New phase-change electrically-controlled photonic neuron device, and preparation method therefor and use thereof

Abstract

The present invention belongs to the technical field related to photoelectrons, and discloses a new phase-change electrically-controlled photonic neuron device, and a preparation method therefor and the use thereof. The device comprises a substrate, a waveguide layer, a phase-change material layer and a heating layer. The general chemical formula of the phase-change material layer is SbxTey, where x+y=1 and 0.6<y<1. The phase-change material layer has an Sb2Te3 phase and a Te phase, wherein and the Sb2Te3 phase is embedded in the Te phase to form an island-shaped structure. When the temperature of the heating layer rises, the Te phase takes on an amorphous state or a molten state from a crystalline state under the thermal induction action of the heating layer, whereas the Sb2Te3 phase maintains its crystalline state; and when the temperature is reduced, the Te phase spontaneously crystallizes at a low temperature, with the crystalline Sb2Te3 phase being taken as a nucleation site, such that rapid and stable crystallization of the phase-change material layer can be promoted, and the function of neurons spontaneously returning to a stable resting potential is achieved.

Description

A novel phase-change electrically controlled photonic neuron device and a preparation method and application thereof

TECHNICAL FIELD

[0001] The present application belongs to the technical field related to optoelectronic technology, and more particularly to a novel phase-change electrically controlled photonic neuron device and a preparation method and application thereof.

BACKGROUND

[0002] Neuromorphic computing simulates the computing mode of the human brain neural network, and is a new computing paradigm of storage and computing integration. The photonic neural network can greatly improve the running efficiency of the system by virtue of the advantages of low power consumption, low delay and super-high parallelism of photons, and has great application potential. The photonic neural network is composed of photonic synapse devices and photonic neuron devices. The neuron device is an important component in the photonic neural network, which is used to integrate the signals of synapses and generate an action potential. The implementation of the photonic neuron device usually includes on-chip and off-chip two ways. The on-chip integrated device can be compatible with other photonic active / passive devices, build an on-chip computing system, and realize more complex functions. Compared with off-chip, it has the advantages of small effective size, high scalability, fast computing speed, and compatibility with electronic CMOS process, and has attracted widespread attention.

[0003] Traditional on-chip integrated passive photonic devices mainly rely on thermal light effect and carrier dispersion effect for adjustment, and the change of refractive index is limited, which requires a large size to realize the change of optical transmittance, and the control loop is complex, which has high energy consumption. The hetero-hybrid integrated waveguide based on phase-change material shows great advantages in constructing photonic devices. Phase-change materials have the advantages of fast phase change, high optical contrast, and easy integration. However, the phase-change materials currently used are mainly Ge2Sb2Te5 ternary and its doped system. The photonic neuron device constructed by them has non-volatility and cannot restore to the resting potential by itself. It needs a feedback loop to control the neuron to return to the resting potential, which increases the complexity and scalability of the photonic neural network. Although a design of a volatile neuron material is mentioned in Chinese patent application CN115034377A, only a specific embodiment of a single-element Te neuron is provided, and the resting potential fluctuation problem exists.

[0004] Therefore, it is a technical problem to be solved at present to design a volatile neuron device that can stably return to the resting potential.

[0005]

SUMMARY

[0006] In view of the above defects or improvement needs of the prior art, the present application provides a novel phase-change electrically controlled photonic neuron device and a preparation method and application thereof, which aims to make the neuron device have volatility and can stably return to the resting potential.

[0007] To achieve the above object, according to one aspect of the present application, a novel phase-change electrically-controlled photonic neuron device is provided, which comprises a substrate, a waveguide layer on the substrate, a phase-change material layer on the waveguide layer, and a heating layer for heating the phase-change material layer;

[0008] wherein the phase-change material layer has a chemical formula of Sb x Te y , x+y=1, 0.6

[0009] When the temperature of the heating layer is raised to be greater than the state transition threshold of the Te phase, the Te phase changes from a crystalline state to an amorphous state or a molten state, and the Sb2Te3 phase maintains a crystalline state; when the temperature of the heating layer is lowered to be less than the state transition threshold of the Te phase, the Te phase changes from an amorphous state or a molten state to a crystalline state with the crystalline Sb2Te3 phase as a nucleation site.

