or threshold voltage.

[0155] The memristor, as described above and in one or more examples, wherein the volatile and non-volatile working regimes may vary with different memristor materials (e.g. RS materials, electrode layer 2 materials, electrode layer 1 materials or substrate materials) or different memristor structures.

[0156] The memristor, as described above and in one or more examples, wherein the electrode layer 2 can be used as a pre-synaptic membrane; the RS material layer and its ions change simulate the synapse cleft; conductance of RS material layer simulates synaptic weight; resistance of RS material layer simulates excited or resting state of biological neurons; RS material layer realizes conversion of multiple resistance states; and the electrode layer 1 simulates post-synaptic membrane.

[0157] The memristor, as described above and in one or more examples, wherein both the electrode layer 1 and electrode layer 2 are connectable to an external power supply.

[0158] The present disclosure and one or more examples describe for a volatile and non-volatile device for storage and computing applications comprising the memristor as described above and in one or more examples.

[0159] A neural bionic device comprising the memristor, as described above and in one or more examples, fully operable to provide a variety of neuromorphic functions of synapses and neurons, including paired-pulse facilitation / depression (PPF / PPD), spike - timing-dependent plasticity (STDP), spikc-amplitudc-dcpcndcnt plasticity (SADP), long-term potentiation / depression (LTP / LTD) of synapses, linear and symmetric programmable conductance modulation.

[0160] The memristor, as described above and in one or more examples, usable in neuromorphic computing, including dynamical virtual reservoir networks, spiking neural network (SNN) and artificial neural networks (ANN) and other neural network computing.

[0161] Memristors, devices, and the method described in the present disclosure advantaegously advance storage and in-memory computing applications of memristor in internet of things (loT), artificial intelligence (Al), consumer electronics, 5G communications, machine learning (ML), servers and data centers, and other fields.

[0162] In summary, the present disclosure provides a class of TMN RS materials, based on which a class of multifunctional memristors that can work in both volatile and non-volatile regimes (even data and analog RS) in a same solid-state device were constructed. These memristors can realize multi-resistance storage, thus simultaneously meet the needs of multiple neural networks, such as dynamical virtual reservoir networks, spiking neural network (SNN) and artificial neural networks (ANN), thereby making diverse computational primitives, multiple time scales signals analyzation in real time possible, therefore, from RS materials to devices, and even to neural computing, systematically upgrading the memristor (as FIG. 8 shows). In addition, the present memristors possess “all-in-one” superior index, such as ultra-fast (< 10 ns) switching, ultra-low power consumption (<7.2 f.J), and large memory window (lager than 106and 103for volatile and non-volatile, respectively), a long-term retention (18.2 years as FIG. 7C), robust endurance at least 1.5xl06for volatile (FIG. 7E) and 1.5xl05for non-volatile (FIG. 7D) RS, and excellent thermal stability up to 613 K (FIG. 7B). In addition, they are compatible with CMOS technology, are suitable for large-scale monolithic integration. More importantly, this class of solid-state memristors can simulate the intrinsic plasticity of synapse due to ionic diffusion or thermal dissipation, which is similar to Ca2+diffusion process in biological synapses, thus they can simultaneously simulate varieties neuromorphic functions and multistore model of human memory in psychology. This invention effectively addresses the single-function, cost, non-biorealistic, speed and power consumption issues of reported memristors, and further facilitates the development of multifunctional memristors.

[0163] While the present disclosure has been particularly shown and described with reference to specific embodiments, it should be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the present disclosure as defined by the appended claims. The scope of the present disclosure is thus indicated by the appended claims and all changeswhich come within the meaning and range of equivalency of the claims are therefore intended to be embraced.

Claims

CLAIMS1. A memristor comprising: a substrate; a first electrode disposed proximal to the substrate; a second electrode disposed distal from the substrate; a resistance switching layer in contact with the first electrode and the second electrode, wherein the resistance switching layer renders the memristor operable between (i) a volatile working regime, (ii) a volatile working regime and a non-volatile working regime, and (iii) a non-volatile working regime, in response to an electrical stimulus, and wherein the resistance switching layer comprises a transition metal nitride halide.

2. The memristor of claim 1, wherein the transition metal nitride halide is represented by a formula:MNX wherein M is a transition metal comprising titanium, zirconium, or hafnium, wherein N is nitrogen, and wherein X is a halogen comprising chlorine, bromine, or iodine.

3. The memristor of claim 1 or 2, wherein the substrate is an electrically insulating substrate, an electrically conductive substrate, or a semiconducting substrate.

