Electronic phase oscillator-based logic device

Abstract

The present invention relates to a thermally coupled correlated oxide oscillator system comprising at least two correlated oxide oscillators coupled through at least one heating element positioned between at least two of the at least two correlated oxide oscillators. The present invention also relates to a logic device comprising a plurality of coupled correlated oxide oscillator / heating element cells, wherein each cell has the dimensions of a 1-column-n-oscillator.

Description

Technology Field

[0001] The present invention relates to connected correlated oxide devices forming basic building blocks for both Boolean and non-Boolean computations. In particular, VO2 (vanadium oxide)-based thermally coupled phase oscillator devices are described, which generate a uniquely coupled oscillation state, which in turn can be used as a basic building block of a logic device, for example, in a (neural) network. Background Technology

[0002] Neural networks can enable a wide range of computing tasks, such as those related to artificial intelligence or machine learning. One example of a neural network is the Oscillator Neural Network (ONN). An Oscillator Neural Network (ONN) is an artificial neural network that uses combined oscillators as neurons. The Oscillator Neural Network is inspired by the neural oscillation phenomenon occurring in the brain. Oscillator Neural Networks have been trained to recognize images. Complex value oscillator networks have been shown to store and retrieve multidimensional non-periodic signals. Oscillator autoencoders using a combination of oscillators and ratio-coded neurons have also been implemented. Neurons composed of two combined oscillators—one with a fixed eigenvalue and the other with an adjustable eigenvalue—have been shown to be capable of executing logic gates, such as the Exclusive OR (XOR), which conventional S-shaped neurons cannot perform.

[0003] Therefore, a network of combined oscillators provides a very efficient means of computation. The problem to be solved

[0004] The key to developing such technology is establishing the ability to manipulate the interactions between oscillators and the coupling effects between neighboring oscillators in an energy-efficient manner. Therefore, an important aspect for the further development of such networks is to find a suitable oscillator system capable of interacting with one another and controlling the interactions (coupling strength) between neighboring oscillators for synchronization.

[0005] Typically, additional electronic components are required to generate oscillations and control coupling. This not only increases the complexity of designing any computational circuit but also limits the degrees of freedom for controlling network dynamics during operation. Therefore, it would be desirable to provide Boolean (logic gate) and non-Boolean (analog) computational techniques that do not require, or at least require, a reduced number of additional electronic components, and preferably use oscillation states based on both frequency and amplitude. This would allow for richer and more adjustable coupling dynamics and, consequently, more complex synchronous oscillation states. means of solving the problem

[0006] It is known that a certain class of correlated oxide materials near room temperature can exhibit thermally coupled oscillating behavior during phase transitions between two distinct states. For example, NbO2 and Ta2O 5-x [Picket and Dr. RS Williams. Thermally driven threshold switching of less than 100 femtojoules (fJ) and less than nanoseconds (one-billionth of a second) in niobium oxide crosspoint nanodevices. Nanotechnology, 23 , 215202 (2012); Pickett, G. Medeiros-Ribeiro and RS Williams, Ph.D. It is an scalable nervous system built with Mott memristors. Nat. Mat. 12, 114, (2013); A. Sharma, JA Bain, JA Weldon. Phase coupling and control of oxide-based oscillators for neuromorphic computing. IEEE Journal of Exploratory Solid State Computing Devices and Circuits, pp (99):1-1, (2015)].

[0007] Very interesting oscillators can be formed from strongly correlated oxide materials, such as VO2[1, 2], which exhibit a transition from insulator to metal.

[0008] In these materials, controlled oscillation between the low resistance (metallic) state and the high resistance (insulator) state is possible under external stimuli such as current, magnetic field, or electric field [3, 4]. Since the material undergoes a phase transition from insulating to metallic near room temperature [5 to 8], vanadium dioxide (VO2)-based oscillators are particularly important. Coupling between VO2 oscillators can be controlled via external electronic components [9–13] or by thermal links

[14] . However, for such thermally coupled oscillators, additional capacitors / resistors are required to induce oscillation. Additionally, the synchronization range between the two oscillators cannot be adjusted.

[0009] Until now, schemes based on the phase relationship between oscillators have been the most common techniques used in coupled VO2 networks [10-12]. These schemes rely on binary logic in which two states correspond to the phase (0° or 180°) of the oscillator relative to the reference oscillator. However, additional electronic components are required to generate oscillations and coordinate the coupling between these VO2 devices.

[0010] The inventors have now discovered a new and effective approach that enables a thermally coupled correlated oxide oscillator system exhibiting adjustable synchronized frequency and different collective coupling behavior.

[0011] This is achieved by inserting an additional heating element between two oscillators, such as a metal wire, preferably a VO2 nanowire, having a diameter in the nano-range of 1 to 3 digits, for example, Fe-, Cu-, Co-, Ni-, W-, V-, or a corresponding alloy wire, which acts as a thermal cell capable of actively changing the surrounding thermal conditions between the two oscillators. In this system, the oscillators can be effectively synchronized and activated even in a sub-threshold current range through the active control of the heating element. Different operating modes of the heating element (V-mode / I-mode) enable rich and controllable coupling dynamics between the oscillators in the system. In V-mode (heating element), more complex forms of oscillation states (spiking and bursting patterns) are generated from AND, NAND, and NOR gates that mimic the behavior of multiple different basic Boolean neurons. Effects of the invention

[0012] Accordingly, the present invention relates to a thermally coupled correlated oxide oscillator system comprising at least two correlated oxide oscillators coupled through at least one heating element positioned between at least two of the at least two correlated oxide oscillators.

