Self-adaptive homogeneous reserve pool computing system based on oxide memristor

Abstract

The invention belongs to the related technical field of micro-nano electronics, and discloses a self-adaptive homogeneous reserve pool computing system based on oxide memristors, and a storage pool layer of the self-adaptive homogeneous reserve pool computing system comprises a plurality of volatile dynamic memristors with nonlinear short-time attenuation memory characteristics; a full-connection readout layer comprises a plurality of non-volatile memristors with a multi-valued long-time-history memory characteristic; the feedback loop is used for adjusting the bias voltage applied during the state attenuation period of the dynamic memristor, and the two types of memristors adopt the same laminated structure containing a titanium oxide layer. According to the method, the problems that material systems of a storage pool layer and a read-out layer are not uniform, and the time scale of a storage pool system is single and cannot be adjusted are solved, and the time scale of the storage pool can be adaptively adjusted, so that the method can better adapt to different types of input signals, the calculation performance of the storage pool is improved, and the calculation efficiency is improved. And a dynamic information identification function is completed.

Description

Technical Field

[0001] The present invention belongs to the technical field related to micro-nanoelectronic technology, and more specifically, relates to an adaptive homogeneous reserve pool computing system based on oxide memristors. Background Art

[0002] With efficient parallel computing capabilities, hardware artificial neural networks have greatly improved the computing power of machine learning. Artificial neural networks mainly include two categories: feedforward neural networks (FNNs) and recurrent neural networks (RNNs). Although RNNs can process time series information, the large number of complex internal loop connections hinders its hardware implementation.

[0003] The reservoir network, as a derivative of RRNs, can nonlinearly map low-dimensional time series into linearly separable reservoir states, providing an optional network framework for processing spatiotemporal information. The main body of the reservoir network consists of two parts: the reservoir layer and the readout layer. The reservoir layer is used to map sequence information to a high-dimensional feature space and is composed of fixed nonlinear nodes. The readout layer is used for simple linear processing and is composed of non-volatile weight units that need to be trained. The network has an extremely low hardware implementation cost, and since only the weights of the readout layer need to be trained, the training cost is greatly reduced.

[0004] However, due to the different functions of the reservoir layer and the readout layer, most of the reservoir hardware systems currently proposed based on new memories use different materials to implement their reservoir layer and readout layer respectively, which limits the improvement of system integration and process compatibility. On the other hand, the time characteristics of most of the current reservoirs based on microelectronic devices are determined by the manufacturing process. Once the device is prepared, the time characteristics cannot be changed. Therefore, most of the current reservoir systems can only process sequence information of a specific time scale, which seriously limits the scalability of the reservoir system in terms of time scale. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides an adaptive homogeneous reserve pool computing system based on oxide memristors, which aims to improve system integration and process compatibility, and improve scalability on a time scale.

[0006] To achieve the above object, the present invention provides an adaptive homogeneous reserve pool computing system based on oxide memristors, which comprises:

[0007] A reservoir layer, comprising a plurality of volatile dynamic memristors with nonlinear short-term decay memory characteristics, for receiving timing signals and extracting characteristic information; the dynamic memristor comprises a top electrode, a bottom electrode and a first stacked layer in the middle;

[0008] A fully connected readout layer, comprising a plurality of non-volatile memristors with multi-valued long-term memory characteristics, for receiving the characteristic information output by the reservoir layer and outputting a recognition result, wherein the non-volatile memristor comprises a top electrode, a bottom electrode and a second stacked layer in the middle;

[0009] A feedback loop, for reversely adjusting the bias voltage applied during the decay of the dynamic memristor in the reservoir layer according to the recognition result output by the fully connected readout layer, so that the decay time constant of the dynamic memristor matches the time scale of the input timing signal;

[0010] The first stack and the second stack are both formed by a first oxide layer and a second oxide layer stacked in contact, the first oxide layer is a titanium oxide layer, the material of the second oxide layer is a material with an electron affinity lower than that of titanium oxide and having non-volatile resistive switching properties, and the oxygen vacancy content of the titanium oxide layer in the first stack is less than a content threshold for generating conductive filaments, and the oxygen vacancy content of the titanium oxide layer in the second stack is greater than or equal to the content threshold for generating conductive filaments.

[0011] Optionally, the fully connected readout layer has a plurality of output terminals, and the recognition result includes recognition of the content category of the timing signal and recognition of the decay speed of the dynamic memristor;

[0012] The reverse adjustment of the bias voltage of the dynamic memristor in the reservoir layer according to the recognition result output by the fully connected readout layer includes: reverse adjustment of the bias voltage of the dynamic memristor in the reservoir layer according to the recognition result of the attenuation speed of the dynamic memristor output by the fully connected readout layer.