[0010] Further, the heating layer is selected from any one of ITO, In2O3, AZO, and TiO2.

[0011] Further, the device further comprises an electrode layer in contact with the heating layer, and the electrode layer is used for connecting an external power supply and passing current to the heating layer to adjust the temperature of the heating layer.

[0012] Further, the electrode layer is one or a combination of Ti, Cr, Pt, Au, Al, and W.

[0013] The present application also provides a preparation method of the novel phase-change electrically-controlled photonic neuron device, which comprises:

[0014] forming a waveguide layer on a substrate;

[0015] forming a phase-change material initial layer on the waveguide layer, and the phase-change material initial layer has a chemical formula of Sb x Te y , x+y=1, 0.6

[0016] forming a heating layer on the phase-change material initial layer;

[0017] forming an electrode layer on both ends of the heating layer;

[0018] applying a pre-treatment pulse voltage to the electrode layer, and heating the phase-change material initial layer by the heating layer to combine Sb atoms and Te atoms in the phase-change material initial layer to form a stable Sb2Te3 phase, and the Sb2Te3 phase is embedded in the Te phase and forms an island structure, thereby obtaining a phase-change material layer.

[0019] Further, the voltage amplitude of the pre-treatment pulse voltage is 0.5V-5V, and the pulse width is 50us-800us.

[0020] The application also provides an on-chip photonic neural network system comprising a nonlinear activation layer and a regulating unit, the nonlinear activation layer being the novel phase-change electrically-controlled photonic neuron device as described above, and the regulating unit being used to apply a voltage pulse to the electrode layer so that the transmittance of the phase-change material layer changes nonlinearly with the size of the voltage pulse.

[0021] Further, the regulating unit is used to regulate the amplitude of the voltage pulse to regulate the light transmittance of the phase-change material layer.

[0022] The application also provides an on-chip photonic neural network system comprising a probability activation layer and a regulating unit, the probability calculation layer being the novel phase-change electrically-controlled photonic neuron device as described above, and the regulating unit being used to apply a voltage pulse to the electrode layer so that the transmittance of the phase-change material layer changes randomly with the size of the voltage pulse.

[0023] Further, the regulating unit is used to regulate the width of the voltage pulse to regulate the probability of random light transmittance of the phase-change material layer.

[0024] Overall, compared with the prior art, the above technical solutions conceived by the application mainly have the following beneficial effects:

[0025] The novel phase-change electrically-controlled photonic neuron device provided by the application is provided with a phase-change material layer, the phase-change material layer adopts two elements of Sb and Te, two phases with different physical properties appear in the phase-change material layer during preparation, which are Sb2Te3 phase and Te phase, the Sb2Te3 phase is embedded in the Te phase to form an island structure, the proportion of the two phases is regulated by regulating the proportion of the two elements of Sb and Te, the phase state transition threshold of the Te phase is lower than that of the Sb2Te3 phase, at normal temperature, the Sb2Te3 phase and the Te phase are both in a crystal state, when the temperature of the heating layer is increased to be greater than the phase state transition threshold of the Te phase, under the heat induction of the heating layer, the Te phase changes from a crystal state to an amorphous state or a molten state, while the Sb2Te3 phase maintains a crystal state; when the temperature is reduced, the Te phase spontaneously crystallizes at low temperature, produces a volatile change, and during the crystallization, an appropriate amount of crystal Sb2Te3 phase is used as a nucleation site, which can promote the phase-change material layer to crystallize rapidly and stably. The difference in optical constants before and after the phase change of the material causes the transmittance of light in the waveguide from the phase-change material layer to change volatily, and the function of the neuron spontaneously returning to a stable resting potential is realized. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a schematic diagram of the structure of a novel phase-change electrically controlled photonic neuron device according to an embodiment of the present invention.