[0013] The present invention further comprises a logic device comprising a plurality of coupled correlated oxide oscillator / column element cells, each cell having the dimensions of a 1-column-n-oscillator, wherein n = 1, 2, 3, 4, 5, 6, 7, 8; preferably n = 2, 3, 4, 5, 6, 7, 8, more preferably n = 2, 3, 4.

[0014] In principle, the number of combined oscillators can be expanded as long as a thermal range is reached. Brief explanation of the drawing

[0015] Figure 1 illustrates the basic transmission behavior of a single VO2 oscillator. a: RT measurements of VO2 bar show a change of at least three orders of magnitude in resistivity. The red curve corresponds to the heating process (from 270 K to 395 K), while the blue curve corresponds to the cooling process (from 395 K to 270 K). b: IV measurement of a single VO2 oscillator (scans current from 0 to 8 mA when measuring voltage). c: A single VO2 oscillator shows voltage oscillation when driven by a constant current source. Here, when a current of 3 mA is supplied, the waveform of the oscillation voltage of the single VO2 oscillator is demonstrated. d: As the current value increases linearly within the negative derivative resistance region, the oscillation frequency increases. Figure 2 illustrates a direct electrical observation of the induction period between two combined VO2 oscillators. a: Schematic diagram of two thermally coupled VO2 oscillators. Oscillator 1 (5×5 μm 2 ) and oscillator 2 (5×5 μm 2 ) is separated by 2 μm (Fig. 2b) and 10 μm (Fig. 2c). d: Oscillation phase when 400μs (pulse period of 800μs) is transmitted to oscillator 1. e: Combined oscillation behavior between oscillator 1 and oscillator 2. In this experiment, oscillator 2 is activated by a constant current of 2.3 mA, which is lower than the oscillation threshold current. Note that the threshold current is 2.5 mA. When oscillator 1 is activated into an oscillating state by a current pulse (2.8 mA - 400 μs), oscillator 2 begins to oscillate at the same frequency. f: Direct observation of the induced delay between two VO2 oscillators spaced 2 μm apart - it takes about 0.53 μs to induce oscillator 2 to the oscillating state. g: Evoked delay time of two VO2 oscillators with different intervals (2 μm, 6 μm, 10 μm) Figure 3 illustrates the adjustable thermal coupling strength between two VO2 oscillators. a: Optical microscope image of a set of two VO2 oscillators with a schematic diagram. A 1.5 μm wide (21 μm long) VO2 strip is placed as a thermal heating element to enhance thermal coupling between two VO2 oscillators of 7×6 μm² dimensions. The distance between oscillator 1 and oscillator 2 is 5 μm. b: Different heating element voltages (V cell I vs. V measurements of the oscillator (1) under ) (sweep current and measured voltage ) . The threshold voltage (V) at which the oscillator enters the negative differential resistance region th ) is V as shown in the inserted drawing cell As this increases, it decreases. c: Different V cell I vs f measurements (sweep current and measurement frequency) from the oscillator (1) under higher V cell In this case, the oscillation frequency of the oscillator (1) increases under DC current. d: V cell = Frequency locking behavior at 0 V. For this measurement, the current to oscillator 1 is fixed at 2.4 mA while increasing the current to oscillator 2. In this case, frequency locking is broken at 23.5 kHz or higher. e: V cell = Frequency locking behavior at 8V. In this case, frequency locking is broken at 25.3kHz or higher. f: Voltage V of different thermal heating elements cell Comparison of frequency synchronization of oscillator 1 (fixed at 2.4 mA) under (0 to 8 V). Figure 4 illustrates a Boolean logic gate based on a thermally coupled VO2 oscillator. a: This is a schematic diagram of logic operations realized by two VO2 oscillators coupled by a thermal heating element. The oscillation state of the oscillator (1) at different currents, I1, functions as an input (A), and the thermal heating element voltage (V cell )Is It functions as an input (B). The oscillation state of oscillator 2 at constant current I2 is the reading (output). An example is as follows. b: AND operation, c: NAND operation, and d: NOR operation Figure 5 illustrates a thermal spike-driven VO2 oscillation with different ignition modes. a: Optical image of a test VO2 oscillator set used to generate spiking potential. VO2 oscillator 1, oscillator 2, and thermal heating element are activated by three independent current sources. b: Oscillation waveform when VO2 oscillator 1 is activated at 3.9 mA, VO2 oscillator 2 is activated at 4.0 mA, and the heating element is activated at 2.3 mA. c: The output waveform from VO2 oscillator 2 is generated by cascade synchronization between three cells. This behavior mimics the generation of spiking potentials in a neuron when activated, as discussed in [7, 34, 35]; region-I: resting state; region-II: stimulus arrives; region-III: depolarization state; region-IV: repolarization and hyperpolarization state (also known as the refractory period). d: While simply changing the current of VO2 oscillator 1 (I1 from 1mA to 5.2mA), VO2 oscillator 2 (I2= 4mA) and thermal heating element (I cell Different firing modes of VO2 neurons by fixing the current for (= 2.3mA). Figure 6 shows the structure of a 2-oscillator 1-heating element. Figure 7 shows the structure of a 4-oscillator 1-heating element. Figure 8 shows a side view of an assembly of buried oxide oscillator / heating element cells formed in a cross-bar type. Specific details for implementing the invention

[0016] Strongly correlated oxide material right before the phase transition temperature 1- Known as correlated oxides It exhibits oscillation characteristics upon external stimulation (e.g., current, magnetic field, electric field, etc.) [1, 2]. In such systems, oscillation of electronic states (e.g., low resistance in the metallic state and high resistance in the insulating state) can be observed, which is considered a fundamental component of the oscillation network. An example of such a correlated oxide is, for example, Ta2O. 5-x (x is preferably between 0 and 1), NbO2, or VO2. Oscillators based on vanadium dioxide (VO2) are particularly important because the material exhibits a transition from insulating to metallic phase near 340K (insulating phase at room temperature) [3-6].