[0013] Optionally, the second oxide layer is made of aluminum oxide or hafnium oxide.

[0014] Optionally, the dynamic memristor includes a substrate and a bottom electrode, a first stack and a top electrode stacked on the substrate from bottom to top, and the titanium oxide layer in the first stack is located on a side close to the top electrode; the non-volatile memristor includes a substrate and a bottom electrode, a second stack and a top electrode stacked on the substrate from bottom to top, and the titanium oxide layer in the second stack is located on a side close to the bottom electrode.

[0015] Optionally, the bottom electrode is adhered to the substrate via a bottom electrode adhesion layer, and the material of the bottom electrode adhesion layer is titanium.

[0016] Optionally, the first stack and the second stack are both prepared by an atomic layer deposition process, wherein:

[0017] The titanium oxide layer in the first stack is deposited at a temperature of 110° C. to 200° C. using tetrakis(dimethylamino)titanium as a titanium source precursor and H2O as an oxygen source;

[0018] The titanium oxide layer in the second stack is deposited at a temperature of 250° C. to 350° C. using tetrakis(dimethylamino)titanium as a titanium source precursor and H 2 O as an oxygen source.

[0019] Optionally, the same material layers in the first stack and the second stack have the same thickness, wherein the thickness of the titanium oxide layer is 35 nm to 45 nm, and the thickness of the second oxide layer is 4 to 6 nm.

[0020] Optionally, the material of the top electrode and the bottom electrode is selected from Al, Pt, Au, W, Ti or TiN.

[0021] Optionally, an input layer is also included, which is used to convert the target to be identified into a time series signal and input it into the reservoir layer.

[0022] Optionally, a control module is further included, configured to apply the bias voltage to the dynamic memristor only during a period in which a voltage pulse without a timing signal is applied to the dynamic memristor.

[0023] In general, compared with the prior art, the above technical solutions conceived by the present invention mainly have the following beneficial effects:

[0024] 1. The adaptive homogeneous storage pool computing system based on oxide memristor proposed in the present invention adopts a stack of layers based on the same material, wherein the first oxide layer is a titanium oxide layer, and the second oxide layer is a material with electron affinity lower than that of titanium oxide and non-volatile resistive switching performance. By controlling the oxygen vacancy content in the titanium oxide layer, the dynamic memristor has a nonlinear short-term decay memory characteristic, and the non-volatile memristor has a long-term memory characteristic. The electron affinity of the second oxide layer is lower than that of titanium oxide, so that the dynamic memristor has a larger resistance window and can realize a richer storage pool state. The present invention A homogeneous reservoir computing system is realized, and the integration potential and process compatibility of the system are improved; when a timing signal is input, a timing signal voltage pulse is applied to the dynamic memristor, the conductance of the dynamic memristor changes, and the time feature is extracted and used for subsequent network reasoning and recognition. When the timing signal voltage pulse is withdrawn, the dynamic memristor enters the conductance decay stage. The present invention proposes a feedback loop, applies a bias voltage during the decay of the conductance state of the dynamic memristor, and utilizes the ion drift and diffusion phenomenon inside the interface resistive layer to realize the short-term memory characteristics with modulatable decay time constant, thereby making the time scale of the reservoir modulatable. In summary, the present invention solves the problems of the inconsistency of the reservoir layer and the readout layer material system and the single and unadjustable time scale of the reservoir system, and the method proposed by the present invention can adaptively adjust the time scale of the reservoir, so that it can better adapt to different types of input signals, thereby improving the calculation performance of the reservoir and completing the dynamic information recognition function.

[0025] 2. In an optional embodiment, the fully connected readout layer can output an identification result of the decay speed of the dynamic memristor, and directly modulate the bias voltage based on the result. In this way, modulation can be performed quickly and the computing cost can be reduced.

[0026] 3. In an optional embodiment, the material used for the second oxide layer is aluminum oxide or hafnium oxide. Aluminum oxide or hafnium oxide are materials with lower electron affinity than titanium oxide and non-volatile resistive switching properties. When stacked with the titanium oxide layer, they can form a larger interface barrier, thereby enabling the dynamic memristor to have a larger resistance window and achieve a richer storage state. Moreover, when stacked with the titanium oxide layer with a higher oxygen content, they can form stable conductive filaments under external voltage stimulation and have non-volatility.

