High-integration low-loss photonic convolution computing unit and photonic convolution computing chip

Abstract

The application provides a high-integration and low-loss photonic convolution calculation unit, comprising a substrate and a cladding layer deposited on the substrate; a side waveguide and a photonic crystal nanobeam are deposited in the cladding layer; wherein a protrusion is deposited on one side of the photonic crystal nanobeam away from the side waveguide in the coupling area of the side waveguide and the photonic crystal nanobeam, so that the side waveguide forms a T-shaped waveguide. In the coupling area of the side waveguide and the photonic crystal nanobeam, a protrusion is deposited on one side of the photonic crystal nanobeam away from the side waveguide, so that the side waveguide forms a T-shaped waveguide to provide a discrete state, thereby realizing a Fano resonance linear spectrum line, effectively reducing the loss of the photonic convolution calculation chip, and realizing a low-loss convolution pool.

Description

Technical Field

[0001] This invention relates to the field of photonic computing technology, and in particular to a highly integrated, low-loss photonic convolution computing unit and a photonic convolution computing chip. Background Technology

[0002] With the increase in national electricity consumption, my country's power grid lines are becoming increasingly complex. The State Grid Corporation of China has higher requirements for the detection and intelligent identification of insulator targets. We need to improve the accuracy and responsiveness of insulator image target recognition to promote the stability of the national power supply. Convolutional Neural Networks (CNNs) have excellent applications in image recognition. Thanks to the continuous maturation of silicon photonics technology, photonic convolutional neural networks have been developed, offering advantages such as fast response, low power consumption, miniaturization, and advanced technology.

[0003] With the increasing demand for computing ultra-large datasets in the era of artificial intelligence, there is a need for more highly integrated photonic convolution computing units to form convolution pools. Currently, photonic convolution computing units are generally micro-ring resonators (MRRs). Although micro-ring-based devices can achieve the μm level, they are limited by their free spectrum range (FSR), making large-scale integration difficult.

[0004] Photonic crystal nanobeams (PCNs) have an ultra-large FSR (free spectral range) in the C-band, enabling large-scale reuse, and their size is much smaller than that of MMR types, allowing for highly integrated convolutional pools.

[0005] Existing photonic convolution computation units are side-waveguide-coupled photonic crystal nanobeams, forming photonic crystal nanobeam cavity (PCNC) structures, such as... Figure 1 The diagram shown is a schematic of a photonic convolution computation unit in the prior art. Figure 2 As shown Figure 1 A cross-sectional view of the AA shows that the photonic convolution computation unit includes a substrate 1, a cladding layer 2 deposited on the substrate 1, a side waveguide 3, and a photonic nanobeam cavity 4. The side waveguide 3 and the photonic crystal nanobeam 4 are deposited within the cladding layer 2. A circular hole 401 with a gradually changing radius is formed in the photonic nanobeam cavity 4. A Ti heater 5 is deposited on the upper surface of the cladding layer 2, in the coupling region between the side waveguide 3 and the photonic crystal nanobeam 4. (The last sentence appears to be incomplete and possibly refers to a different topic.) Figure 3The diagram shows the spectrum output of a photonic convolution computation unit in the prior art. The spectrum output of the aforementioned photonic convolution computation unit in the prior art has sharp resonance peaks. Although this structure can also be used as a photonic convolution computation unit, it requires higher modulation power consumption. Summary of the Invention

[0006] To address the technical problems of low integration and high power consumption in existing photonic convolution computing units, one objective of this invention is to provide a highly integrated, low-loss photonic convolution computing unit, comprising: a substrate, and a cladding layer deposited on the substrate; and a side waveguide and a photonic crystal nanobeam deposited within the cladding layer.

[0007] In the coupling region between the side waveguide and the photonic crystal nanobeam, a protrusion is deposited on one side of the side waveguide away from the photonic crystal nanobeam, so that the side waveguide forms a T-shaped waveguide.

[0008] Preferably, the thickness of the substrate is 2 micrometers, and the thickness of the coating layer is 4.72 micrometers;

[0009] The thickness of the side waveguide and the photonic crystal nanobeam is 220 nanometers; the thickness of the protrusion is 220 nanometers.

[0010] Preferably, the width of the protrusion is 450 nanometers and the length is 450 nanometers.

[0011] Preferably, the gradual change period of the circular hole with a gradually changing radius in the photonic crystal nanobeam is 26, and the gradual change constant is 330 nanometers.

[0012] The central circular hole of the photonic crystal nanobeam with a gradually changing radius has a radius of 40 nanometers, and the side circular holes of the photonic crystal nanobeam with a radius of 110 nanometers have a radius of 110 nanometers.

[0013] Preferably, the upper surface of the cladding layer is located in the coupling region between the side waveguide and the photonic crystal nanobeam, where a Ti heater is deposited.

[0014] Another objective of this invention is to provide a highly integrated, low-loss photonic convolution computing chip, wherein the photonic convolution computing chip includes a convolution pool array formed by a highly integrated, low-loss photonic convolution computing unit array provided by this invention.