[0017] To develop efficient oscillator-based computational networks, interaction (synchronization) and adjustable coupling strength are two important aspects of such systems. It is already well known that when VO2 oscillators are connected to external electronic components such as resistors, capacitors, or / and inductors, VO2 can exhibit various oscillation characteristics and coupling effects [7-17]. However, these oscillators require additional electronic components to generate oscillation characteristics or to adjust the coupling between VO2 oscillators, which not only reduces the oscillator density but also makes the architecture of a fully functional circuit extremely complex. In addition to oscillation coupling with external electronic complex circuits, it has been reported that VO2 oscillators can be thermally connected using additional capacitors and resistors and used for computational tasks with different VO2 oscillator [18-20] low harmonic ratios. So far, calculations based on the phase relationship between oscillators (binary phase state 0° or 180° relative to the reference oscillator) are the most common techniques in combined VO2 networks [8-16], whereas basic Boolean (logic gate operation) and non-Boolean (analog) calculation techniques using oscillation states based on frequency and amplitude have not yet been explored.

[0018] Unlike phase-relationship-based calculations, richer coupling dynamics from network systems can lead to computational techniques based on both the frequency domain and amplitude. In principle, this can generate rich synchronous oscillation states from coupled oscillators, a challenge that has remained in this domain to date.

[0019] The present invention introduces a novel and effective approach that enables thermally coupled oscillator systems to have adjustable synchronized frequencies and different collective coupling behaviors. To achieve this, an additional heating element, such as a metal wire like Fe-, Cu-, Co-, Ni-, W-, V-, or the corresponding alloy wire, or a nanowire made of the same material as the oscillator, such as VO2 (preferably having a diameter in the 1 to 3 digit nano-range), is implemented to act as a heating element capable of actively changing the surrounding thermal conditions between at least two correlated oxide oscillators. The resistance of the heating element is determined by its size and resistivity. For this purpose, a material having adjustable resistivity, such as Ti 1-x N x Alternatively, a nitride formed from Ta1-xNx may be preferred, where x can vary over a wide range, but is preferably in the range of 0.45 to 0.55 or more preferably in the range of 0.48 to 0.52. The resistivity can be adjusted to optimize the current required to supply energy to the heater. For example, it may be desirable to have the resistance of the heater to be comparable to the resistance of the oscillator itself, which depends on the respective size, shape, and thickness of each device. This allows similar voltages to be applied to the heater and oscillator elements, thereby enabling the system of the oscillator and heater to be controlled more easily. Ti 1-x N x and Ta 1-x N x The resistivity of can vary widely, with the highest resistivity for x to 0.5.

[0020] In such a system, it has been found that the oscillator can be effectively synchronized and induced to operate even in sub-threshold current ranges by controlling the thermal heating element. Different operating modes of the thermal heating element (V-mode / I-mode) enable rich and controllable coupling dynamics between oscillators within the system. In V-mode (heating element), various oscillation states can be generated by these coupled oscillators to realize 12 different basic Boolean logic operations from AND, NAND, and NOR operation gates. Additionally, when the thermal heating element operates in I-mode, it can induce cascade synchronization between at least three coupled correlated oxide oscillators, which generates more complex forms of oscillation states (such as spiking and bursting patterns) that mimic the behavior of biological neurons. Specifically, in this case, it is preferable to use VO2 nanowires.

[0021] Accordingly, the present invention relates to a thermally coupled correlated oxide oscillator system comprising at least two correlated oxide oscillators coupled through at least one heating element that is mutually positioned between at least two of the at least two correlated oxide oscillators.

[0022] The present invention further comprises a logic device comprising a plurality of coupled correlated oxide oscillator(s) / heating element(s) cells, wherein each cell has the dimensions of a 1-column-n-oscillator, where n = 1, 2, 3, 4, 5, 6, 7, 8; preferably n = 2, 3, 4, 5, 6, 7, 8, more preferably n = 2, 3, 4. In principle, the number of coupled oscillators can be extended as long as a column range is reached.

[0023] Correlated oxide oscillators can be fabricated using known deposition processes and photolithography techniques. Using these techniques, a film of correlated oxide material is first deposited on a substrate, for example, a sapphire substrate such as c-plane sapphire. Conventional pulsed laser deposition may be used for this deposition. Typical thicknesses for this deposition of the correlated oxide oscillator range from 10 to 100 nm, preferably from 20 to 50 nm, more preferably from 25 to 40 nm. The fabricated thin film is patterned onto an oscillator device by conventional photolithography techniques, as shown, for example, in FIG. 2b, 2c, 3a, or 4a. If the material of the heating element is the same as one of the correlated oxide materials used in the oscillator, the complete oscillator / heating element pattern can be formed at once. If the material of the heating element is different from one of the correlated oxide materials used in the oscillator, they can be formed individually in a subsequent photolithography patterning step; For example, a VO2 oscillator is formed first, then a heating element is formed, and finally an electrical contact is formed. The shape of the correlated oxide oscillator is not important, but is typically a square or rectangular having a size of 0.0001 to 1000 μm², preferably 0.005 to 50 μm², and more preferably 0.005 to 10 μm². Likewise, the shape of the heating element is not important, but is typically a square or rectangular, preferably a rectangular having a size of 0.0001 to 1000 μm², preferably a rectangular having a size of 0.005 to 50 μm², and more preferably a rectangular having a size of 0.005 to 10 μm².

