Multi-path photon storage and calculation integrated device and photon neural network

Abstract

The invention belongs to the technical field of micro-nano photoelectrons, and particularly relates to a multi-path photon storage and calculation integrated device and a photon neural network. The multi-path photon storage and calculation integrated device comprises a multi-path cross structure and an on-chip photon device structure of a photon storage material from bottom to top. The photon storage material layer is located at the central position of the multi-path cross structure; the multi-channel light-substance interaction characteristic of a multi-channel cross structure is utilized and combined with the structure of a material / device on the waveguide, so that the multi-channel storage and calculation integrated function is realized; electrode layers are added above and below the photon storage material layer, and multi-channel photon storage and calculation functions are realized through at least one of optical excitation, electric excitation and thermal excitation. The invention further comprises a photon neural network, and the structure of the photon neural network is formed by cascading a plurality of photon storage and calculation integrated devices. The invention provides a novel device and a theoretical basis for developing on-chip integrated optical path applications such as a high-speed, low-energy-consumption, high-precision and high-expandability photon neural network.

Description

Technical Field

[0001] The present invention belongs to the field of micro-nano optoelectronic technology, and specifically relates to a multi-channel photon storage and computing integrated device and a photon neural network. Background Art

[0002] With the rapid development of the Internet of Things (IoT), big data analytics, and cloud computing, artificial neural networks (ANNs) are playing an increasingly important role in fields such as computer vision, speech recognition, and natural language processing. However, existing electronic computing systems, limited by the von Neumann architecture, cannot fully meet the large-scale matrix-vector multiplication (MVM) requirements of ANNs. Furthermore, CMOS-based electronic computing hardware is approaching the limits of computational speed and energy efficiency. Optical technology, by contrast, offers a promising approach for achieving rapid data transmission and processing, offering advantages such as high speed, low energy consumption, low heat dissipation, and high bandwidth. Over the past decade, optical interconnects for multi-core on-chip low-latency communication and nanosecond-scale ultrafast optical computing components have been developed. Optical technology offers an upgrade path for ANNs that can significantly advance the speed and energy efficiency of parallel data processing, namely optical neural networks (ONNs). Summary of the Invention

[0003] The purpose of the present invention is to provide a new type of multi-channel photonic storage and computing integrated device to greatly improve the parallelism of the device; at the same time, it provides a photonic neural network to improve the training efficiency, test accuracy and scalability of the neural network.

[0004] The multi-path photon storage and computing device provided by the present invention has a structure as shown in FIG. Figure 1 As shown, from bottom to top, the multi-channel cross-connect structure and the photon storage material or device are shown. The photon storage material layer is located at the center of the multi-channel cross-connect structure. The multi-channel light-matter interaction characteristics of the multi-channel cross-connect structure are combined with the structure of the materials or devices on the waveguide to achieve multi-channel storage and computing functions. The device can also be used independently as an optical computing device or optical storage device.

[0005] The storage material of the multi-path photonic storage and computing device can be a non-volatile reversible material (such as phase change material, two-dimensional material, ferroelectric material, magnetic material, etc.), which is used to realize matrix-vector multiplication and (weight) matrix storage; it can also be a reversible material with nonlinear optical response or photoelectric response (such as phase change material, two-dimensional material, ferroelectric material, magnetic material, etc.), which is used to realize nonlinear activation function.

[0006] Optionally, the photonic storage and computing integrated device is covered with a protective layer material, such as a dielectric material such as ITO, Al2O3, HfO2, etc. This can be achieved by methods such as PVD, PECVD, and ALD.

[0007] Optionally, the waveguide material of the device must have a transparent window in the used optical band (eg, C band and L band), and the material may be one of Si3N4, Si, LiNbO3, etc.

[0008] Optionally, electrode layers are added above and below the photon storage material layer to form a multi-way cross structure / lower electrode / photon storage material (device) / upper electrode structure, and the material state is changed by electrical or thermal excitation to complete light modulation and realize the integrated photon storage and computing function.

