Opto-electro-optic module and optical artificail neural network system inlcuding the same
The optical-optical module addresses the challenge of high-accuracy nonlinear calculations in optical neural networks by converting input light into electrical and optical signals using activation functions, enabling efficient and low-power nonlinear computations for deep learning applications.
Patent Information
- Application Number
- KR1020240093941
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2044-07-16
AI Technical Summary
Existing optical neural networks face challenges in performing nonlinear calculations with high accuracy, which are crucial for deep learning applications, due to the limitations of semiconductor-based electrical computers in terms of power consumption and computational complexity.
An optical-optical module comprising a photodetector, an electronic device, and a light source device that converts input light into electrical and then optical signals, applying activation functions like Relu or sigmoid functions to perform nonlinear calculations with high accuracy.
The module enables fast and efficient nonlinear calculations with reduced power consumption, facilitating the integration of multi-channel optical artificial neural networks and serving as a wavelength converter.
Smart Images

Figure 112024077240606-PAT00005_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to an optical artificial neural network system, and more specifically, to an optical-optical module configured to perform nonlinear calculations of an optical artificial neural network and an optical artificial neural network system including the same. Background Technology
[0002] Artificial Neural Networks are machine learning techniques that process information in a manner similar to how the human brain processes and learns information. In particular, due to the improvement of big data processing capabilities through the development of computer hardware and the advancement of deep learning algorithms, artificial neural network technology is rapidly expanding and being applied to various fields such as speech recognition, pattern recognition, natural language processing, and medical data analysis.
[0003] An artificial neural network consists of an input layer, a hidden layer, and an output layer, and each layer is composed of nodes that process signal information like neurons. The nodes in the hidden layer sum the weights of the signals received from the nodes in the input layer, perform a non-linear transformation, and transmit the result to the output layer to generate the final output signal.
[0004] The signal processing process of an artificial neural network is similar to the signal processing method of a neuron, and while there may be only one hidden layer, it may also consist of multiple hidden layers. An artificial neural network composed of multiple hidden layers is called a deep neural network (or deep learning artificial neural network), and learning using it is called deep learning.
[0005] Deep neural networks using deep learning require a large amount of computation to optimize weights as the number of hidden layers increases. In particular, as the number of neurons and hidden layers increases, the amount of computation increases exponentially. Consequently, computations using semiconductor-based electrical computers become more complex and require greater power consumption, eventually leading to a technological limit.
[0006] To overcome these limitations, optical neural computing technology, which possesses excellent characteristics for parallel processing, has been proposed. Optical computing technology can not only perform inherently efficient parallel computation but also process vast amounts of computation at high speeds with low power consumption.
[0007] The computational process of optical neural computing consists of linear and nonlinear calculations. Linear calculations can be implemented optically using optical devices. However, nonlinear calculations utilize the optical nonlinearity of materials, which presents a problem of low accuracy. Therefore, an optical method capable of implementing nonlinear calculations with high accuracy is required. The problem to be solved
[0008] The objective of the present invention is to provide an optical-optical module configured to perform nonlinear calculations of an optical artificial neural network with high accuracy, and an optical artificial neural network system including the same. means of solving the problem
[0009] A photoelectric module according to an embodiment of the present invention includes a photodetector that receives input light and converts the input light to generate a first current, an electronic device that amplifies the first current to generate a second current, and a light source device that converts the second current to generate output light having characteristics in which a non-linear calculation of an optical artificial neural network is performed.
[0010] For example, the output light has characteristics to which an activation function is applied, and the light source element generates output light based on a second current and the threshold current of the light source.
[0011] For example, the activation function is a Relu function, and when the second current is greater than the threshold current, the light source element oscillates such that the intensity of the output light is proportional to the intensity of the second current.
[0012] For example, it further includes a resistor that determines the maximum value of the output light intensity, the activation function is a sigmoid function, and when the second current is greater than the threshold current, the light source element oscillates such that the output light has an intensity corresponding to the maximum value.
[0013] For example, the photodetector includes a photodiode, the electronic device includes at least one of a transistor and an operational amplifier, and the light source device includes a surface light-emitting laser.
[0014] For example, a photovoltaic module includes a unit cell array composed of a plurality of unit cells, and each of the plurality of unit cells includes a photodetector, an electronic element, and a light source element.