[0024] The spacing (center-to-center) between the correlated oxide oscillator(s) and the column element(s) is preferably no more than five times the width of a single correlated oxide oscillator (preferably 10 nm to about 30,000 nm), preferably two times, more preferably 1.5 to 1.9 times. The typical thickness for the column element is in the range of 10 to 100 nm, preferably 20 to 50 nm, more preferably 25 to 40 nm. In a logic device comprising a plurality (i.e., more than two) of combined correlated oxide oscillator / column element cells, i.e., an assembly of cells, the spacing between every individual correlated oxide oscillator and every column element is preferably equidistant in the plane of the plurality of combined correlated oxide oscillator / column element cells. An assembly of oxide oscillator / column element cells can be formed, for example, in a cross-bar type (see FIG. 6, 7 and 8). For example, the assembly can be formed as (1) two oscillator-one column element. In this case, one heating element is placed between two oscillators. Another example is (2) an n-oscillator-1-row element. In this case, one heating element is placed at the center of n oscillators. An example of a 4-oscillator-1-row element assembly is shown in FIG. 7. The size of the assembly depends on the number of combined oscillators. - The dimensions of each combined cell (cell spacing, diameter, thickness) are as described above.

[0025] Heating elements can be placed on a layer above or below the oscillator, especially when formed from a different class of material as the nano-oscillators, for ease of manufacturing and for more effective combination of two or more oscillators.

[0026] In a preferred embodiment of an assembly of oscillator / heating element cells, as well as the heating elements, are embedded in a material that is electrically insulating but thermally conductive, as illustrated in Fig. 8. The embedded material may be important to the present application, specifically when it is critical to distinguish between heating and electrical conductivity, that is, to allow better thermal conductivity while blocking electrical connections between the cells. The preferred embedded material may be selected from one or more of, for example, silicon nitride (Si₃N₄), alumina (Al₂O₃), silica (SiO₂), hexagonal boron nitride (h-BN), Teflon, mica, polyvinyl chloride (PVC), and thermal lubricants (e.g., silicone grease).

[0027] The required electrodes can be placed, for example, by using a conventional lift-off method, for example, ion beam deposition. Typical materials for the electrodes are metals or alloys, for example, Ti / Au. The thickness of the electrodes is typically in the range of 10 to 1000 nm, preferably 50 to 100 nm.

[0028] Definition of Terms

[0029] The following definitions of terms apply to terms used in both the singular and plural forms.

[0030] "Combined" and / or "thermally coupled" means that at least two oscillators (or cells) are placed within a range where the input of one oscillator can affect the output of the other oscillator(s) through Joule heating.

[0031] "Correlated oxides" are oxides that exhibit an electrical phase transition between an insulator and a metal. An example of a correlated oxide that can be used in the present invention is Ta2O 5-x (x is preferably between 0 and 1), NbO2, or VO2, preferably VO2.

[0032] An "oscillator" is a system that exhibits an oscillating phase transition between a metallic state and an insulating state through the injection of current or voltage.

[0033] A "heating element" or "thermal element" is an element capable of generating thermal energy through Joule heating.

[0034] "Multiple coupled correlated oxide oscillators / heating elements" refers to a set of coupled correlated oxide oscillators / heating elements arranged next to each other.

[0035] "I-mode" is a type of operation method that uses electric current to operate cells.

[0036] "V-mode" is a type of operation method that uses voltage to operate cells.

[0037] "Logic device" refers to a digital device that generates an output based on a combination of two inputs, either 1 or 0.

[0038] For a single set of two correlated oxide oscillators (two oscillators driven by two independent current sources; with no thermal element between them), the further the oscillators are separated, the weaker the thermal coupling will be. When the two oscillators are far apart, heat emitted from one oscillator will be completely dissipated to the substrate before reaching the other. Furthermore, the coupling strength of the two oscillators is limited, and synchronization can be maintained only within a certain frequency range. To realize adjustable thermal coupling strength, a heating element, preferably a VO2 thermal element, is placed in the middle of the two correlated oxide oscillators, as shown in FIG. 3a. This heating element is designed to alter the surrounding thermal environment between oscillator 1 and oscillator 2. The heating element voltage V that induces Joule heating. cell By applying differently, the ambient temperature between oscillator 1 and oscillator 2 can be changed. Higher heating element voltage V cellThis can generate a higher ambient temperature, which will result in a decrease in the critical switching voltage (as shown in Fig. 3b) and a higher oscillation frequency (as shown in Fig. 3c, under the same applied current). This means that the starting point of the oscillation circle in the RT hysteresis curve (phase transition from high resistance to low resistance and then back to high resistance) is T c This is because it is biased toward a higher temperature closer to . Therefore, enhanced thermal coupling strength (higher V cell As such, Oscillator 1 can be assisted to synchronize with the higher frequency of Oscillator 2. When the heating element is not activated (V cell When V = 0 V), a constant low current I1 (2.4 mA) is supplied to oscillator 1 as shown in Fig. 3d, and it oscillates at a low frequency. By gradually increasing the supply current I2 for oscillator 2, the frequency of oscillator 1 (f1) and the frequency of oscillator 2 (f2) are synchronized up to 23.5 kHz. As f2 increases further, due to the limited thermal coupling strength, f1 first drops to a value approximately half of f2, and then slowly increases while maintaining a ratio of approximately 1:2 (f1:f2) to f2. When the heating element is activated with a very strong thermal coupling strength (as shown in Fig. 3e, V cell (when V = 8), f1 can now be fully synchronized to frequency f2 up to 25.3 kHz. Furthermore, the synchronization frequency already exhibits a sharp increase (from approximately 11 kHz to 14 kHz) at a lower current value I2 (approx. 2.2 mA). Weak / Medium (V cell < 8 V) and strong (V cell = 8 V) A significant difference in the synchronization behavior of oscillator 1 in the coupling effect can be observed in Fig. 3f.