[0027] Figure 2 is an optical response diagram of a neuron device in one embodiment of the present invention.

[0028] Figure 3 is a flowchart of the fabrication method of a novel phase-change electrically controlled photonic neuron device according to an embodiment of the present invention.

[0029] Figure 4 shows the response of a neuron device to an optical signal in one embodiment, before and after heating pretreatment. Figure a shows the response of the neuron device to an optical signal before heating pretreatment, and Figure b shows the response of the neuron device to an optical signal after heating pretreatment.

[0030] Figure 5 shows the responses of four different neuronal devices to optical signals after heat pretreatment. Figure a shows the response of the SbTe7 (y = 0.875) device to optical signals, Figure b shows the response of the Sb2Te3 (y = 0.6) device to optical signals, and Figure c shows the response of the Te device (y = 1) to optical signals.

[0031] Figure 6 is a schematic diagram of the probabilistic response of a neuron implemented by a neuron device in one embodiment of this application.

Detailed Implementation Methods

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0033] Figure 1 shows a schematic diagram of the structure of a novel phase-change electrically controlled photonic neuron device according to an embodiment of the present invention. It includes a substrate, a waveguide layer on the substrate, a phase-change material layer on the waveguide layer, and a heating layer for heating the phase-change material layer. The phase-change material layer has the general chemical formula Sb. x Te y, x+y=1, 0.6

[0034] As shown in FIG. 2, it is an optical response diagram of the neuron device in an embodiment of the present application. After the electric pulse is applied, the heat is conducted to the phase change material layer, the light transmission is changed, and the neuron can spontaneously return to the resting potential, showing the volatile change.

[0035] In the embodiment, the temperature of the heating layer is regulated by applying electric energy. Specifically, the device further comprises an electrode layer in contact with the heating layer, the electrode layer is used to connect an external power supply and pass electric current to the heating layer to regulate the temperature of the heating layer. The two electrodes have no polarity and are used to apply an electric signal to the device. The external electric field is input into the device through one electrode, flows through the heating layer, and is output from the device through the other electrode, forming an electric circuit. The heating layer has a certain resistance, and when the electric current flows through the heating layer, the Joule heating effect is generated, the heat is conducted to the phase change material layer, and the phase transition of the material is induced. By controlling the electric pulse signal, the temperature of the phase change material layer can be controlled, thereby controlling the phase state of the material and further affecting the firing of the neuron.

[0036] Specifically, the shapes of the phase change material layer, the heating layer, and the electrode layer can be determined according to the heating efficiency and the performance requirements of the neuron, and can be one or a combination of rectangular, trapezoidal, triangular, circular, and conical.

[0037] Specifically, the heating layer can be made of an optically transparent material, such as one of ITO, In2O3, AZO, and TiO2, to reduce the interference with the light transmittance. Specifically, the thickness of the heating layer can be selected to be 50-500 nm. The electrode layer is made of a material with good conductivity, such as one or a combination of Ti, Cr, Pt, Au, Al, and W, to conduct electricity. Specifically, the thickness of the electrode layer can be selected to be 50-300 nm.

[0038] Further, isolation layers can be introduced above and below the phase change material layer to enhance the heat transfer effect of the heating layer or limit the volume expansion during the phase change of the material. The isolation layer material is one of SiO2 and Al2O3.

[0039] In specific embodiments, the thickness of the phase change material layer can be selected to be 5-100 nm, and the width can be selected to be 1-20 pm.

[0040] As shown in FIG. 3 is a step flow chart of a preparation method of a novel phase change electrically controlled photonic neuron device in an embodiment of the present application, for preparing the novel phase change electrically controlled photonic neuron device introduced above, mainly including steps S1 to S5, which will be introduced in detail below.

[0041] Step S1: forming a waveguide layer on a substrate.

