An ultra-compact non-volatile photonic synapse device

Abstract

The application discloses an ultra-compact non-volatile photonic neural synapse device and belongs to the field of optical computing. The device comprises a substrate, a ridge waveguide, a phase change film and a metal film. The ridge waveguide is arranged above the substrate and covers the whole substrate area. The phase change film is distributed above the ridge type area of the ridge waveguide. The metal film covers the phase change film and has the same geometric parameters as the phase change film. The base mode in the ridge waveguide and the surface plasmon polariton (SPPs) mode of the interface between the phase change film and the metal film form a superimposed light field. The geometric structure of the ridge waveguide will affect the effective mode refractive index of the superimposed mode. The length of the modulation area is further optimized to realize different normalized intensity outputs. By changing the phase state (crystalline state and amorphous state) of the phase change film, the photonic neural synapse has a large output intensity difference under the condition that the phase change film has different phase states.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of optical computing, and more particularly, relates to an ultra-compact non-volatile photonic synapse device. BACKGROUND

[0002] Traditional electrical computing systems are gradually unable to meet the requirements of modern computing and communication systems due to their slow computing speed, high energy consumption, and the gradual approach of semiconductor integration process to the Moore's limit. As the size of transistors gradually decreases, the problems of chip power consumption and heat dissipation become more and more serious, which seriously limits the development of large-scale integrated chips and the further improvement of chip computing efficiency. In order to solve this problem, researchers propose to use optical signal processing systems to solve the problems of energy consumption and computing speed, such as optical processing chips based on neuromorphic computing network architecture. For example, a brain-like chip based on neuromorphic computing technology can simulate the working principle of the human brain to achieve fast learning, and can meet complex information processing conditions through continuous autonomous learning. Generally, a neuromorphic computing network is composed of about 10 11 personal artificial neurons, each of which is based on an adjustable synapse to adjust the weight. Therefore, the adjustability of a single synapse will greatly affect the computing efficiency and energy consumption of an optical neuromorphic network. Synapses are important structures for signal transmission and information exchange of neurons, and are also the basis of neuromorphic computing technology. Optical synapses can be designed by taking advantage of the difference in optical properties between the crystalline and amorphous states of phase change materials (PCMs), and can be directly regulated by optical pulses. PCMs have excellent optical property contrast between covalent bonding amorphous state and resonance bonding crystalline state, and have been widely used in reconfigurable photonic applications, such as optical switches, optical routers and metasurfaces. Photonic synapses based on PCMs can maintain the on-off state without static power supply and have a modulation speed of picoseconds, and are considered as the key unit device for designing optical neuromorphic networks.

[0003] However, when PCMs are directly sputtered on the surface of silicon waveguide, the output transmittance intensity changes little in different phase change states. This is because the waveguide mode is mainly concentrated in the silicon-based waveguide region, only a small part of the light field energy can interact with PCMs, resulting in that the phase change of PCMs is difficult to cause a greater impact on the light field. Therefore, researchers designed non-volatile photonic synapse devices more sensitive to the phase change of PCMs, such as non-volatile optical switches based on Mach-Zehnder (MZ) structure and non-volatile optical switches based on ring coupler. This non-volatile neural photonic synapse using interference effect and coupler effect is more sensitive to the change of PCM phase state, but has a large physical size, which seriously hinders the large-scale integration of non-volatile optical neural morphological network. In order to solve this problem, some reports propose to design special small-size waveguide structures, such as periodic PCM array, ridge waveguide and surface plasmon waveguide, to enhance the interaction effect between waveguide and PCM. But the output transmittance intensity difference between different PCM states is still small, about 40%, which is not conducive to the design of multi-level adjustable neural photonic synapse, and limits the development of non-volatile photonic neural network. SUMMARY

[0004] In view of the defects of the prior art, the purpose of the present application is to provide an ultra-compact non-volatile photonic neural synapse device, which aims to solve the problem of too small normalized output intensity range of photonic neural synapse.