[0039] Implementation of Boolean logic gates based on coupled oscillators

[0040] Based on the adjustable thermal coupling mechanism discussed above, three different Boolean logic gates (AND, NAND, and NOR) are realized by the set of devices shown in FIG. 3a. The oscillation state of the oscillator, expressed by the oscillation frequency (f) and amplitude (A), is taken as logic "0" or "1". The threshold frequency f th is while I2 gradually increases while I1 is fixed at 2.5mA (f th = 23.5kHz is taken for the following calculation), V cell = is defined as the frequency at which Oscillator 1 and Oscillator 2 desynchronize at 0V. f th Lower frequencies are defined as 'low frequencies', and f th Higher frequencies are defined as 'high frequencies'. Peak-to-peak value V pk-pk = 1 V is selected as the critical amplitude. V pk-pk An amplitude smaller than is defined as "small amplitude," and V pk-pk A larger amplitude is defined as "large amplitude".

[0041] Basically, there are four oscillation states: high frequency with large amplitude, high frequency with small amplitude, low frequency with large amplitude, and low frequency with small amplitude. Here, only the high frequency and high amplitude oscillation state is set to logic "1". The other states are taken as logic "0".

[0042] As shown in FIG. 4a, the oscillation state of oscillator 1 under different given current values ​​(I2) is taken as input A, while the heating element voltage V cell The value is expressed as input B (V cell = 0 V represents B = 0, and V cell = 7 V indicates B = 1). While the oscillation state of oscillator 1 under input A and input B is taken as the output, the current value (I1) of oscillator 1 is kept constant during each operation.

[0043] In the AND operation, I1 is fixed at 2.5mA, while I2 = 5mA is taken as input logic "1" and I2 = 2.5mA is taken as input logic "0". In the NAND operation, I1 is fixed at 6.2mA, while I2 = 5mA is taken as input logic "1" and I2 = 2.4mA is taken as input logic "0". In the NOR operation, I1 is fixed at 6.4mA, while I2 = 5.2mA is taken as input logic "1" and I2 = 2.5mA is taken as input logic "0". As demonstrated in the diagram, the output logic "0" (indicated by the red curve) from the oscillator (2) can be represented by four different oscillation states:

[0044] Low frequency with large amplitude (Fig. 4b); or high frequency with small amplitude (Fig. 4c); or low frequency with small amplitude / no oscillation (Fig. 4d).

[0045] Control of firing modes in a coupled oscillator

[0046] Unlike operating the thermal heating element in V-mode, a constant current (I) to the thermal heating element cell ), that is, applying the I-mode can form cascade synchronization between the correlated oxide oscillator and the thermal heating element. In this case, the thermal heating element itself must consist of the oscillator itself; that is, the thermal heating element is preferably made of a correlated oxide, for example, VO2. This phenomenon occurs because the thermal heating element now also oscillates (I within the NDR region). cell (along with), this occurs because the heat periodically emitted therefrom strongly couples the oscillations of Oscillator 1 and Oscillator 2 to their own cycles. As shown in Fig. 5a, Oscillator 1 (Cell 1), Oscillator 2 (Cell 2), and the thermal heating element (thermal cell) are activated by three independent current sources: I1 = 4 mA, I2 = 3.9 mA, and I cellWhen = 2.3 mA, frequency synchronization between the three devices can be observed in Fig. 5b.

[0047] Conceptually, the output waveform of oscillator 2 is similar to the spiking potential when a neuron is stimulated. Neurons have rich and complex responses to external stimuli in biological systems. Based on different responses in neurons, various spiking neuron models have been established, such as rigid spiking and bursting, and phase spiking and bursting

[31] .

[0048] As illustrated in Fig. 5c, four typical regions can be identified when a neuron transitions from a resting state to an activated state [7, 32-34]: Region I corresponds to the resting state when the neuron is not activated and its potential remains at the resting voltage level. Region II is the time when a stimulus arrives and causes the neuron potential to rise. Region III corresponds to the depolarization state where the potential exceeds a threshold and emits a spike. Region IV is the repolarization and hyperpolarization state (known as the refractory period) where the potential returns to the resting state. The currents of oscillator 2 and the heating element are kept constant (I2 = 4 mA, I cell By simply adjusting the current of oscillator 1 (between 1 mA and 5.2 mA) while maintaining it at 2.3 mA, different numbers of spikes (5 different neuron firing modes) can be generated from oscillator 2 within a single firing period, as shown in FIG. 5d. Among these 5 firing modes, mode 2 can be considered as a tonic spiking neuron that fires a spike once within a cycle, and modes 3 and 5 can be classified as tonic bursting neurons that periodically fire bursts of multiple spikes [31, 35-39].