[0009] Optionally, the shape and size of the multi-way cross structure and the waveguide structure can be set according to actual application requirements; preferably, the waveguide structure can be one or more of a strip waveguide, a ridge waveguide, a slot waveguide, etc.; each branch in the multi-way cross structure is elliptical, and the major and minor axes of the ellipse can be 0.1 ~ 100 μm.

[0010] Optionally, the lower / upper electrode material is an electrode material that is transparent in the used light band (such as ITO, graphene, carbon nanotubes, etc.), or an electrode material that has weak absorption and is less affected by light intensity and temperature.

[0011] Optionally, the photon storage material can achieve reversible state transition of the material through at least one of heat generated by light absorption, Joule heat generated by electricity, direct heating by a thermal field, and the like.

[0012] The present invention also provides a photonic neural network implemented using the photonic storage and computing integrated device, the structure of which is composed of a cascade of multiple photonic storage and computing integrated devices, wherein: The multi-channel photonic storage and computing integrated device as a photonic neural network structural unit is composed of a multi-channel cross structure and photonic storage materials / devices, and the photonic storage materials / devices must have non-volatile storage characteristics.

[0013] The neural network structure unit can be directly connected to the reading waveguide of the next neural network structure unit through the reading waveguide of the unit, thereby realizing the scale expansion of the neural network.

[0014] The front-stage multi-channel photonic storage and computing device obtains the optical pulse signal of the corresponding working band from the erasing waveguide for weight matrix storage, and modulates the reading light obtained from the reading waveguide for matrix-vector operation, and then transmits the modulated reading light to the back-stage multi-channel photonic storage and computing device through the reading waveguide.

[0015] There are at least two multi-channel photonic storage and computing devices.

[0016] Optionally, the photon storage material / device of the multi-channel photon storage and computing device can be one or more of non-volatile phase change materials / devices, two-dimensional materials / devices, ferroelectric materials / devices, magnetic materials / devices, thermo-optical effect materials / devices and electro-optical effect materials / devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the waveguide structure of the novel multi-path photonic storage and computing integrated device of the present invention.

[0018] Figure 2 Schematic diagram of the waveguide structure of the photonic neural network of the present invention.

[0019] In the figure, reference numerals 10 denote a multi-path cross structure, 11 denotes a photon storage material layer (or device), 12 denotes an erasing waveguide, 13 denotes an erasing light, 14 denotes a reading light, and 15 denotes a reading waveguide. DETAILED DESCRIPTION

[0020] Example 1, a photon storage and computing integrated device, the structure of which is composed of a multi-way cross structure (10) and a photon storage material layer (11), such as Figure 1 shown.

[0021] As an example, the optical waveguide layer material of the waveguide structure (12, 15) may be one of Si3N4, Si, LiNbO3, etc. Preferably, in this embodiment, the optical waveguide layer material of the waveguide structure (12, 15) is Si3N4.

[0022] As an example, the waveguide structure dimensions of the multi-way cross structure (10) and the waveguide structure (12, 15) can be set according to actual application requirements; preferably, the waveguide structure can be one or more of a strip waveguide, a ridge waveguide, a slot waveguide, etc., the total height of the waveguide structure can be 0.05 ~ 3 μm, the waveguide structure width can be 0.05 ~ 10 μm, and the major semi-axis and minor semi-axis of the ellipse in the multi-way cross structure can be 0.1 ~ 100 μm; more preferably, in this embodiment, the total height of the waveguide structure is 0.34 μm, the height of the ridge waveguide structure is 0.23 μm, the waveguide width is 1.3 μm, and the major semi-axis and minor semi-axis of the ellipse in the multi-way cross structure are 3 μm and 10 μm, respectively.

[0023] As an example, the multi-way cross structure (10) and the waveguide structure (12, 15) can be prepared using Si3N4 wafers through ultraviolet lithography, electron beam lithography (EBL), reactive ion etching (RIE), inductively coupled plasma etching (ICP), etc.