[0027] 4. In an optional embodiment, the conditions of the titanium oxide layer preparation process are restricted to form a stack with different oxygen vacancy concentrations, thereby achieving different resistive switching characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic structural diagram of an oxide memristor-based adaptive homogeneous reserve pool computing system in one embodiment of the present invention;

[0029] Figure 2 is a schematic diagram of the conductance change of a dynamic memristor in one embodiment of the present invention under the combined action of a timing signal and a bias voltage;

[0030] Figure 3 is a schematic structural diagram of a dynamic memristor in one embodiment of the present invention;

[0031] Figure 4 is a schematic diagram of attenuation changes of a dynamic memristor under different bias voltages in one embodiment of the present invention;

[0032] Figure 5 is a schematic structural diagram of a dynamic memristor in one embodiment of the present invention;

[0033] Figure 6 is a long-term inhibition cycle measurement diagram of a non-volatile memristor in one embodiment of the present invention;

[0034] Figure 7 is a schematic diagram of system prediction accuracy in one embodiment of the present invention;

[0035] Figure 8 This is a comparison chart of the time scale range that can achieve accurate prediction under different bias voltages. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0037] like Figure 1 Schematic diagram of the structure of an oxide memristor-based adaptive homogeneous reservoir computing system in one embodiment of the present invention, which includes a reservoir layer, a fully connected readout layer and a feedback loop.

[0038] The reservoir layer includes a plurality of volatile dynamic memristors with nonlinear short-term decay memory characteristics, which are used to receive timing signals and extract characteristic information; the dynamic memristor includes a top electrode, a bottom electrode and a first stacked layer located in the middle.

[0039] The fully connected readout layer includes a plurality of non-volatile memristors with multi-valued long-term memory characteristics, which are used to receive feature information output by the storage layer and output recognition results. The non-volatile memristors include a top electrode, a bottom electrode and a second stacked layer located in the middle.

[0040] Among them, the first stack and the second stack are both formed by stacking a first oxide layer in contact with a second oxide layer, the first oxide layer is a titanium oxide layer, the material of the second oxide layer is a material with an electron affinity lower than that of titanium oxide and having non-volatile resistive switching properties, and the oxygen vacancy content of the titanium oxide layer in the first stack is less than the content threshold for generating conductive filaments, and the oxygen vacancy content of the titanium oxide layer in the second stack is greater than or equal to the content threshold for generating conductive filaments.

[0041] The feedback loop is used to reversely adjust the bias voltage applied during the decay of the dynamic memristor in the reservoir layer according to the recognition result output by the fully connected readout layer, so that the decay time constant of the dynamic memristor matches the time scale of the input timing signal.

[0042] In the above structure, the reservoir layer is composed of volatile dynamic memristors. After being stimulated by external voltage, the conductance of the dynamic memristor will change briefly, and decay according to the nonlinear short-term decay memory characteristics and return to the original state, thereby realizing the nonlinear mapping function of the reservoir layer to the time series. The fully connected readout layer is composed of non-volatile memristors. A non-volatile memristor is used as a non-volatile weight unit of the readout layer. The non-volatile memristor can realize multi-value storage, thereby ensuring that the weight it represents can be adjusted during training, and due to its long-term memory characteristics, that is, non-volatility, it can be ensured that the adjusted weight will not change automatically, ensuring the stability of its weight.

[0043] When carrying out specific structural design, the structures of dynamic memristors and non-volatile memristors are basically the same. Both contain a stack between the top electrode and the bottom electrode. The stack is composed of a first oxide layer and a second oxide layer. Oxygen vacancies are formed inside the oxide layer, and the materials of the first oxide layers of the two stacks are the same, both are titanium oxide layers, and the materials of the second oxide layers of the two stacks are the same, both are materials with lower electron affinity than titanium oxide and have non-volatile resistive switching properties.

[0044] In a dynamic memristor, its resistance switching mechanism is as follows: when a reverse voltage is applied to one electrode of the titanium oxide layer, the positively charged oxygen vacancies in the titanium oxide layer move away from the interface, so the conduction band on the titanium oxide side at the interface increases. Similarly, the oxygen vacancies in the second oxide layer migrate toward the interface, and the conduction band on the second oxide layer side at the interface decreases, ultimately resulting in a decrease in the interface potential barrier and an increase in the effective conductance of the memristor. At this time, since the oxygen vacancy content of titanium oxide is less than the content threshold for producing conductive filaments, stable conductive filaments will not be formed under external excitation. Therefore, when the external excitation is removed, due to the concentration gradient of oxygen vacancies between the interface and the body, the oxygen vacancies diffuse spontaneously, the interface potential barrier gradually increases, and the conductance gradually decreases until it returns to its original state. The above process is the short-term attenuation memory characteristic. It can be seen that the conductance value of the dynamic memristor is determined by the barrier height of the titanium oxide layer / second oxide layer contact interface in the first stack. In the present invention, the electron affinity of the material used for the second oxide layer is lower than that of titanium oxide, so the titanium oxide layer / second oxide layer can form a higher interface barrier, thereby enabling the dynamic memristor to have a larger resistance window and achieve a richer reservoir state.