[0015] This invention provides a highly integrated, low-loss photonic convolution computing unit and a photonic convolution computing chip. In the coupling region between the side waveguide and the photonic crystal nanobeam, a protrusion is deposited on one side of the side waveguide that is offset from the photonic crystal nanobeam, so that the side waveguide forms a T-shaped waveguide to provide discrete states, thereby realizing Fano resonance linear spectral lines. This can effectively reduce the loss of the photonic convolution computing chip and realize a low-loss convolution pool. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A schematic diagram of a photonic convolution computation unit in the prior art is shown.

[0018] Figure 2 It shows Figure 1 Sectional view of AA.

[0019] Figure 3 This diagram illustrates the spectrum output of a photonic convolution computation unit in the prior art.

[0020] Figure 4 A schematic diagram of a highly integrated, low-loss photonic convolution computation unit of the present invention is shown.

[0021] Figure 5 It shows Figure 4 A cross-sectional view of BB.

[0022] Figure 6 This diagram illustrates the spectrum output of a highly integrated, low-loss photonic convolution computation unit according to the present invention.

[0023] Figure 7 The diagram illustrates the principle of a highly integrated, low-loss photonic convolution computation unit according to the present invention.

[0024] Figure 8 A schematic diagram of a highly integrated, low-loss photonic convolution computing chip according to the present invention is shown.

[0025] Figure 9 A schematic diagram of a test system for testing a photonic convolution computing chip is shown in one embodiment of the present invention.

[0026] Figure 10 The output spectrum of a highly integrated, low-loss photonic convolution computing chip of the present invention is shown at different temperatures. Detailed Implementation

[0027] To make the above and other features and advantages of the present invention clearer, the invention will be further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explanation to those skilled in the art and are exemplary only, not restrictive.

[0028] To address the technical problems of low integration and high power consumption in existing photonic convolution computing units, this invention provides a highly integrated, low-loss photonic convolution computing unit and a photonic convolution computing chip.

[0029] like Figure 4 The diagram shown is a schematic of a highly integrated, low-loss photonic convolution computation unit according to the present invention. Figure 5 As shown Figure 4 A cross-sectional view of BB, according to an embodiment of the present invention, provides a highly integrated, low-loss photonic convolution computing unit fabricated on silicon-on-insulator (SOI), comprising: a substrate 1, and a cladding layer 2 deposited on the substrate 1. Side waveguides 3 and photonic crystal nanobeams 4 are deposited within the cladding layer 2.

[0030] A circular hole 401 with a gradually changing radius is formed in the photonic nanobeam cavity 4. A Ti heater 5 is deposited on the upper surface of the cladding layer 2, in the coupling region between the side waveguide 3 and the photonic crystal nanobeam 4.

[0031] According to an embodiment of the present invention, in the coupling region between the side waveguide 3 and the photonic crystal nanobeam 4, a protrusion 301 is deposited on one side of the side waveguide 3 offset from the photonic crystal nanobeam 4, thereby forming a T-shaped waveguide. The Fabry-Perot (FP) cavity generated by the T-shaped waveguide formed by the side waveguide 3 is taken as a continuous state, and the photonic crystal nanobeam 4 is placed on one side of the T-shaped waveguide formed by the side waveguide 3 to provide a discrete state, thereby achieving Fano resonance, such as... Figure 6 The diagram shows the spectrum output of a highly integrated, low-loss photonic convolution computing unit according to the present invention. The Fano line has an asymmetric sharp resonance peak near the resonance wavelength, and the wavelength range for adjusting the light intensity from 0 to 1 is much narrower, thereby reducing power consumption.

[0032] According to an embodiment of the present invention, the substrate 1 is made of Si material, and the cladding layer 2 is made of SiO2 material. The thickness of the substrate 1 is 2 micrometers, and the thickness H of the cladding layer 2 is 4.72 micrometers. The side waveguide 3, the protrusion 301, and the photonic crystal nanobeam 4 are made of Si material. The thickness h of the side waveguide 3, the protrusion 301, and the photonic crystal nanobeam 4 is 220 nanometers. The width and length of the protrusion 301 are 450 nanometers.

[0033] The circular aperture 401 of the photonic crystal nanobeam 4, with a gradually changing radius, has a height of 220 nanometers. This is determined by Bragg's equation a = λ / 2n. effWhen the working wavelength is 1550nm, the gradient period of the circular hole 401 with the gradually changing radius of the photonic crystal nanobeam 4 is calculated to be 26, the gradient constant is 330nm, the central circular hole radius of the circular hole 401 with the gradually changing radius of the photonic crystal nanobeam 4 is 40nm, and the side circular hole radius of the circular hole 401 with the gradually changing radius of the photonic crystal nanobeam 4 is 110nm. That is, the circular hole 401 with the gradually changing radius of the photonic crystal nanobeam 4 gradually changes from a central circular hole diameter of 40nm to 110nm on both sides.