[0042] Specifically, Klayout software is used to draw a waveguide and grating pattern, a 1 cm x 1 cm SOI silicon wafer is spin-coated with photoresist, developed and fixed after electron beam lithography (EBL), to obtain a waveguide and grating pattern window, etched by inductively coupled plasma (ICP) for 220 nm, and the photoresist is removed to obtain a waveguide.

[0043] Step S2: forming a phase change material initial layer on the waveguide layer, the chemical formula of the phase change material initial layer is Sb x Te y , x+y = 1, 0.6 < y < 1.

[0044] Specifically, Klayout software is used to draw a material size pattern, S1 silicon wafer is spin-coated with photoresist, developed after ultraviolet light exposure, to obtain a material pattern window, a phase change material initial layer is prepared by magnetron sputtering method, and then the photoresist is stripped using acetone and ethanol to obtain the phase change material initial layer. At this time, Sb atoms and Te atoms have not yet combined.

[0045] Step S3: forming a heating layer on the phase change material initial layer.

[0046] Specifically, Klayout software is used to draw a heating layer size pattern, the heating layer needs to completely cover the phase change material layer, S2 silicon wafer is spin-coated with photoresist, developed after ultraviolet light exposure, to obtain a heating layer pattern window, a heating layer material is prepared by magnetron sputtering method, and then the photoresist is stripped using acetone and ethanol to obtain the heating layer.

[0047] Step S4: forming an electrode layer on both ends of the heating layer.

[0048] Specifically, Klayout software is used to draw an electrode layer size pattern, the electrode layer is located at both ends of the heating layer pattern and cannot contact the waveguide, S3 silicon wafer is spin-coated with photoresist, developed after ultraviolet light exposure, to obtain an electrode layer pattern window, an electrode layer material is prepared by magnetron sputtering method, and then the photoresist is stripped using acetone and ethanol to obtain the electrode layer.

[0049] Step S5: a pre-treatment pulse voltage is applied to the electrode layer, and the initial phase change material layer is heated by the heating layer to combine Sb atoms and Te atoms in the initial phase change material layer to form a stable Sb2Te3 phase, the Sb2Te3 phase is embedded in the Te phase and forms an island structure, and a phase change material layer is obtained.

[0050] According to the conventional method, the phase change layer obtained by sputtering is generally directly used and a neural network is introduced, but experiments show that the neuron device obtained at this time cannot stably realize switching, and the neuron cannot return to a stable resting potential. After analysis, it is found that the crystal structure formed after crystallization each time is very different because the phase change region, the grain growth orientation, and the growth speed during the phase change process are not completely the same, so the neuron cannot stably return to the resting potential. In the present application, the initial phase change material layer obtained by sputtering is heated and pre-treated in step S5 to combine Sb atoms and Te atoms to form a stable Sb2Te3 phase, the Sb2Te3 phase is embedded in the Te phase and forms an island structure, which provides nucleation sites for the crystallization process of Te, and the crystallization is preferentially performed on the grain edges, thereby regulating the crystallization orientation and grain size of Te grains, promoting the rapid and stable crystallization of the phase change material layer, and realizing the function of the neuron spontaneously returning to the stable resting potential. Specifically, the Sb atoms are generally uniformly and dispersedly distributed, and therefore, the island structure after pre-treatment is also uniformly distributed.

[0051] As shown in FIG. 4, the response of the neuron device in an embodiment to a light signal before and after heating pre-treatment is shown, wherein a diagram is the response of the neuron device to the light signal without heating pre-treatment, and b diagram is the response of the neuron device to the light signal after heating pre-treatment. As can be seen, the device without pre-treatment has free Sb atoms and Te atoms, which can produce volatile changes, but the switching is unstable, and after pre-treatment, the Sb2Te3 phase is embedded in the Te phase and forms an island structure, and the switching of the neuron device is more stable.