[0045] In a non-volatile memristor, since the second oxide layer is a material with non-volatile resistive switching properties, the following resistive switching mechanism can be achieved: when a reverse voltage is applied to an electrode on one side of the second oxide layer, positively charged oxygen vacancies migrate from the second oxide layer to the titanium oxide layer. Since the oxygen vacancy content of the titanium oxide layer is greater than or equal to the content threshold for producing conductive filaments, a large number of oxygen vacancies can be locally aggregated to form a conductive channel, that is, to form conductive filaments, and the memristor exhibits low resistance; when a forward voltage is applied to an electrode on one side of the second oxide layer, oxygen vacancies migrate, the conductive filaments break, and the memristor exhibits high resistance. Once the filaments are formed, when the external excitation is removed, their conductivity can remain stable, exhibiting long-term memory characteristics.

[0046] In the present invention, the dynamic memristor and the non-volatile memristor designed as above both adopt stacking and the materials used for the stacking are the same, except for the oxygen vacancy content of the titanium oxide layer. The same materials make the dynamic memristor and the non-volatile memristor homogeneous, thereby improving system integration and process compatibility.

[0047] At the same time, the present invention also designs a feedback loop for reversely adjusting the bias voltage applied to the dynamic memristor in the reservoir layer according to the recognition result output by the fully connected readout layer, so that the decay time constant of the dynamic memristor matches the time scale of the input timing signal. In other words, the dynamic memristor will receive the pulse excitation of the timing signal, and the conductivity will change under the pulse excitation to realize the extraction of the high-dimensional spatiotemporal signal of the pulse. When the voltage pulse of the timing signal is withdrawn, the dynamic memristor enters the nonlinear decay stage according to its short-term decay memory characteristics. At this time, the bias voltage is applied to the dynamic memristor for modulation, and its decay speed can be adjusted so that its decay time constant matches the time scale of the input timing signal.

[0048] like Figure 2 The figure shows the conductivity change of the dynamic memristor under the combined action of the timing signal and the bias voltage. signal Under excitation, the conductivity increases, and the timing signal voltage pulse V signal After the removal, apply the bias voltage V bias , modulation attenuation constant. In the timing signal voltage pulse V signal In the interval, the conductivity level G of the device will show a double exponential decay process. This phenomenon is the short-term memory characteristic of the dynamic memristor in the present invention. When the dynamic memristor is used to implement the reserve pool, the short-term memory characteristic of the device determines the time characteristics of the reserve pool. In order to more accurately describe the time characteristics of the reserve pool, a specific number of square wave voltage pulses are applied to the top electrode of the device to make the device reach a specific conductivity state, and then the conductivity evolution curve over time is fitted with a double exponential to obtain the fast decay time constant τ1 and the slow decay time constant τ2 and the corresponding coefficients A1 and A2. These parameters reflect the strength of the association between adjacent virtual nodes in the reserve pool, and can therefore be used to describe the time characteristics of the reserve pool. In the present invention, by applying a bias voltage during the decay period, the fast decay time constant τ1 and the slow decay time constant τ2 can be modulated, thereby modulating the decay speed during the period.

[0049] Generally speaking, the higher the matching degree between the two, the more appropriate the correlation between the nodes, (too strong correlation: leads to information overlap, the reservoir state is called the linear superposition of time series information, and cannot reflect the order information; too weak correlation: the reservoir state can only reflect the current input, and cannot reflect historical information) the more accurate the prediction. In addition, considering that the time scale of the input time series signal is variable, when the time scale of the input time series signal is different, the bias voltage required in the attenuation stage will be different. In order to adapt to this variability, the present invention introduces a feedback loop, which can automatically adjust the bias voltage according to the output result of the readout layer, so that it can adapt to time series signals of different time scales and realize the scalability of the reservoir system in terms of time scale.