[0049] Discussion and Outlook

[0050] The present invention experimentally demonstrated and described the tunable thermal coupling effect between two closely correlated oxide oscillators without external electronic components such as resistors, capacitors, or inductors, as previously discussed [9-16]. By changing the voltage applied to a thermal heating element located between the two oscillators, the thermal coupling strength can be enhanced, and the range of the synchronization frequency can be effectively extended. A Boolean-type calculation method was introduced by setting thresholds in frequency and amplitude based on different oscillation states generated from the coupled oscillators. The application of these synchronized oscillators with tunable thermal coupling strength to AND, NAND, and NOR logic gates was demonstrated.

[0051] The heating element can also provide adjustable thermal coupling between two or more oscillator elements. For example, the heating element may couple three, four, or more oscillators by placing the element next to three, four, or more oscillators. It may also be desirable to provide asymmetric adjustable thermal coupling by placing the element at different distances from the coupled elements.

[0052] This thermally assisted frequency synchronization process can also be regarded as analogous to signal propagation in the nervous system. Action potentials combined with information modulated by frequency (or amplitude, or both) are propagated through synapses from a presynaptic neuron to a postsynaptic neuron [40-44]. In this process, action potentials are transmitted through the space between synapses, and neurotransmitters are released from the axon of the preceding neuron to the dendrite of the next neuron

[45] . In Fig. 4a, Oscillator 1 (Cell 1) acts as a presynaptic neuron, and Oscillator 2 (Cell 2) acts as a postsynaptic neuron, respectively. These two neurons communicate through oscillating potentials, much like neurons in biological systems. The thermal heating element is equivalent to a synapse. The postsynaptic neuron (Oscillator 2) will output a low-frequency signal only when supplied with a constant low current and thus not synchronized with the presynaptic neuron (Oscillator 1). Information is V between Oscillator 1 and Oscillator 2 cell Encoded in a domain above the threshold frequency desynchronizing at 0 V, and supplied to the presynaptic neuron (oscillator 1) as a continuously varying current (frequency modulation). V cell At = 0 V, there is no synchronization between neurons in the frequency range where information is encoded.

[0053] As a result, all information encoded at the modulated frequency in the presynaptic neuron (oscillator 1) is lost as it propagates to the postsynaptic neuron (oscillator 2). V cell With ≠ 0 V, this means that a neural connection is established and the two neurons are synchronized, and information encoded in frequency can be transmitted to the postsynaptic neuron. This mechanism can also be applied for frequency-modulated data transmission for non-Boolean operations in analog networks [46, 47].

[0054] In addition, cascade synchronization between three oscillators has been demonstrated. By simply changing the current supplied to one of the coupled correlated oxide oscillators, five different firing modes, such as spiking and bursting, can be generated from one of the coupled oscillators. These current-driven firing behaviors can be used to code an artificial spiking neural network, where the output spikes (number and firing time) depend on the intensity and spatiotemporal distribution of the input signal

[48] .

[0055] A set of thermally coupled oscillators capable of generating multiple synchronized states is suitable for building oscillator computation networks that require dynamic richness and coordination. This technique is necessary to solve different types of computationally difficult problems, such as image processing and combinatorial problems [49-54].

[0056] Experiment Details

[0057] Sample and device preparation

[0058] A VO2 film with a thickness of approximately 30 nm (determined by XRR measurements) was deposited on a c-plane sapphire

[0001] substrate by pulsed laser deposition (PLD) technology. A KrF excimer laser (Coherent LPX pro) beam with a frequency of 3 Hz was focused on a VO2 target (99.9% purity, plasma material) in a test vessel at 450°C under ambient O2 of 0.020 mbar. The energy and influence of the laser beam at the target surface were 44 mJ and 587 mJ / cm2, respectively. After deposition, the sample was cooled under ambient O2 of 0.045 mbar. Subsequently, the fabricated thin film was patterned into a device pattern by conventional photolithography technology. A negative-tone photosensitive resin (ARN-4340, all-resistive) and a maskless aligner (Heidelberg MLA 150) were used. Afterwards, a lift-off method is used to place the electrode. For the contact, approximately 77 nm Ti / Au was deposited via ion beam deposition.

[0059] Electrical transport measurement

[0060] To measure resistance as a function of temperature, a physical properties measurement system (PPMS) was used in conjunction with a conventional 4-point resistance measurement scheme with a cooling and heating rate of 5 K / min. For the 4-point measurement, a 1 μA DC current (Keithley 6221 AC / DC source and 2182A nanovoltmeter set) was applied to the sample to minimize the effect of Joule heating.

[0061] To measure the oscillation state using various types of devices, measurements were performed at a multi-probe cryogenic probe station (Lakeshore). A current source (Keithley 6221 AC / DC current source) and a source meter (Keithley 2636B) were used to drive the oscillator and heating element, and the oscillation voltage was collected by an oscilloscope (DSO5052A, InfiniVision) with a home-built LV code.

[0062] References

[0063] [1] Asmara, T., Wan, D., Zhao, Y. et al. Tunable and low-loss correlated plasmons in Mott-type insulating oxides. Nat Commun 8, 15271 (2017).

[0064] [2] Gao, L., Chen, P. and Yu, S. NbOx-based oscillating neuron for neuromorphic computing. Appl. Phys. Lett. 4, 103503 (2017).