4. The memristor of claim 3, wherein the electrically insulating substrate comprises silicon dioxide, glass, ceramic, or plastic, wherein the electrically conductive substrate comprises a metal or graphite, or wherein the semiconducting substrate comprises silicon, oxide, or nitride.

5. The memristor of any one of claims 1 to 4, wherein the first electrode comprises an inert and electrically conductive material, wherein the inert and electrically conductive material comprises a metal, an oxide, a nitride, and / or a carbon material.

6. The memristor of any one of claims 1 to 5, wherein the first electrode comprises gold, palladium, or platinum.

7. The memristor of any one of claims 1 to 6, wherein the second electrode comprises an active material or an inert and electrically conductive material, wherein the active material comprises a transition metal, an alkali metal, an alkaline earth metal, a rare-earth metal, and / or aluminum, wherein the inert and electrically conductive material comprises a metal, an oxide, a nitride, and / or a carbon material.

8. The memristor of any one of claims 1 to 7, wherein the second electrode comprises gold, palladium, copper, silver, titanium, nickel, and / or chromium.

9. The memristor of any one of claims 1 to 8, wherein the first electrode and / or the second electrode and / or the resistance switching layer each has a thickness in a range of 1 nm to 500 nm.

10. The memristor of any one of claims 1 to 9, further comprising an adhesive layer on the first electrode, wherein the adhesive layer is between the first electrode and the substrate and is distal from the resistance switching layer, wherein the adhesive layer comprises chromium or titanium.1 1. The memristor of any one of claims 1 to 10, further comprising a protective layer on the second electrode and distal from the resistance switching layer, wherein the protective layer comprises an inert and electrically conductive material, wherein the inert and electrically conductive material comprises a metal, an oxide, a nitride, and / or a carbon material.

12. The memristor of claim 11, wherein the inert and electrically conductive material comprises gold, palladium, or platinum.

13. The memristor of any one of claims 1 to 12, wherein the first electrode, the second electrode, and the resistance switching layer are structurally configured to have a sandwich structure, a T-type structure, an 1-type structure, a pyramid-type structure, a lateral structure, a through hole structure, a cross structure, a combination thereof, or a modification thereof.

14. A device comprising the memristor of any one of claims 1 to 13, wherein the device is a neural bionic device or a memory storage device or a logic computing device.

15. A method for forming the memristor of any one of claims 1 to 13, the method comprising: forming the first electrode proximal to the substrate; forming the resistance switching material; having the resistance switching material formed in contact with the first electrode; and depositing the second electrode to be in contact with the resistance switching material.

16. The method of claim 15, wherein forming the resistance switching material comprises: having an ammonium halide disposed in a reactor disposed proximal to an air inlet, and a transition metal disposed proximal to an opening of the reactor and proximal to an air outlet; heating the reactor in the presence of ammonia or an inert environment to render a reaction between the transition metal and the ammonium halide so as to form amorphous transition metal nitride halide; transferring the amorphous transition metal nitride halide to a sealed reactor; and heating the sealed reactor with a temperature gradient to form crystals of the transition metal nitride halide.

17. The method of claim 15, wherein forming the resistance switching material comprises: heating the transition metal and ammonium halide in a sealed reactor to render a reaction between the transition metal and the ammonium halide so as to form amorphous transition metal nitride halide; and heating the sealed reactor with a temperature gradient to form crystals of the transition metal nitride halide.

18. The method of claim 16 or 17, wherein heating the reactor in an inert environment and heating the transition metal with the ammonium halide in a sealed reactor are carried out at a temperature in a range of 500°C to 900°C. and wherein the temperature gradient comprises a temperature of 700°C to 850°C.

19. The method of any one of claims 16 to 18, wherein the ammonium halide comprises ammonium chloride, ammonium bromide, or ammonium iodide.

20. The method of any one of claims 15 to 19, further comprising: forming the adhesive layer on the first electrode in a manner where the adhesive layer is distal from the resistance switching layer in the memristor of any one of claims 1 to 13; and / or forming the protective layer on the second electrode in a manner where the protective layer is distal from the resistance switching layer in the memristor of any one of claims 1 to 13.

Citation Information

Patent Citations

  • Multilayer device having intrinsic layered-type single crystal material and method for preparing the same

    KR1020150087913A

  • High Temperature Resistant Memristor Based on Two-Dimensional Covalent Crystal and Preparation Method Thereof

    US20180277754A1