[0050] In one embodiment, the fully connected readout layer has multiple output terminals, and the recognition result includes the recognition of the content category of the timing signal and the recognition of the decay speed of the dynamic memristor. The bias voltage of the dynamic memristor in the reservoir layer can be reversely adjusted according to the recognition result of the decay speed of the dynamic memristor by the fully connected readout layer. If the decay speed of the output terminal feedback is too fast, the bias voltage is adjusted to slow down the decay speed of the dynamic memristor. If the decay speed of the output terminal feedback is too slow, the bias voltage is adjusted to speed up the decay speed of the dynamic memristor. It can be understood that before being put into use, the network needs to be trained, and the network has the function of content recognition and the function of recognizing the decay speed that is too fast and too slow through deep learning. Taking image recognition as an example, during the training phase, two types of labels can be added to the input images. One type of label indicates the content of the image, and the other type of label indicates whether the decay time constant of the dynamic memristor is too fast or too slow compared to the time scale of the time series signal when the image is converted into a time series signal and input into the storage pool. Through multiple batches of training, the weights of each node in the readout layer are determined, that is, the conductance of each non-volatile dynamic memristor is determined. After the training is completed, it can be directly put into use. When put into use, the state of each non-volatile dynamic memristor in the readout layer is stable, and each output end of the readout layer will output the content and recognition results as well as the recognition results of the dynamic memristor in the storage layer decaying too fast or too slowly. The feedback loop adjusts the bias voltage based on the recognition result feedback to match it with the time scale of the time series signal as much as possible.

[0051] In one embodiment, the material used for the second oxide layer is aluminum oxide or hafnium oxide. Aluminum oxide or hafnium oxide are materials with lower electron affinity than titanium oxide and non-volatile resistive switching properties. When stacked with the titanium oxide layer, they can form a larger interface barrier, thereby enabling the dynamic memristor to have a larger resistance window and achieve a richer storage state. Moreover, when stacked with the titanium oxide layer with a higher oxygen content, they can form stable conductive filaments under external voltage stimulation and are non-volatile.

[0052] In a specific embodiment, the dynamic memristor includes a substrate, a bottom electrode, a first stack and a top electrode stacked on the substrate from bottom to top, and the titanium oxide layer in the first stack is located on a side close to the top electrode.

[0053] like Figure 3 FIG. 1 is a schematic diagram of the structure of a dynamic memristor in an embodiment of the present invention, which includes, from bottom to top, a substrate 1, a bottom electrode adhesion layer 2, a bottom electrode 3, an AlO y Layer 4, TiO x layer 5 and a top electrode 6.

[0054] The above dynamic memristor can be prepared by the following method:

[0055] 1. Select a SiO2 / Si(100) substrate with a size of 1.5 cm×1.5 cm as the substrate, clean the surface and back with acetone, alcohol, and deionized water respectively to remove dust particles, organic and inorganic impurities, and obtain a substrate 1 to be deposited;

[0056] 2. The bottom electrode 3 is prepared by magnetron sputtering process; the bottom electrode material is preferably Pt, and the bottom electrode thickness is 80nm-120nm, for example, 100nm can be selected; the bottom electrode adhesion layer 2 is Ti, and the thickness ranges from 8nm to 12nm, for example, 10nm can be selected;

[0057] 3. Preparation of TiO by atomic layer deposition process x / AlO y Laminated (AlO y Layer 4 / TiO x Layer 5), the Al source precursor is trimethylaluminum (TMA), the oxygen source is H2O, the deposition temperature is 250°C-350°C, for example, 250°C can be selected, and the deposition thickness is preferably 3nm-7nm, for example, 5nm can be selected; the Ti source precursor is tetrakis(dimethylamino)titanium (TDMAT), the oxygen source is H2O, the deposition temperature is 110°C-200°C, for example, 120°C can be selected, and the deposition thickness is 35nm-45nm, for example, 40nm can be selected; the bottom electrode is retained by a hard mask method during the deposition process; TiO x / AlO y The introduction of the interface enables the dynamic memristor to have modulatable short-term memory characteristics;

[0058] 4. Prepare the pattern of the top electrode on the interface resistive layer by photolithography, and prepare the pattern mask layer of the top electrode by coating, pre-baking, pre-exposure, post-baking, post-exposure, and developing processes;

[0059] 5. Take the sample after photolithography and use the sputtering process to prepare the top electrode 6; the top electrode material is preferably Pt, and the thickness ranges from 90nm to 110nm, for example, 100nm can be selected;

[0060] 6. Take the sample prepared by sputtering the top electrode, soak it in acetone solution for 20 minutes, peel off the pattern mask layer of the top electrode, then wash it with ethanol and deionized water, and blow it dry with a nitrogen gun to obtain a patterned top electrode. The size of the top electrode is preferably 50μm, 100μm, 150μm, and 200μm.