[0005] To achieve the above-mentioned purpose, the present application provides an ultra-compact non-volatile photonic neural synapse device, which comprises a substrate, a ridge waveguide, a phase change film and a metal film. The ridge waveguide is arranged above the substrate and covers the entire substrate area. The phase change film is distributed above the ridge waveguide ridge area. The metal film covers the phase change film and has the same geometric parameters as the phase change film.

[0006] After the input light passes through the coupling waveguide, it enters the photonic neural synapse device for transmission. Different cross-sectional geometric parameters of the ridge waveguide will affect the light field distribution in the ridge waveguide. At the same time, the input light excites surface plasmon polariton (SPPs) wave on the interface between the metal film and the phase change film. The wavelength and light field distribution of the SPPs wave are related to the refractive index of the metal film and the phase change film. Therefore, the light field in the ridge waveguide and the SPPs light field at the interface between the metal film and the phase change film finally form a superimposed light field. Part of the energy of the superimposed light field is distributed in the phase change film, increasing the interaction intensity between the light field and the phase change material. When the phase state of the phase change film is changed, the output intensity of the photonic neural synapse will change significantly.

[0007] Further, by the cross-sectional geometric parameters (height, width) of the ridge waveguide, the super-compact non-volatile photonic neural synapse device realizes the maximum optical output intensity modulation range when the phase change film is in different phase states. The thickness of the flat layer of the ridge waveguide is greater than or equal to 0, and the specific value is related to the actual optimization of the cross-sectional geometric parameters.

[0008] As preferred, the structure of the super-compact non-volatile photonic neural synapse device is a T-shaped structure, and the initial cross-sectional geometric parameters are: the height of the flat layer is 100 nm, the height of the ridge-shaped region is 240 nm, and the width of the silicon waveguide of the ridge-shaped region is 300 nm. By changing these parameters, when the phase change film is in different phase states, the effective mode refractive index difference of the photonic neural synapse is larger, and the modulation range of the phase change film on the output intensity of the photonic neural synapse is larger. When the phase change film is in an amorphous state, more optical field energy can be stably transmitted in the waveguide; when the phase change film is in a crystalline state, most of the optical field energy needs to be lost during transmission, thereby achieving the purpose of modulating the output intensity of the photonic neural synapse.

[0009] Further, the material of the phase change film is Ge2Sb2Se4Te1 (GSST).

[0010] Further, when the phase change film is in an amorphous state and a crystalline state, the real part of the refractive index is less than and greater than the real part of the refractive index of the metal film, respectively. As preferred, when the phase change film is in an amorphous state, the imaginary part of the refractive index should be approximately equal to 0, so as to reduce the insertion loss of the device.

[0011] Further, the selection of the metal film should ensure that the SPPs excited at the interface between the phase change film and the metal film have the smallest transmission loss.

[0012] Further, the material of the ridge waveguide is silicon or other low-loss semiconductor materials.

[0013] Further, a cladding layer is further included, which is arranged above the metal film. The cladding layer can be air or other materials, such as silicon dioxide.

[0014] Further, the cross-sectional geometric parameters of the ridge waveguide, the phase change film and the metal film are fixed or slowly changed in the propagation direction.

[0015] Further, the length of the phase change film and the metal film in the propagation direction should be selected so that the SPPs wave generated at the incident end and the SPPs wave reflected at the interface of the exit end satisfy the constructive interference condition.

[0016] Further, the super-compact non-volatile photonic neural synapse device can be processed on a 340 nm high silicon-on-insulator (SOI) platform or other high semiconductor platforms.

[0017] Compared with the prior art, the present application has the following beneficial effects:

[0018] 1. In order to achieve greater output transmission intensity difference between different phase states of PCMs in a shorter modulation distance, the present application provides an ultra-compact non-volatile photonic neural synapse device on a silicon platform, which significantly increases the modulation range of output intensity by designing a hybrid waveguide structure on the silicon platform, and has important value in a non-volatile adjustable photonic neural network.