[0015] For example, each of the multiple unit cells further includes a resistor that determines the maximum value of the output light intensity.
[0016] For example, a photovoltaic module includes a photodetector array in which photodetectors are arranged in an array form, and a light source array in which light source elements are arranged in an array form.
[0017] For example, the photodetector array and the light source array are arranged so that the input light and the output light travel in the same direction.
[0018] For example, a photovoltaic module includes a first PCB substrate including a photodetector array, a second PCB substrate including a light source array and an electronic element, and a connector connecting the first PCB substrate and the second PCB substrate.
[0019] For example, a photodetector array is placed on a first surface of a PCB substrate, and a light source array is placed on a second surface of a PCB substrate.
[0020] For example, the photodetector array and the light source array are arranged so that the input light and the output light travel in opposite directions.
[0021] For example, a photodetector array and a light source array are placed on a first surface of a PCB substrate.
[0022] An optical artificial neural network system according to an embodiment of the present invention includes a linear calculator that receives input light containing input data and performs a linear calculation on the input data to generate a first processed light, and a non-linear calculator that performs a non-linear calculation on the data included in the first processed light to generate an output light having characteristics to which an activation function is applied. The non-linear calculator includes a photodetector that converts the first processed light to generate a first current, an electronic device that amplifies the first current to generate a second current, and a light source device that converts the second current to generate an output light having characteristics to which an activation function is applied.
[0023] For example, a light source element generates output light based on a second current and the threshold current of the light source.
[0024] For example, the activation function is a Relu function, and when the second current is greater than the threshold current, the light source element oscillates such that the intensity of the output light is proportional to the intensity of the second current.
[0025] For example, a nonlinear calculator further includes a resistor that determines the maximum value of the output light intensity, the activation function is a sigmoid function, and when the second current is greater than the threshold current, the light source element oscillates so that the output light has an intensity corresponding to the maximum value.
[0026] For example, the photodetector includes a photodiode, the electronic device includes at least one of a transistor and an operational amplifier, and the light source device includes a surface light-emitting laser.
[0027] For example, a nonlinear calculator includes a unit cell array composed of a plurality of unit cells, and each of the plurality of unit cells includes a photodetector, an electronic element, and a light source element.
[0028] For example, a nonlinear calculator includes a photodetector array in which photodetectors are arranged in an array form, and a light source array in which light source elements are arranged in an array form. Effects of the invention
[0029] According to the present invention, a photoelectric module can convert a linear input optical signal into a non-linear output optical signal in the form of an activation function and amplify the intensity of the optical signal.
[0030] According to the present invention, the optoelectronic circuit has a fast driving speed and can be easily integrated by minimizing electronic components. Therefore, a multi-channel optical artificial neural network system can be easily configured.
[0031] According to the present invention, since the optoelectronic circuit can convert the wavelength of an optical signal, it can be utilized as a wavelength converter. Brief explanation of the drawing
[0032] Figure 1 illustrates the information processing process of an optical artificial neural network. Figures 2a and 2b illustrate graphs of output data. FIG. 3 shows an example of an optical artificial neural network system according to one embodiment of the present invention. FIG. 4 exemplarily shows a photoelectric module according to one embodiment of the present invention. FIG. 5 exemplarily shows a circuit diagram of a photovoltaic module according to one embodiment of the present invention. Figure 6 exemplarily shows a graph of the output light signal output from the photoelectric module of Figure 5. FIG. 7 exemplarily shows a circuit diagram of a photovoltaic module according to one embodiment of the present invention. Figure 8 exemplarily shows a graph of the output light signal output from the photoelectric module of Figure 7. FIG. 9 exemplarily shows a multichannel photovoltaic module according to one embodiment of the present invention. FIG. 10 exemplarily shows a multichannel photovoltaic module according to one embodiment of the present invention. FIG. 11 exemplarily shows a multichannel photovoltaic module according to one embodiment of the present invention. FIG. 12 exemplarily shows a multichannel photovoltaic module according to one embodiment of the present invention. Specific details for implementing the invention
[0033] In the following, embodiments of the present invention will be described clearly and in detail so that a person skilled in the art can easily practice the present invention.