[0061] Experiments show that when a voltage pulse is input to the top electrode of the dynamic memristor, its conductivity state will change accordingly. When the applied voltage pulse is removed, the conductivity level will gradually relax back to the initial state. Experiments further show that the above short-term memory effect can be modulated by the bias voltage of the relaxation process. As the bias voltage value changes, the relaxation process is also accelerated or slowed down, which is ultimately reflected in different decay time constants of the device.

[0062] like Figure 4 The figure shows the attenuation change diagram of the dynamic memristor under different bias voltages. It can be seen that the attenuation speed of the device will be different when the applied bias voltage is different. When the bias voltage increases, the attenuation speed is also significantly accelerated. This proves the effectiveness of the scheme in modulating the attenuation process of the device by adjusting the bias voltage.

[0063] In a specific embodiment, the non-volatile memristor includes a substrate, and a bottom electrode, a second stacked layer, and a top electrode stacked on the substrate from bottom to top, wherein the titanium oxide layer in the second stacked layer is located on a side close to the bottom electrode.

[0064] like Figure 5 The structure diagram of the dynamic memristor in one embodiment of the present invention is shown, which includes a substrate 7, a bottom electrode adhesion layer 8, a bottom electrode 9, a TiO x Layer 10, AlO y layer 11 , and a top electrode 12 .

[0065] The above non-volatile memristor can be prepared by the following method:

[0066] 1. Select a SiO2 / Si(100) substrate with a size of 1.5 cm×1.5 cm as the substrate, clean the surface and back with acetone, alcohol, and deionized water respectively to remove dust particles, organic and inorganic impurities, and obtain a substrate 7 to be deposited;

[0067] 2. The bottom electrode 9 is prepared by magnetron sputtering process; the bottom electrode material is preferably Pt, and the bottom electrode thickness is 80nm-120nm, for example, 100nm can be selected; the bottom electrode adhesion layer 8 is Ti, and the thickness ranges from 8nm to 12nm, for example, 10nm can be selected;

[0068] 3. Preparation of AlO by atomic layer deposition process y / TiO z Bulk resistive switching layer (TiO z Layer 10 / AlO y Layer 11), the Ti source precursor is tetrakis(dimethylamino)titanium (TDMAT), the oxygen source is H2O, the deposition temperature is 250℃-350℃, for example, 250℃ can be selected, and the deposition thickness is 35nm-45nm, for example, 40nm can be selected; the Al source precursor is trimethylaluminum (TMA), the oxygen source is H2O, the deposition temperature is 250℃-350℃, for example, 250℃ can be selected, and the deposition thickness is preferably 3nm-7nm, for example, 5nm can be selected; the bottom electrode is retained by a hard mask method during the deposition process; TiO z The deposition temperature of the layer is adjusted so that it can serve as an oxygen storage layer, giving the device non-volatile resistive switching properties;

[0069] 4. Prepare the pattern of the top electrode on the interface resistive layer by photolithography, and prepare the pattern mask layer of the top electrode by coating, pre-baking, pre-exposure, post-baking, post-exposure, and developing processes;

[0070] 5. Take the sample after photolithography and use the sputtering process to prepare the top electrode 12; the top electrode material is preferably Pt, and the thickness ranges from 90nm to 110nm, for example, 100nm can be selected;

[0071] 6. Take the sample prepared by sputtering the top electrode, soak it in acetone solution for 20 minutes, peel off the pattern mask layer of the top electrode, then wash it with ethanol and deionized water, and blow it dry with a nitrogen gun to obtain a patterned top electrode. The size of the top electrode is preferably 50μm, 100μm, 150μm, and 200μm.

[0072] like Figure 6 The figure shows the long-term inhibition cycle measurement diagram of the non-volatile memristor. The non-volatile memristor is subjected to a long-term depression (LTD) / potentiation (LTP) cycle test. The pulse form used to achieve LTD in the test is a reverse step pulse, that is, a continuous pulse with a constant amplitude change in pulse amplitude; the pulse used to achieve LTP is a single forward pulse. With the addition of the pulse, the normalized conductance is linearly modulated and shows very high cycle stability. In each cycle, the conductance of the non-volatile memristor decreases monotonically, and a total of 32 intermediate resistance states are obtained, and a one-step setting from low conductance to high conductance is achieved by a setting pulse. Linear fitting of the LTD part gives a Pearson correlation coefficient of -0.945, which shows that the conductance state has a strong linear correlation with the number of pulses.