[0019] 2. The ultra-compact non-volatile photonic neural synapse device provided by the present application has a modulation distance of about one micrometer, which is much smaller than the modulation distance required for existing non-volatile photonic neural synapse devices to achieve the maximum output intensity modulation range, and realizes large-scale change of normalized output intensity.

[0020] 3. On the basis of the output intensity difference of adjacent control levels being more than 1%, the ultra-compact non-volatile photonic neural synapse device provided by the present application can realize more than 64-level (6-bit) coding, which is twice as much as other output intensity modulation type non-volatile photonic neural synapse devices.

[0021] 4. The ultra-compact non-volatile photonic neural synapse device provided by the present application has a very small cell size, is beneficial to large-scale integration, and will have very important significance and value in future optical neural network design and optical computing applications. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a two-dimensional cross-sectional schematic view of an ultra-compact non-volatile photonic neural synapse device.

[0023] Figure 2 is a simulation result graph of an ultra-compact non-volatile photonic neural synapse device, (a) and (b) respectively represent the normalized electric field distribution of a traditional ridge waveguide and the hybrid waveguide provided by the present application when the phase change film 3 is in an amorphous state; (c) and (d) respectively represent the normalized electric field distribution of a traditional ridge waveguide and the hybrid waveguide provided by the present application when the phase change film 3 is in a crystalline state.

[0024] Figure 3 is a simulation result graph of an ultra-compact non-volatile photonic neural synapse device, (a) and (b) respectively represent the functional relationship between the normalized output intensity of the photonic neural synapse and the ridge type region width and flat layer height of the ridge waveguide 2 when the phase change film 3 is in an amorphous state and a crystalline state, and (c) represents the functional relationship between the normalized output intensity difference between different phase states and the ridge type region width and flat layer height of the ridge waveguide.

[0025] Figure 4is a simulation result diagram of an ultra-compact non-volatile photonic neural synapse device, and (a) and (b) respectively represent the transmission electric field distribution when the phase change film 3 is in amorphous state and crystalline state.

[0026] Figure 5 is a curve diagram of the relationship between modulation distance and normalized optical output intensity.

[0027] Figure 6 is a simulation result diagram of the multi-stage adjustable capability of an ultra-compact non-volatile photonic neural synapse device. DETAILED DESCRIPTION

[0028] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0029] The present application provides an ultra-compact non-volatile photonic neural synapse device, comprising a substrate 1, a ridge waveguide 2, a phase change film 3, and a metal film 4. The ridge waveguide 2 is arranged above the substrate 1 and covers the entire substrate 1 area. The phase change film 3 is distributed above the ridge type area of the ridge waveguide 2. The metal film 4 is arranged above the phase change film 3 and has the same geometric parameters as the phase change film 3.

[0030] Hereinafter, the silicon waveguide width of the ridge waveguide 2 is 140 nm, the flat layer height is 100 nm, the ridge type area height is 240 nm, the thickness of the phase change film 3 and the metal film 4 is 50 nm, and the length is 1 mm, which are taken as the initial geometric parameters of an ultra-compact non-volatile photonic neural synapse device.

[0031] Specifically, the phase change film material is Ge2Sb2Se4Te1 (GSST).

[0032] Specifically, the metal film material is gold.

[0033] Specifically, the sum of the flat layer height and the ridge type area height of the ridge waveguide 2 is always 340 nm.