[0034] The components described by reference to terms such as part or unit, module, block, ~or, ~er used in the detailed description and the functional blocks illustrated in the drawings may be implemented in the form of software, hardware, or a combination thereof. For example, software may be machine code, firmware, embedded code, and application software. For example, hardware may include electrical circuits, electronic circuits, processors, computers, integrated circuits, integrated circuit cores, pressure sensors, inertial sensors, microelectromechanical systems (MEMS), passive components, or a combination thereof.
[0035] FIG. 1 illustrates an exemplary information processing process of an optical artificial neural network. Referring to FIG. 1, the optical artificial neural network may include an input layer (10), a hidden layer (20), and an output layer (30).
[0036] The input layer (10) may include a plurality of nodes. Each of the plurality of nodes may transmit input data (ID) to the hidden layer (20).
[0037] The hidden layer (20) can generate output data (OD) by performing linear and non-linear calculations on input data (ID) received from the input layer (10). For example, the linear calculation unit (21) of the hidden layer (20) can perform linear calculations on input data (ID) received from the input layer (10). The non-linear calculation unit (23) of the hidden layer (20) can generate output data (OD) to which an active function is applied or output data (OD) having the characteristics of an active function by performing non-linear calculations on linearly calculated data (LCD).
[0038] In an exemplary embodiment, the linear calculation may include weight calculation. For example, the hidden layer (20) may include a plurality of nodes. The plurality of nodes of the hidden layer (20) may each be connected to a plurality of nodes of the input layer (10). The plurality of nodes of the hidden layer (20) may perform a linear calculation to assign weights to input data (ID) received from each of the plurality of nodes of the input layer (10). However, the present invention is not limited thereto, and the linear calculation unit (21) may be configured to perform various linear calculations, such as matrix calculation and convolution calculation.
[0039] The hidden layer (20) can transmit output data (OD) to the output layer (30). The output data (OD) to which the activation function is applied is described in more detail with reference to FIGS. 2a and FIGS. 2b.
[0040] The output layer (30) can generate a prediction result based on the output data (OD) received from the hidden layer (20).
[0041] In FIG. 1, the optical artificial neural network is illustrated as having one hidden layer (20), but the optical artificial neural network may include multiple hidden layers. Based on the calculations of the multiple hidden layers, the optical artificial neural network can learn weights for input data (ID) and continuously adjust the values of the weights. Therefore, as the number of hidden layers increases, the accuracy of the prediction of the optical artificial neural network can be improved.
[0042] In FIG. 1, the optical artificial neural network can transmit data using an optical signal (OS). For example, the input layer (10) transmits a first optical signal (OS1) containing input data (ID) to a linear calculation unit (21), the linear calculation unit (21) transmits a second optical signal (OS2) containing linearly calculated data (LCD) to a non-linear calculation unit (23), and the non-linear calculation unit (23) transmits a third optical signal (OS3) containing output data (OD) to an output layer (30).
[0043] FIGS. 2A and 2B illustrate a graph of output data. In FIGS. 2A and 2B, the horizontal axis represents the value of the input data, and the vertical axis represents the value of the output data. In FIG. 2A, the input data corresponds to the linearly calculated data (LCD) of FIG. 1, and the first output data (OD1) may correspond to the output data (OD) of FIG. 1. In FIG. 2B, the input data corresponds to the linearly calculated data (LCD) of FIG. 1, and the second output data (OD2) may correspond to the output data (OD) of FIG. 1.
[0044] Referring to FIG. 2a, the first output data (OD1) may have the characteristics of a Relu function. For example, if the value of the input data is less than or equal to the first threshold (TH1), the value of the first output data (OD1) may be '0'. If the value of the input data exceeds the first threshold (TH1), the value of the first output data (OD1) may increase in proportion to the value of the input data. In an exemplary embodiment, the first threshold (TH1) may be '0'.
[0045] Referring to FIG. 2b, the second output data (OD2) may have the characteristics of a sigmoid function. For example, if the value of the input data is less than or equal to the second threshold (TH2), the value of the second output data (OD2) may be '0'. If the value of the input data exceeds the third threshold (TH3), the value of the second output data (OD2) may be a constant value (e.g., '1').
[0046] FIG. 3 shows an example of an optical artificial neural network system according to an embodiment of the present invention. Referring to FIG. 3, the optical artificial neural network system (100) may include an input unit (110), a hidden unit (120), and an output unit (130).