[0073] In the above preparation method, the deposition temperatures of the titanium oxide layer in the first stack and the second stack are different. In the first stack, the deposition temperature of the titanium oxide layer is 110°C ~ 200°C, and in the second stack, the deposition temperature of the titanium oxide layer is 250°C - 350°C. The titanium oxide deposited under low temperature conditions has a lower oxygen vacancy concentration and cannot participate in the formation of stable conductive filaments; the titanium oxide deposited under high temperature conditions has a higher oxygen vacancy concentration and can be used as an oxygen storage layer to provide sufficient oxygen vacancies for the resistive layer. Therefore, although the materials of the first stack and the second stack are the same, different resistive properties can be achieved by controlling the oxygen vacancy concentration of the titanium oxide layer. In addition, since the deposition temperature of the titanium oxide layer in the first stack is lower, the influence of high temperature on the titanium oxide layer can be avoided by first preparing another oxide layer with a higher deposition temperature and then preparing the titanium oxide layer. When preparing the second stack, the experiment found that the second oxide layer is arranged on the side away from the bottom electrode, and it is easier to modulate under external stimulation than the side close to the bottom electrode to reach the corresponding high resistance state. The reason may be that if the second oxide layer is close to the bottom electrode, the adhesion layer Ti of the bottom electrode will adsorb the oxygen ions of the second oxide layer, so that the second oxide layer will produce oxygen vacancies without relying on titanium oxide, and the obtained device will always show low resistance and be difficult to modulate. According to the above method, when building the system, all the bottom electrodes can be grounded, and the timing signal and bias voltage can be directly or processed and applied to the top electrode of the dynamic memristor to realize the resistance change mechanism of the dynamic memristor.

[0074] In one embodiment, the material of the top electrode and the bottom electrode is selected from Al, Pt, Au, W, Ti or TiN.

[0075] In one embodiment, the system further comprises an input layer for converting the target to be identified into a time series signal and inputting the signal into the reservoir layer.

[0076] In one embodiment, the system further includes a control module, configured to apply a bias voltage to the dynamic memristor only during a period in which a voltage pulse without a timing signal is applied to the dynamic memristor.

[0077] When operating, first set an initial bias voltage V bias , and reversely apply the timing signal to the top electrode of the dynamic memristor, ground the bottom electrode, and use the current of the dynamic memristor as the extracted feature to be transmitted to the subsequent recognition network or other classification algorithms such as ridge regression, so as to complete the classification recognition or prediction of the input timing signal, and reversely adjust the bias voltage V based on the output result bias , and so on and so forth, the prediction accuracy can be improved.

[0078] like Figure 7 The figure shows a schematic diagram of the prediction accuracy of the system in one embodiment of the present invention, wherein the prediction root mean square error MSE is ≤ 0.1, indicating that the prediction accuracy is relatively high.

[0079] like Figure 8 The figure shows a comparison of the time scale ranges that can achieve accurate prediction under different bias voltages, where the hybrid means that the bias voltage can be adaptively adjusted, which is the solution of the present invention. It can be seen that the adaptive adjustment of the bias voltage significantly expands the time scale of accurate identification, and the decay time constant of the reservoir system is adaptively adjusted through the feedback loop to match the time scale of the specific input signal, which can significantly improve the time scale of accurate network identification. The advantage of this system is that compared with the traditional reservoir, the system can process a wider range of time series information with a wider frequency range.

[0080] In summary, the adaptive homogeneous reservoir computing system based on oxide memristor proposed in the present invention, the reservoir layer and the readout layer are both stacked layers based on titanium oxide materials, realizing a homogeneous reservoir computing system, and improving the integration potential and process compatibility of the system; a single dynamic memristor constitutes a reservoir unit, when a timing signal is applied to the dynamic memristor, the device current is used as the output signal of the reservoir calculation result, based on the virtual node method in the reservoir calculation, the device current is used for subsequent network reasoning and identification, by applying a bias voltage during the attenuation of the conductivity state of the dynamic memristor, using the ion drift and diffusion phenomenon inside the interface resistive layer, the short-term memory characteristics with modulatable decay time constant are realized, and the time scale of the reservoir can be modulated. In summary, the present invention solves the problems of the inconsistency of the reservoir layer and the readout layer material system and the single and unadjustable time scale of the reservoir system, and the method proposed in the present invention can adaptively adjust the time scale of the reservoir, so that it can better adapt to different types of input signals, thereby improving the reservoir computing performance and completing the dynamic information recognition function.