[0065] [3] McGee, R. et al. Modulation of clarity and intensity of metal-insulator transitions in ultrathin VO2 films by interfacial structure manipulation. Phys. Rev. Mat. 2, 034605 (2018).

[0066] [4] Yang, T. et al. Growth of epitaxial vanadium oxide thin films on sapphire (0001). J. Mater. Res. 25, 422-426 (2010).

[0067] [5] Hattori, AN et al. Investigation of the statistical metal-insulator transition properties of electronic domains in spatially confined VO2 nanostructures. Crystal 10, 631 (2020).

[0068] [6] Jeong, J. et al. Suppression of metal-insulator transition in VO2 by oxygen vacancy formation by an electric field. Science 339, 1402-1405 (2013).

[0069] [7] Yi, W. et al. Biological validity and probabilistics in scalable VO2-activated memristor neurons. Nat. Commun. 9, 4661 (2018).

[0070] [8] Shukla, N., et al. Pairwise coupled hybrid vanadium dioxide-MOSFET (HVFET) oscillators for non-Boolean associative computing. IEDM 2014, San Francisco, CA, USA (2014).

[0071] [9] Corti, E., Gotsmann, B., Moselund, K., Stolichnov, I., Ionescu, A. and Karg, S. VO2 oscillator for resistance-coupled image recognition. ICRC, McLean, VA, USA (2018).

[0072]

[10] Corti, E., Gotsmann, B., Moselund, K., Stolichnov, I., Ionescu, A. and Karg, S., Robertson, J. scaled a resistively coupled VO2 oscillator for neuromorphic computing. Solid State Electronics. Lett. 168, 107729 (2020).

[0073]

[11] Corti, E., Gotsmann, B., Moselund, K., Stolichnov, I., Karg, S., Robertson, J. Time-delay encoded image recognition is encoded in a network of VO₂ resistively coupled to a Si oscillator. IEEE Electronic Devices Lett. 41, 629-632 (2020).

[0074]

[12] Corti, E. et al. incorporated a VO2 oscillator circuit as an analog first-layer filter in a convolutional neural network. Front. Neurosci 15, 628254 (2021).

[0075]

[13] Carapezzi, S. et al. Multiscale modeling and simulation flow of oscillatory neural networks for edge computing. 2021 NEWCAS, Toulon, France (2021).

[0076]

[14] Todri-Sanial, A. et al. A method for training an oscillating neural network to operate in phase using frequency injection synchronization techniques. IEEE Trans Neural Network Learn Syst. 33, 1996-2009 (2022).

[0077]

[15] Delacour, C. et al. Oscillatory neural networks for edge AI computing. ISVLSI, Tampa, FL, USA (2021).

[0078]

[16] N ez, J. et al. Oscillating neural network using VO2-based topologically encoded logic. Front. Neurosci. 15, 655823 (2021).

[0079]

[17] Lee, Y. et al. Metal-insulator transition-induced vanadium dioxide thin film electrooscillation. Appl. Phys. Lett. 92, 162903 (2008).

[0080]

[18] Veliko, A., Belyayev, M., Putrolaynen, V., Perminov, V. and Pergament, A. Modeling of thermal coupling in VO2-based oscillator neural networks. Solid State Electronics. Lett. 139, 8 (2018).

[0081]

[19] Veliko, A., Belyayev, M. and Boriskov, P. Model of an oscillatory neural network with multi-level neurons for pattern recognition and computing. E.T. 8, 75 (2019).

[0082]

[20] Veliko, A., Belyayev, M., Putrolayenen, V., Perminov, V. and Pergament, A. Effects of subharmonic synchronization in a system of thermal coupling and two VO2-based oscillators. Solid State Electronics. 141, 40-49 (2018).

[0083]

[31] Izhikevich, Model for use in EM cortical spiking neurons. IEEE Trans. Neural Netw. 15, 1063-1070 (2004).

[0084]

[32] Korson, N., Aziz-Al-Laoi, M. Hindmarsh-Rose Asymptotic dynamics of neuronal systems. Dynamics of continuous, discrete and impulsive systems Series B: Applications and Algorithms 16, 535 (2009).

[0085]

[33] Hille, B. Ionic channels of excitable membranes, Sinauer associates Inc. (1992).

[0086]

[34] Gerstner, W., Kistler, WM, Naud, R. and Paninski, L. Neuronal Dynamics: From Single Neuron to Networks and Cognitive Models. University of Cambridge Press (2014).

[0087]

[35] Agmon, A. and Connors, B. Repetitive burst-firing neurons in the deep layer of the somatosensory cortex of mice. Neurosis. Lett. 99, 137-141 (1989).

[0088]

[36] McCormick, DA, Conos, BW, Lighthall, JW, and Prince, DA Comparative electrophysiology of pyramidal and sparse spin astrocytes in the neocortex. J. Neurophytol 54, 782-806 (1985).

[0089]

[37] Gibson, JR, Beierlein, M. and Connors, BW Two networks of electrically coupled inhibitory neurons in the neocortex. Nature. 402, 75 (1999).

[0090]

[38] Gray, CM and McCormick, DA. Chatty cells: Superficial pyramidal neurons contributing to the development of synchronous oscillations in the visual cortex. Science. 274, 109-113 (1996).

[0091]

[39] Brumberg, JC, Nowak, LG and McCormick, DA. The ion mechanism underlies repetitive high-frequency burst firing in pyogenic cortical neurons. J. Neurosis 20, 4829-4843(2000).

[0092]

[40] Frequency modulation of large oscillator neural networks by Weifels, F., Li, J., Waegeman, T. et al. Viol Cyberen 108, 145-157 (2014).