[0047] In FIG. 3, the optical artificial neural network system (100) may be configured to perform the operation of the optical artificial neural network of FIG. 1. That is, the input unit (110) may be configured to perform the operation of the input layer (10) of FIG. 1, the hidden unit (120) may be configured to perform the operation of the hidden layer (20) of FIG. 1, and the output unit (130) may be configured to perform the operation of the output layer (30) of FIG. 1.
[0048] The input unit (110) may be configured to generate a first optical signal (OS1). The first optical signal (OS1) may include input data (ID). The input unit (110) may transmit the first optical signal (OS1) to a linear calculator (120a) of a concealed unit (120).
[0049] In an exemplary embodiment, a plurality of pixel regions corresponding to a plurality of nodes may be defined in a first optical signal (OS1). Each of the plurality of pixel regions may include corresponding input data (ID). For example, each of the plurality of pixel regions may store input data (ID) by adjusting the intensity of the input optical signal.
[0050] The linear calculator (120a) can receive a first optical signal (OS1) from the input unit (110). The linear calculator (120a) can perform a linear calculation on the input data (ID) included in the first optical signal (OS1) to generate linearly calculated data (LCD). The linear calculator (120a) can transmit a second optical signal (OS2) containing the linearly calculated data (LCD) to the non-linear calculator (120b).
[0051] In an exemplary embodiment, the linear calculator (120a) may include an optical device configured to perform linear calculations. For example, the linear calculator (120a) may include photoelectric elements such as a light modulator or lens comprising a plurality of pixels, and photoelectric systems such as a Vector Matrix Multiplication (VMM) system, a 4F system, etc. that combine the same.
[0052] The nonlinear calculator (120b) can receive a second optical signal (OS2) from the linear calculator (120a). The nonlinear calculator (120b) can perform a nonlinear calculation on the linearly calculated data (LCD) included in the second optical signal (OS2) to generate output data (OD) to which an active function is applied or output data (OD) having the characteristics of an active function. The nonlinear calculator (120b) can transmit a third optical signal (OS3) containing the output data (OD) to the output unit (130). The nonlinear calculator (120b) is described in more detail with reference to FIGS. 4 to 6b.
[0053] FIG. 4 illustrates an exemplary photoelectric module according to an embodiment of the present invention. In FIG. 4, the photoelectric module (200) corresponds to the nonlinear calculator (120b) of FIG. 3, the input optical signal (IOS) corresponds to the second optical signal (OS2) of FIG. 3, and the output optical signal (OOS) corresponds to the third optical signal (OS3) of FIG. 3.
[0054] The photoelectric module (200) may include a photodetector (210), an electronic element (220), and a light source (230). The photodetector (210) may receive an input light signal (IOS). The photodetector (210) may convert the received input light signal (IOS) into an electrical signal. For example, the photodetector (210) may convert the received input light signal (IOS) to generate a first current (I1).
[0055] The photodetector (210) can transmit the converted electrical signal to the electronic element (220). For example, the photodetector (210) can transmit the first current (I1) to the electronic element (220).
[0056] In an exemplary embodiment, the photodetector (210) may be a device configured to convert a light signal into an electrical signal. For example, the photodetector (210) may be a photodiode (PD). However, the scope of the present invention is not limited thereto.
[0057] The electronic device (220) can receive an electrical signal from the photodetector (210). The electronic device (220) can electrically process the received electrical signal. For example, the electronic device (220) can amplify the first current (I1) received from the photodetector (210) to generate a second current (I2).
[0058] The electronic element (220) can transmit the processed electrical signal to the light source (230). For example, the electronic element (220) can transmit a second current (I2) to the light source (230).
[0059] In an exemplary embodiment, the electronic element (220) may be a device configured to amplify a received electrical signal. For example, the electronic element (220) may be a transistor or an operational amplifier. However, the scope of the present invention is not limited thereto.
[0060] The light source (230) can receive an electrical signal processed from an electronic element (220). The light source (230) can convert the processed electrical signal into an optical signal. For example, the light source (230) can convert a second current (I2) to generate an output optical signal (OOS) to which an activation function is applied, or an output optical signal (OOS) having the characteristics of an activation function. The light source (230) can output an output optical signal (OOS). The output optical signal (OOS) is described in detail with reference to FIGS. 5a through 6b.