[0034] EMBODIMENT

[0035] As Figure 1As shown, the multi-level tunable optical switch of the embodiment includes a substrate 1, a ridge waveguide 2, a phase change film 3 and a metal film 4. The ridge waveguide 2 covers the entire area of the substrate 1, the phase change film 3 is distributed on the upper side of the partial area of the ridge waveguide 2, and the metal film 4 covers the upper side of the phase change film 3 and has the same geometric parameters as the phase change film 3. After the input light passes through the coupling waveguide, it enters the photonic neural synapse device for transmission. Different cross-sectional geometric parameters of the ridge waveguide 2 will affect the light field distribution in the ridge waveguide 2. At the same time, the input light excites SPPs waves on the interface between the metal film 4 and the phase change film 3. The wavelength and light field distribution of the SPPs waves are related to the refractive index of the metal film 4 and the phase change film 3. Therefore, the light field in the ridge waveguide 2 and the SPPs light field at the interface between the metal film 4 and the phase change film 3 eventually form a superimposed light field. Part of the energy of the superimposed light field is distributed in the phase change film, thereby increasing the interaction strength between the light field and the phase change material. When the phase state of the phase change film 3 is changed, the output intensity of the photonic neural synapse will change significantly.

[0036] For a traditional ridge waveguide, the change in mode energy distribution caused by the change in phase state of the phase change film 3 is not obvious. The super-compact non-volatile photonic neural synapse device proposed in the present application can achieve significant changes in mode energy distribution, as shown in Figure 2 The geometric shape of the hybrid waveguide cross-section, such as the height of the flat layer, the height of the ridge region and the width of the ridge waveguide, will affect the real part and the imaginary part of the effective mode refractive index of the fundamental mode, and further affect the output transmittance intensity of the hybrid waveguide. By scanning these parameters, the photonic neural synapse device with the best multi-level tunable ability can be determined. For the super-compact non-volatile photonic neural synapse proposed in the present application, the structure optimization effect is analyzed and quantified by monitoring the normalized output intensity. The change in normalized output intensity difference (ΔT) is obtained by using three-dimensional finite difference time domain (FDTD) simulation, so as to quantify the multi-level tunable ability of our non-volatile photonic neural synapse device.

[0037] Specifically, the silicon waveguide width and the flat layer height of the ridge waveguide 2 are changed to observe the change in optical output intensity difference when the phase change film 3 is in different phase states. The functional relationship between the normalized output intensity and the silicon waveguide width and the flat layer height of the ridge waveguide 2 is as follows Figure 3It can be found that the normalized output intensity of the non-volatile photonic synapse slowly increases with the increase of the silicon waveguide width and slab height of the ridge waveguide 2 when the GSST is in amorphous state and crystalline state. And the normalized output intensity reaches the maximum value when the silicon waveguide width is 700 nm and the slab height is 200 nm. The reason why the optical output intensity difference gradually increases in the whole modulation range is that when the silicon waveguide width and the slab height of the ridge waveguide 2 are small, the mode field cannot stably exist in the silicon waveguide region, but is radiated in the external environment, thereby reducing the interaction between the superimposed light field and the phase change film 3, and thus the transmission loss caused by the GSST material is reduced. In addition, it can be found that the normalized output intensity difference of the phase change film 3 in the amorphous state and the crystalline state reaches the maximum value of 87.9%, which is almost twice that of the existing structure, and is suitable for the design of multi-level adjustable non-volatile photonic synapses in practical applications.