[0052] For an on-chip structure, the entire neural network chip cannot be heated, and the present application heats the initial phase change material layer by applying a low-energy long-time electric pulse to combine Sb atoms and Te atoms to form a stable Sb2Te3 phase without affecting other structures. Specifically, the voltage amplitude is 0.5-5V, and the pulse width is 50-800μs, so that the heating layer heats the phase change material to a certain temperature for a long time without phase change, thereby pre-treating the neuron device. The purpose is to make the Sb in the phase change material layer form an island-shaped Sb2Te3 phase by bonding with the surrounding atoms, and then the obtained photonic neuron device can be used for a neural network.

[0053] Figure 5 shows the responses of three different neuron devices to light signals after heating pretreatment, wherein a is the response of the SbTe7 (y = 0.875) device to light signals, b is the response of the Sb2Te3 device (y = 0.6) to light signals, and c is the response of the Te device (y = 1) to light signals, from which it can be seen that the SbTe7 device has stable and volatile optical response performance; the Sb2Te3 device has a non-volatile switch and cannot meet the function of the neuron device; the Te device has a volatile switching response but the switch is unstable, that is, the Te device presents volatile changes under the action of an electric field, but due to the unstable crystallization of Te, it cannot stably and volatily return to the same resting potential.

[0054] Correspondingly, the application also relates to an on-chip photonic neural network system, which comprises a nonlinear activation layer and a regulation unit, the nonlinear activation layer is the novel phase-change electrically-controlled photonic neuron device with an electrode layer introduced above, and the regulation unit is used for applying a voltage pulse to the electrode layer, and the transmittance of the phase-change material layer changes nonlinearly with the size of the voltage pulse.

[0055] The neuron device obtained by the method introduced above needs to provide heat energy for the heating layer material through an external electric field excitation to drive the phase-change material layer to occur phase transition when performing phase modulation, which is essentially a direct heat-induced phase change of the heating layer, but the heat is affected by the electric field, and different electric pulse parameters will affect the generation and distribution of heat.

[0056] In the neural network, the nonlinear activation layer realizes a nonlinear function, which is a relatively key structure, and the optical nonlinear function is expressed by the light transmittance of the neuron device, which changes nonlinearly with the regulation parameters under different regulation parameters, and the adjustable parameters include a voltage amplitude, a pulse width, a pulse period, a duty cycle, a rising / falling edge, a pulse waveform and a pulse number. In the embodiment, the neuron device realizes a nonlinear activation function by adjusting the voltage amplitude. Under different voltages, the phase-change material reaches different degrees of phase transition, and then responds to light signals differently, has different optical contrasts, and changes nonlinearly with the pulse width, which can be used for realizing the nonlinear activation layer in the neural network.

[0057] Correspondingly, the application also relates to another on-chip photonic neural network system, which comprises a probabilistic activation layer and a regulation unit, the probabilistic activation layer is the novel phase-change electrically-controlled photonic neuron device with an electrode layer introduced above, and the regulation unit is used for applying a voltage pulse to the electrode layer, and the transmittance of the phase-change material layer changes randomly with the size of the voltage pulse.

[0058] Due to the randomness of the distribution of Sb2Te3 phase in the volatile phase change material, the heat distribution is also random in the process of heating-induced phase change. Under high-voltage short-time pulse, the heat reaching the surface of the material is limited, and the heat will be randomly absorbed by part of the grains to make part of the grains produce phase transition, thereby generating the probabilistic firing characteristics of neurons. Specifically, under the condition that other parameters remain unchanged, the voltage amplitude has a greater impact on the temperature of the phase change material layer, and the phase change material temperature cannot be controlled more accurately by controlling the voltage amplitude. The embodiment adopts a pulse width adjustment method, which adjusts the pulse width near the phase change threshold voltage to adjust the temperature of the phase change material layer, controls the change amount of heat reaching the material to be smaller, and more accurately controls the firing probability of neurons. Due to the two-phase structure inside the material, adjusting the pulse width under a fixed voltage amplitude can make the temperature reaching the inside of the material randomly distributed, so that part of the material randomly phase changes, which will produce probabilistic modulation on the optical signal, realize the probabilistic firing of neurons, and can be used in the probabilistic calculation of neural network neurons.