[0061] The light source (230) can convert an electrical signal into a light signal based on a threshold current. For example, if the second current (I2) received from the electronic element (220) is smaller than the threshold current of the light source (230), the light source (230) may not oscillate. That is, the intensity of the output light signal (OOS) of the light source (230) may be '0'.
[0062] For example, if the second current (I2) received from the electronic element (220) is greater than the threshold current of the light source (230), the light source (230) can oscillate to generate an output light signal (OOS) in the form of an active function.
[0063] In an exemplary embodiment, the light source (230) may be a light source element configured to convert a light signal into an electrical signal. For example, the light source (230) may be one of a laser, a light-emitting diode (LED), and a surface light-emitting laser (VCSEL). However, the scope of the present invention is not limited thereto.
[0064] In an exemplary embodiment, the photoelectric module (200) may not amplify the electrical signal. For example, if the intensity of the first current (I1) output from the photodetector (210) exceeds a threshold value, the photoelectric module (200) may not amplify the first current (I1). Accordingly, the electronic element (220) may transmit the received first current (I1) as is to the light source (230).
[0065] In an exemplary embodiment, the photoelectric module (200) may be implemented as an electric circuit or an electronic circuit.
[0066] Unlike as illustrated in FIG. 1, in an exemplary embodiment, the photoelectric module (200) may not include an electronic element (220). For example, if the photodetector (210) is configured to output a first current (I1) exceeding a threshold intensity, the photoelectric module (200) may not include an electronic element (220).
[0067] Although not illustrated, the photoelectric module (200) may further include a resistor for controlling the intensity of the output light signal (OOS). For example, the photoelectric module (200) may further include a resistor configured to determine the maximum intensity of the output light signal (OOS).
[0068] As described above, the photoelectric module (200) may be configured to perform nonlinear calculations of an optical artificial neural network. The photoelectric module (200) may output a nonlinear optical signal based on a linear optical signal. For example, the photoelectric module (200) may convert a linear optical signal into an electrical signal, process (e.g., amplify) the electrical signal, and convert the processed electrical signal to output a nonlinear optical signal. The nonlinear optical signal may have the characteristics of an activation function and may have an amplified intensity compared to the linear optical signal. At this time, the characteristics of the activation function and the amplification ratio of the intensity may be controlled by the elements included in the photoelectric module (200).
[0069] FIG. 5 illustrates an exemplary circuit diagram of a photovoltaic module according to one embodiment of the present invention. In FIG. 5, the photodiode (PD), transistor (TR), and surface light-emitting laser (VCSEL) may correspond to the photodetector, electronic device, and light source of FIG. 4, respectively.
[0070] Referring to FIG. 5, the photovoltaic module (300) may include a photodiode (PD), a transistor (TR), and a surface light-emitting laser (VCSEL).
[0071] The photoelectric module (300) can operate based on the voltage (Vbb) applied to the photodiode (PD) and the voltage (Vcc) applied to the surface light-emitting laser (VCSEL) from a voltage source (VS). In an exemplary embodiment, the voltage (Vcc) may be generated by an electronic circuit, such as a voltage generator, located inside or outside the photoelectric module (300). In an exemplary embodiment, the voltage (Vcc) may be generated from a voltage source located inside or outside the photoelectric module (300).
[0072] A photodiode (PD) converts an input optical signal (IOS) into an electrical signal, which is current, and can output current. A transistor (TR) amplifies the current output from the photodiode (PD) and can output the amplified current. A surface light-emitting laser (VCSEL) converts the amplified current output from the transistor (TR) into an output optical signal (OOS) and can output the output optical signal (OOS).
[0073] In an exemplary embodiment, the wavelength of the input optical signal (IOS) may be '850' nm. In an exemplary embodiment, the surface light-emitting laser (VCSEL) may be a surface light-emitting laser (VCSEL) of '670' nm. For example, the wavelength of the output optical signal (OOS) may be '670' nm. However, the scope of the present invention is not limited thereto.