[0038] Specifically, the influence of the modulation distance on the normalized output intensity of the non-volatile photonic synapse device proposed by the present application is observed by changing the length of the phase change film 3 and the metal film 4. In fact, part of the light field energy existing in the phase change film 3 will be reflected at the interface between the device and the air, and when the input light and the reflected light interfere with each other, the real part of the mode effective refractive index has a great influence on the normalized output intensity of the output device. The cross-sectional normalized transmission electric field distribution of the phase change film 3 in the amorphous state and the crystalline state under the initial geometric parameters is shown in FIG. 6. Figure 4 When the GSST is in the amorphous state, the electric field is not uniformly attenuated along the propagation direction, but periodically enhanced and weakened. This is due to the reflection of SPPs waves transmitted at the interface between the gold film and the output end GSST, and the interference between the reflected signal and the SPPs waves, resulting in a periodic increase or decrease of the light field distribution. Therefore, if the real part and the imaginary part of the mode effective refractive index are used to modulate the output transmittance intensity of the waveguide at the same time, the output transmittance intensity difference of the waveguide will be much larger when the GSST is in different phase states. Since the interference effect is closely related to the propagation distance, the influence of the modulation distance on the normalized output intensity of the non-volatile photonic synapse device is simulated and analyzed. Figure 5 To change the influence of the modulation distance on the normalized output intensity of the device. From Figure 5 It can be seen that in the process of gradually increasing the modulation distance, the output transmittance intensity difference presents a periodic change, and the period is about 245 nm, and the maximum value is obtained when the modulation distance is 1010 nm, which is 89.2%.

[0039] Specifically, in the multi-level adjustable test of the ultra-compact non-volatile photonic synapse, the optical refractive index of the phase change film 3 can be changed by changing the crystallization degree of the phase change film 3, and the relationship between the crystallization degree and the optical refractive index satisfies:

[0040]

[0041] where p represents the degree of crystallization, e a and e c represent the dielectric constants of the phase-change film 3 in the amorphous and crystalline states, respectively. As a preferred example, the phase-change material used in the present application is GST, which has a refractive index of 2.9817+0.005i and 4.9956+0.2938i at a wavelength of 1550 nm in the amorphous and crystalline states, respectively. Simulation results show that the multi-level tunable optical switch can control the light output intensity at 65 different levels, which corresponds to a 6-bit programming resolution, as shown in Figure 6

[0042] It is to be understood that the above description is merely a preferred example of implementation of the application and is not intended to limit the application, and any modification, equivalent replacement and improvement made within the spirit and principle of the application shall be included in the scope of the application.​

Claims

1. An ultra-compact non-volatile photonic synaptic device, characterized in that: The invention comprises a substrate (1), a ridge waveguide (2), a phase change film (3), and a metal film (4); the ridge waveguide (2) is arranged above the substrate (1); the phase change film (3) is distributed above the ridge region of the ridge waveguide (2); and the metal film (4) is distributed above the phase change film (3); and the lengths of the phase change film (3) and the metal film (4) in the propagation direction should be selected so that the SPPs wave generated at the incident end and the SPPs wave reflected at the interface of the output end meet the constructive interference condition.

2. The ultra-compact non-volatile photonic neural synaptic device according to claim 1, characterized in that: The structure of the ridge waveguide (2) includes a T-type structure.

3. The ultra-compact non-volatile photonic neural synaptic device according to claim 1, characterized in that: The material of the phase change film (3) is Ge2Sb2Se4Te1.

4. The ultra-compact non-volatile photonic neural synaptic device according to claim 1, characterized in that: When the phase change film (3) is in an amorphous state and a crystalline state, the real part of its refractive index is respectively smaller than and larger than the real part of the refractive index of the metal film (4).

5. The ultra-compact non-volatile photonic neural synaptic device according to claim 1, characterized in that: The metal film (4) is gold.

6. The ultra-compact non-volatile photonic neural synaptic device according to claim 2, characterized in that: The material of the ridge waveguide (2) is silicon.

7. The ultra-compact non-volatile photonic neural synaptic device according to claim 1, characterized in that: It also includes a cladding layer (5), which is arranged above the metal film (4).

8. The ultra-compact non-volatile photonic neural synaptic device according to claim 1, characterized in that: The cross-sectional geometric parameters of the ridge waveguide (2), phase change film (3) and metal film (4) are fixed or change slowly in the propagation direction.

9. The ultra-compact non-volatile photonic neural synaptic device according to claim 1, characterized in that: The ultra-compact non-volatile photonic synapse is fabricated on a silicon-on-insulator (SOI) platform or other semiconductor platforms.

Images