[0034] According to an embodiment of the present invention, a Ti heater 5 with a thickness of 100 nanometers is deposited on the upper surface of the cladding layer 2 in the coupling region between the side waveguide 3 and the photonic crystal nanobeam 4. The Ti heater 5 is connected to a metal electrode to adjust the temperature and change the weight value of the photonic convolution calculation unit.

[0035] like Figure 7 The diagram shown illustrates the principle of a highly integrated, low-loss photonic convolution computation unit according to the present invention. In an embodiment of the present invention, this highly integrated, low-loss photonic convolution computation unit is a side-coupled type. Light enters from the side waveguide input port, passes through the resonant cavity, and finally exits from the side waveguide output port. Its transmission characteristics are analyzed using temporal coupled-mode theory, and the equations are as follows:

[0036]

[0037]

[0038]

[0039] Where A is the amplitude of the 4-resonance mode of the photonic crystal nanobeam, and S... I+ S I- and S R- These represent the input, output, and reflection amplitudes, respectively; w0 is the resonant frequency; and τ... a and τ r These are the amplitude coupling attenuation coefficients when the photonic crystal nanobeam 4 and the side waveguide 3 are coupled, respectively, and θ is the phase factor introduced by the side waveguide 3 to achieve a Fano linear shape. For any frequency, the sum of transmittance and reflectivity satisfies the law of conservation of energy and equals one if and only if τ a =τ r Only when the transmittance can it be equal to zero, therefore let τ a =τ r =τ wav .

[0040] By solving equations (1)-(3), the transmittance expression can be obtained:

[0041]

[0042] Heating alters the effective refractive index of a device, thus changing its conductivity. Conductivity affects the electric field strength. From the energy formula E = □v, where v is the electromagnetic wave frequency (i.e., the wavelength of light), it can be deduced that changing the effective refractive index will change its wavelength, thereby altering the weight.

[0043] like Figure 8 The diagram shows a highly integrated, low-loss photonic convolution computing chip according to the present invention. According to an embodiment of the present invention, a highly integrated, low-loss photonic convolution computing chip is provided, including a modulator 6 formed by a micro-ring array, and a convolution pool array 7 formed by a highly integrated, low-loss photonic convolution computing unit array provided by the present invention.

[0044] After the light intensity is modulated by the micro-ring of modulator 6, it is multiplied by the photonic convolution calculation unit of convolution pool array 7, and the resulting light intensity values ​​are summed to realize the photonic convolution multiplication operation.

[0045] like Figure 9 The diagram shows a test system for testing a photonic convolution computation chip according to one embodiment of the present invention. The present invention tests the photonic convolution computation chip by constructing a test system, which includes a tunable laser 8, a pulse signal generator 9, and a spectrometer 10. The tunable laser 8 serves as the input light source. The voltage is adjusted by the pulse signal generator 9 to change or maintain the weights of the photonic convolution computation units of the photonic convolution computation chip. The output spectrum is observed by the spectrometer 10 to verify the feasibility of the convolution unit.

[0046] When a voltage is applied to the heater 5 of the photonic convolution calculation unit of the photonic convolution calculation chip using the signal generator 9, the resistor is heated, and the temperature is changed, thus altering the overall effective refractive index and achieving phase shift. This allows the function of changing the weights of the photonic convolution calculation unit to be modified. Figure 10 The image shows the output spectrum of a highly integrated, low-loss photonic convolution computing chip of the present invention at different temperatures.

[0047] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A highly integrated, low-loss photonic convolution computation unit, characterized in that, The photonic convolution computing unit includes: a substrate, and a cladding layer deposited on the substrate; a side waveguide and a photonic crystal nanobeam are deposited within the cladding layer; In the coupling region between the side waveguide and the photonic crystal nanobeam, a protrusion is deposited on one side of the side waveguide away from the photonic crystal nanobeam, so that the side waveguide forms a T-shaped waveguide.

2. The photonic convolution computation unit according to claim 1, characterized in that, The substrate has a thickness of 2 micrometers, and the coating layer has a thickness of 4.72 micrometers; The thickness of the side waveguide and the photonic crystal nanobeam is 220 nanometers; the thickness of the protrusion is 220 nanometers.

3. The photonic convolution computation unit according to claim 2, characterized in that, The protrusion has a width of 450 nanometers and a length of 450 nanometers.

4. The photonic convolution computation unit according to claim 1, characterized in that, The central circular hole of the photonic crystal nanobeam with a gradually changing radius has a radius of 40 nanometers, and the side circular holes of the photonic crystal nanobeam with a radius of 110 nanometers have a radius of 110 nanometers.

5. The photonic convolution computation unit according to claim 1, characterized in that, The upper surface of the cladding layer is where a Ti heater is deposited in the coupling region between the side waveguide and the photonic crystal nanobeam.

6. A highly integrated, low-loss photonic convolution computing chip, characterized in that, The photonic convolution computing chip includes a convolution pool array formed by the photonic convolution computing unit array of any one of claims 1 to 5.

Images