[0024] As an example, the photon storage material layer (11) can be one or more of phase change materials, two-dimensional materials, ferroelectric materials, magnetic materials, etc. Preferably, in this embodiment, the photon storage material layer is crystalline GST, that is, crystalline Ge2Sb2Te5.

[0025] As an example, the thickness, width and height of the photon storage material layer (11) can be set according to actual application requirements; preferably, the thickness of the photon storage material layer (11) can be 0.4 to 100 nm, the width can be 0.05 to 50 μm, and the length can be 0.05 to 50 μm; more preferably, in this embodiment, the thickness of the photon storage material layer (11) is 10 nm, the width is 1.3 μm, and the length is 0.5 to 3 μm.

[0026] As an example, the photon storage material layer (11) can be deposited by sputtering, evaporation, chemical vapor deposition (CVD), plasma enhanced chemical vapor deposition (PECVD), low pressure chemical vapor deposition (LPCVD), metal compound vapor deposition (MOCVD), molecular beam epitaxy (MBE), atomic vapor deposition (AVD) or atomic layer deposition (ALD).

[0027] As an example, the photon storage material layer (11) may be covered with a layer of protective material, which may be a transparent material such as ITO, Al2O3, HfO2, graphene, etc. Preferably, in this embodiment, the protective layer material is ITO.

[0028] As an example, the thickness of the protective layer material may be 0.4 to 1000 nm, and the width and length are not less than the photon storage material layer (11); preferably, in this embodiment, the thickness of the protective layer material is 10 nm.

[0029] As an example, the protective layer material can be prepared by sputtering, evaporation, chemical vapor deposition (CVD), plasma enhanced chemical vapor deposition (PECVD), low pressure chemical vapor deposition (LPCVD), metal compound vapor deposition (MOCVD), molecular beam epitaxy (MBE), atomic vapor deposition (AVD) or atomic layer deposition (ALD).

[0030] Specifically, the photon storage material layer (11) can produce a strong absorption effect on high-power input light pulses (13) within the working band of 1530 to 1560 nm, and undergo a crystallization or amorphization phase transition under the action of the photothermal effect, thereby changing the absorption coefficient of the photon storage material layer (11) and realizing the photon storage function.

[0031] Specifically, the photon storage material layer (11) can produce a weak absorption effect on low-power input light pulses (14) within the working wavelength range of 1530 to 1560 nm, change the power of the light pulse and output it, and is insufficient to induce a crystallization or amorphization phase transition, thereby realizing the optical computing function. In this way, the photon storage and computing integrated device has a multi-channel storage and computing integrated function. Example 2

[0032] This embodiment provides a solution for realizing a photonic neural network by using the photonic storage and computing integrated device, and its structure is composed of three multi-path photonic storage and computing integrated devices cascaded from the reading waveguide (15) port, such as Figure 2 shown.

[0033] As an example, the multi-channel photonic storage and computing integrated device is composed of a multi-channel cross structure (10) and a photonic storage material layer (11), and the photonic storage material / device (11) must have non-volatile storage characteristics.

[0034] As an example, the photon storage material / device (11) may be one or more of a non-volatile phase change material / device, a two-dimensional material / device, a ferroelectric material / device, a magnetic material / device, a thermo-optical effect material / device, and an electro-optical effect material / device. Preferably, the photon storage material / device (11) is a Ge2Sb2Te5 (GST) thin film.

[0035] As an example, the output end of the readout waveguide (15) of the neural network structure unit can be directly connected to the input end of the readout waveguide (15) of the next neural network structure unit.

[0036] Specifically, the training / test set data can be simultaneously encoded into input optical pulses (13) of different wavelengths and transmitted to each multi-channel photon storage and computing integrated device through the erasable waveguide (12); each multi-channel photon storage and computing integrated device obtains the optical pulse signal of the corresponding working band from the erasable waveguide (12) for weight matrix storage, and modulates the read light (14) obtained from the read waveguide (15) to perform matrix-vector operations, and then transmits the modulated read light (14) to the subsequent multi-channel photon storage and computing integrated device through the read waveguide (15); the modulated read light (14) of the last multi-channel photon storage and computing integrated device is used as output light; the output light is used as the read light (14) of the next neural network structural unit, or the training / test results are obtained from it after decoding. In the above manner, the neural network structural unit can effectively reduce the crosstalk between different working wavelengths and significantly improve the accuracy and scalability of the neural network.