[0059] As shown in FIG. 6, it is a schematic diagram of the probabilistic response of neurons in the neuron device in an embodiment of the present application. Under the same electrical pulse, the generation of neuron action potential has randomness, which is used for the probabilistic neuron layer in the neural network.

[0060] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application. It should be noted that the "in an embodiment of the present application", "for example", "such as" and the like in the present application are intended to illustrate the present application, and are not used to limit the present application.

[0061] The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the application. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application.

Claims

1. A novel phase change electrically controlled photonic neuron device, characterized by, A device comprising a substrate, a waveguide layer on the substrate, a phase change material layer on the waveguide layer, and a heating layer for heating the phase change material layer; The chemical general formula of the phase change material layer is Sb x Te y , x+y=1, 0.6 The phase change material layer has two phases with different physical properties, which are Sb2Te3 phase and Te phase respectively, and the Sb2Te3 phase is inlaid in the Te phase to form an island structure. When the temperature of the heating layer is raised to be greater than the Te phase state transition threshold, the Te phase changes from a crystalline state to an amorphous state or a molten state, and the Sb2Te3 phase remains in a crystalline state; when the temperature of the heating layer is lowered to be less than the Te phase state transition threshold, the Te phase changes from an amorphous state or a molten state to a crystalline state with the Sb2Te3 phase as a nucleation site.

2. The novel phase change electrically controlled photonic neuron device of claim 1, wherein, The heating layer is selected from any one of ITO, In2O3, AZO, and TiO2.

3. A novel phase change electrically controlled photonic neuron device as claimed in claim 1 or 2, characterized in that, The device further comprises an electrode layer in contact with the heating layer, the electrode layer being used to access an external power source and pass a current to the heating layer to adjust the temperature of the heating layer.

4. The novel phase change electrically controlled photonic neuron device of claim 3, wherein, The electrode layer is one or a combination of Ti, Cr, Pt, Au, Al, and W.

5. A method for preparing a novel phase change electrically controlled photonic neuron device, characterized by, The device comprises: forming a waveguide layer on a substrate; forming an initial layer of phase change material on the waveguide layer, the initial layer of phase change material having a chemical formula of Sb x Te y , x+y = 1, 0.6 < y < 1; forming a heating layer on the phase change material initial layer; forming an electrode layer on both ends of the heating layer; applying a pre-treatment pulse voltage to the electrode layer, and heating the phase change material initial layer through the heating layer to combine Sb atoms and Te atoms in the phase change material initial layer to form a stable Sb2Te3 phase, the Sb2Te3 phase being embedded in the Te phase and forming an island structure to obtain a phase change material layer. The voltage amplitude of the pre-treatment pulse voltage is 0.5V-5V, and the pulse width is 50μs-800μs.

6. The method of preparing a novel phase change electrically controlled photonic neuron device according to claim 5, wherein, The device comprises a non-linear activation layer and a regulating unit, the non-linear activation layer being the novel phase change electrically controlled photonic neuron device as claimed in claim 3 or 4, and the regulating unit being used to apply a voltage pulse to the electrode layer to cause the transmittance of the non-linear activation layer to change nonlinearly with the size of the voltage pulse.

7. A photonic neural network system on chip, characterized by The regulating unit is used to regulate the amplitude of the voltage pulse to regulate the light transmittance of the phase change material layer.

8. The on-chip photonic neural network system of claim 7, wherein, The device comprises a probabilistic activation layer and a regulating unit, the probabilistic activation layer being the novel phase change electrically controlled photonic neuron device as claimed in claim 3 or 4, and the regulating unit being used to apply a voltage pulse to the electrode layer to cause the transmittance of the phase change material layer to change randomly with the size of the voltage pulse.

9. A photonic neural network system on chip, characterized by The regulating unit is used to regulate the width of the voltage pulse to regulate the probability of random light transmittance of the phase change material layer.

10. The on-chip photonic neural network system of claim 9, wherein, ​

Images