[0093]

[41] Wyffels, F. and Schrauwen, B. Design of a central pattern generator using repository computing to learn human motion. 2009 High Technology for Improved Quality of Life, Iasi, Romania (2009).

[0094]

[42] Soman, K., Muralidharan, V. and Chakravarthy, VS A Oscillatory neural autoencoder based on frequency modulation and multiplexing. Front. Neurosis. 12,00052 (2018).

[0095]

[43] D. Cumin and CP Unsworth, “Generalization of the Kuramoto model for neural synchronization studies in the brain”, Phys. D Nonlin. Physics 226, 181-196(2006).

[0096]

[44] Hoppensted, FC and Izkevich, EM thalamus-cortical interactions modeled by weakly coupled oscillators: the brain can use FM radio principles. Biosystems. 48, 85-94 (1998).

[0097]

[45] Faber, DS and Peredda, two forms of electrical transmission between AE neurons. Front. Mole. Neurosis. 11,00427 (2018).

[0098]

[46] Guy, CG Analog Data Transmission. Data Communication for Engineers Macmillan New Electronic Series. Palgrave, London (1992).

[0099]

[47] Smith, DR Digital Transmission System over Analog Network. Springer, Boston (2004)

[0100]

[48] ​​Yu, Q., Tang, H., Hu, J., Tan, KC Rapid feedforward computation by time encoding and learning with spiking neurons. Neuromorphic Cognitive Systems, 19-41. Springer, Berlin (2017).

[0101]

[49] Csaba, G., Porod, W. Combined Oscillators for Computing: A Review and Perspective. June 7, 2020 Phys. Rev. 7.

[0102]

[50] Heger, D. and Krischer, K. Powerful auto-associative memory with a coupled network of Kuramoto-type oscillators. Revision E. 94(2), 022309 (2016).

[0103]

[51] Vodenicarevic, D., Locatelli, N., Grollier, J. and Querlioz, D. Synchronization detection in a network of combined oscillators for pattern recognition. IJCNN, Vancouver, BC, Canada (2016).

[0104]

[52] Smith, KA Neural Networks for Combinatorial Optimization: A Review of Over 10 Years of Research. Notified in J. Comput. 15-34(1999)

[0105]

[53] Graph coloring through synchronization of Wu, CW combined oscillators. IEEE Trans. Circuits Syst. I. 45(9), 974-978 (1998).

[0106]

[54] Wu, J., Jiao, L., Li, R. and Chen, W. Clustering dynamics of nonlinear oscillator networks: application to the graph coloring problem. Physica D 240 (24), 1972-1978 (2011).

Claims

Claim 1 A thermally coupled correlated oxide oscillator system comprising at least two correlated oxide oscillators coupled through at least one heating element, wherein the heating element is positioned between at least two of the at least two correlated oxide oscillators. Claim 2 In claim 1, the correlated oxide is Ta2O 5-x A thermally coupled correlated oxide oscillator system selected from one or more of , NbO2, or VO2. Claim 3 A thermally coupled correlated oxide oscillator system according to claim 1 or 2, wherein the heating element is composed of Fe, Cu, Co, Ni, W, V, or an alloy of two or more of these metals. Claim 4 In claim 1 or 2, the heating element is Ti 1-x N x titanium nitride or Ta 1-x N x A thermally coupled correlated oxide oscillator system characterized by being composed of tantalum nitride. Claim 5 A thermally coupled correlated oxide oscillator system according to claim 1 or 2, wherein the heating element is composed of VO2. Claim 6 A thermally coupled correlated oxide oscillator system characterized in that, in any one of claims 3 to 5, the heating element is a wire. Claim 7 A thermally coupled correlated oxide oscillator system according to claim 6, wherein the heating element is a nanowire. Claim 8 A thermally coupled correlated oxide oscillator system according to any one of claims 1 to 7, wherein the correlated oxide has a thickness of 10 to 100 nm. Claim 9 A thermally coupled correlated oxide oscillator system according to any one of claims 1 to 8, wherein the shape of the correlated oxide oscillator is square or rectangular. Claim 10 A thermally coupled correlated oxide oscillator system according to claim 9, wherein the size of the correlated oxide oscillator is 0.0001 to 1000 μm². Claim 11 A thermally coupled correlated oxide oscillator system according to any one of claims 1 to 10, wherein the center-to-center spacing between the correlated oxide oscillators and the thermal element or heating element is not greater than five times the width of a single correlated oxide oscillator. Claim 12 A method for manufacturing a thermally coupled correlated oxide oscillator system according to claim 1, wherein the method comprises the steps of: depositing a film of the correlated oxide material on a substrate; patterning an oscillator / heating element pattern using photolithography techniques when the material of the heating element is the same as the correlated oxide material used in the oscillator; or, when the material of the heating element is different from the correlated oxide material used in the oscillator, individually patterning an oscillator pattern and a heating element pattern through subsequent lithography patterning steps; and then forming a thermal element and finally forming an electrical contact. Claim 13 A logic device comprising a plurality of thermally coupled correlated oxide oscillator systems according to claim 1. Claim 14 A logic device according to claim 13, wherein the plurality of thermally coupled correlated oxide oscillator systems form cells, each cell having the dimensions of a 1-column-n-oscillator, and n = 1, 2, 3, 4, 5, 6, 7 or 8. Claim 15 A neural network comprising a plurality of thermally coupled correlated oxide oscillator systems according to claim 1.

Images