Claims

1. A multi-channel photon storage and computing device, characterized in that: The structure consists of a multi-channel cross structure and photon storage materials or devices from bottom to top. The photon storage material layer is located at the center of the multi-channel cross structure. It utilizes the multi-channel light-matter interaction characteristics of the multi-channel cross structure combined with the structure of the materials / devices on the waveguide to achieve multi-channel storage and computing functions. The storage material of the multi-path photonic storage and computing device is a non-volatile reversible material, which is used to realize matrix-vector multiplication and storage of weight matrices; or it is a reversible material with nonlinear optical response or photoelectric response, which is used to realize nonlinear activation function.

2. The multi-channel photon storage and computing device according to claim 1, characterized in that: The non-volatile reversible material is selected from phase change materials, two-dimensional materials, ferroelectric materials, and magnetic materials; the reversible material with nonlinear optical response or photoelectric response is selected from phase change materials, two-dimensional materials, ferroelectric materials, and magnetic materials.

3. The multi-channel photon storage and computing integrated device according to claim 2, characterized in that: The photonic storage and computing integrated device is covered with a protective layer material, which is selected from ITO, Al2O3, and HfO2 dielectric materials.

4. The multi-channel photon storage and computing device according to claim 3, characterized in that: The waveguide material of the device has a transparent window in the used optical band, and the window material is selected from one of Si3N4, Si, and LiNbO3.

5. The multi-channel photon storage and computing integrated device according to claim 4, characterized in that: Electrode layers are added above and below the photon storage material layer to form a multi-way cross structure / lower electrode / photon storage material or device / upper electrode structure. The material state is changed by electrical or thermal excitation to complete light modulation and realize the integrated photon storage and computing function.

6. The multi-channel photon storage and computing integrated device according to claim 5, characterized in that: The lower electrode and upper electrode materials are transparent electrode materials in the light band used, specifically selected from ITO, graphene, carbon nanotubes, or electrode materials with weak absorption and little influence by light intensity and temperature.

7. The multi-channel photon storage and computing integrated device according to claim 6, characterized in that: The waveguide structure may be one or more of a strip waveguide, a ridge waveguide, and a slot waveguide; and each branch in the multi-path cross structure is elliptical.

8. The multi-channel photon storage and computing integrated device according to claim 7, characterized in that: The photon storage material realizes reversible state transition of the material through at least one of heat generated by light absorption, Joule heat generated by electricity, and direct heating by a thermal field.

9. A photonic neural network implemented by the multi-channel photonic storage and computing device according to any one of claims 1 to 8, characterized in that: The structure consists of a cascade of multiple multi-channel photonic storage and computing devices, including: The multi-channel photon storage and computing device as a photon neural network structural unit is composed of a multi-channel cross structure and a photon storage material / device, and the photon storage material / device has non-volatile storage characteristics; The neural network structure unit is directly connected to the reading waveguide of the next neural network structure unit through the reading waveguide of the unit, thereby realizing the scale expansion of the neural network; The front-stage multi-channel photon storage and computing device obtains the optical pulse signal of the corresponding working band from the erasing waveguide to store the weight matrix, and modulates the reading light obtained from the reading waveguide to perform matrix-vector operations, and then transmits the modulated reading light to the back-stage multi-channel photon storage and computing device through the reading waveguide; There are at least two cascaded multi-channel photonic storage and computing devices.

10. The photonic neural network according to claim 9, wherein: The photon storage material / device of the multi-path photon storage and computing device is one or more of non-volatile phase change materials / devices, two-dimensional materials / devices, ferroelectric materials / devices, magnetic materials / devices, thermo-optical effect materials / devices and electro-optical effect materials / devices.

Images