[0081] The technical features of the above-mentioned embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. It should be noted that "in one embodiment", "for example", "for example", etc. of the present invention are intended to illustrate the present invention, rather than to limit the present invention.

[0082] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent application. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. An adaptive homogeneous reservoir computing system based on oxide memristor, characterized in that: include: A storage layer, comprising a plurality of volatile dynamic memristors with nonlinear short-term decay memory characteristics, for receiving timing signals and extracting characteristic information; The dynamic memristor comprises a top electrode, a bottom electrode and a first stacked layer in the middle; A fully connected readout layer, comprising a plurality of non-volatile memristors with multi-valued long-term memory characteristics, for receiving the characteristic information output by the reservoir layer and outputting a recognition result, wherein the non-volatile memristor comprises a top electrode, a bottom electrode and a second stacked layer in the middle; A feedback loop, for reversely adjusting the bias voltage applied during the decay of the dynamic memristor in the reservoir layer according to the recognition result output by the fully connected readout layer, so that the decay time constant of the dynamic memristor matches the time scale of the input timing signal; The first stack and the second stack are both formed by a first oxide layer and a second oxide layer stacked in contact, the first oxide layer is a titanium oxide layer, the material of the second oxide layer is a material with an electron affinity lower than that of titanium oxide and having non-volatile resistive switching properties, and the oxygen vacancy content of the titanium oxide layer in the first stack is less than a content threshold for generating conductive filaments, and the oxygen vacancy content of the titanium oxide layer in the second stack is greater than or equal to the content threshold for generating conductive filaments.

2. The oxide memristor-based adaptive homogeneous reservoir computing system according to claim 1, characterized in that: The fully connected readout layer has a plurality of output terminals, and the recognition result includes recognition of the content category of the timing signal and recognition of the decay speed of the dynamic memristor; The reverse adjustment of the bias voltage of the dynamic memristor in the reservoir layer according to the recognition result output by the fully connected readout layer includes: reverse adjustment of the bias voltage of the dynamic memristor in the reservoir layer according to the recognition result of the attenuation speed of the dynamic memristor output by the fully connected readout layer.

3. The adaptive homogeneous reservoir computing system based on oxide memristor as claimed in claim 1, characterized in that: The material used for the second oxide layer is aluminum oxide or hafnium oxide.

4. The adaptive homogeneous reservoir computing system based on oxide memristor as claimed in claim 1, characterized in that: The dynamic memristor comprises a substrate and a bottom electrode, a first stack and a top electrode stacked on the substrate from bottom to top, wherein the titanium oxide layer in the first stack is located on a side close to the top electrode; the non-volatile memristor comprises a substrate and a bottom electrode, a second stack and a top electrode stacked on the substrate from bottom to top, wherein the titanium oxide layer in the second stack is located on a side close to the bottom electrode.

5. The oxide memristor-based adaptive homogeneous reservoir computing system according to claim 4, characterized in that: The bottom electrode is adhered to the substrate through a bottom electrode adhesion layer, and the material of the bottom electrode adhesion layer is titanium.

6. The oxide memristor-based adaptive homogeneous reservoir computing system according to any one of claims 1 to 5, characterized in that: The first stack and the second stack are both prepared by an atomic layer deposition process, wherein: The titanium oxide layer in the first stack is deposited at a temperature of 110° C. to 200° C. using tetrakis(dimethylamino)titanium as a titanium source precursor and H2O as an oxygen source; The titanium oxide layer in the second stack is deposited at a temperature of 250° C. to 350° C. using tetrakis(dimethylamino)titanium as a titanium source precursor and H 2 O as an oxygen source.

7. The oxide memristor-based adaptive homogeneous reservoir computing system according to any one of claims 1 to 5, characterized in that: The thickness of the same material layer in the first stack and the second stack is the same, wherein the thickness of the titanium oxide layer is 35nm-45nm, and the thickness of the second oxide layer is 4-6nm.

8. The oxide memristor-based adaptive homogeneous reservoir computing system according to any one of claims 1 to 5, characterized in that: The material of the top electrode and the bottom electrode is selected from Al, Pt, Au, W, Ti or TiN.

9. The adaptive homogeneous reservoir computing system based on oxide memristor as claimed in claim 1, characterized in that: It also includes an input layer, which is used to convert the target to be identified into a time series signal and input it into the reservoir layer.

10. The adaptive homogeneous reservoir computing system based on oxide memristor as claimed in claim 1, characterized in that: The system also includes a control module, which is used to apply the bias voltage to the dynamic memristor only during a period when a voltage pulse without a timing signal is applied to the dynamic memristor.

Images