[0074] In an exemplary embodiment, the photoelectric module (300) may operate based on the input light signal (IOS), the current delivered to the surface light emitting laser (VCSEL) (hereinafter referred to as 'delivery current'), and the threshold current of the surface light emitting laser (VCSEL). For example, if the delivery current is smaller than the threshold current of the surface light emitting laser (VCSEL), the surface light emitting laser (VCSEL) may not oscillate. In this case, the intensity of the output light signal (OOS) may be '0'. If the delivery current is larger than the threshold current of the surface light emitting laser (VCSEL), the surface light emitting laser (VCSEL) may oscillate such that the intensity of the output light signal (IOS) is proportional to the intensity of the delivery current.
[0075] In an exemplary embodiment, the photoelectric module (300) may operate based on the intensity of an input light signal (IOS) and the threshold current of a surface light-emitting laser (VCSEL). For example, if the intensity of the input light signal (IOS) is below a threshold value, the transfer current may be smaller than the threshold current of the surface light-emitting laser (VCSEL). In this case, the surface light-emitting laser (VCSEL) may not oscillate, and the intensity of the output light signal (OOS) may be '0'. If the intensity of the input light signal (IOS) exceeds a threshold value, the transfer current may be larger than the threshold current of the surface light-emitting laser (VCSEL). In this case, the surface light-emitting laser (VCSEL) may oscillate such that the intensity of the output light signal (IOS) is proportional to the intensity of the input light signal (IOS), and the output light signal (OOS) may have the form of a Relu function among the activation functions.
[0076] As described above, the photoelectric module (300) can output an output optical signal (OOS) to which a Relu function among the active functions is applied, or an output optical signal (OOS) having the characteristics of a Relu function, based on the intensity of the transmission current or input optical signal (IOS). However, the scope of the present invention is not limited thereto, and the photoelectric module (300) can output an output optical signal (OOS) having the characteristics of various active functions.
[0077] FIG. 6 illustrates an exemplary graph of an output light signal output from the photovoltaic module of FIG. 5. In FIG. 6, the horizontal axis represents the intensity of the input light signal (IOS), and the vertical axis represents the intensity of the output light signal (IOS).
[0078] Referring to FIGS. 5 and 6, the output light signal (OOS) output from the photoelectric module (300) may have the characteristics of a Relu function. That is, when the intensity of the input light signal (IOS) is below a threshold value (TH), the intensity of the output light signal (OOS) may be '0' or an approximation of '0'. When the intensity of the input light signal (IOS) exceeds the threshold value (TH), the intensity of the output light signal (OOS) may increase in proportion to the intensity of the input light signal (IOS). In this case, the threshold value (TH) may be related to the threshold current of the surface light emitting laser (VCSEL).
[0079] In FIG. 6, the intensity of the output optical signal (OOS) is shown as being related to the intensity of the input optical signal (IOS), but the scope of the invention is not limited thereto. For example, the intensity of the output optical signal (OOS) may be related to the intensity of the transmission current.
[0080] As described above, the graph of the output light signal (OOS) output from the photovoltaic module (300) may include one inflection point related to the threshold current of the surface light-emitting laser (VCSEL).
[0081] FIG. 7 illustrates an exemplary circuit diagram of a photoelectric module according to an embodiment of the present invention. In FIG. 7, the photoelectric module (400) may correspond to the photoelectric module (300) of FIG. 5, except that the photoelectric module (400) further includes a resistor (R). That is, the photodiode (PD), transistor (TR), and surface light-emitting laser (VCSEL) of the photoelectric module (400) may correspond to the photodiode (PD), transistor (TR), and surface light-emitting laser (VCSEL) of the photoelectric module (300) of FIG. 5, respectively. Therefore, redundant descriptions are omitted.
[0082] Referring to FIG. 7, the resistor (R) can determine the maximum intensity of the output optical signal (OOS). For example, the maximum intensity of the output optical signal (OOS) can be adjusted according to the value of the resistor (R).
[0083] In an exemplary embodiment, the maximum intensity of the output optical signal (OOS) may be determined based on resistance (R) and voltage (Vcc). However, the scope of the present invention is not limited thereto.
[0084] The photoelectric module (400) can output the maximum intensity of the output light signal (OOS) determined by the resistance (R) (hereinafter referred to as the 'maximum value') as the intensity of the output light signal (OOS).
[0085] In an exemplary embodiment, the photovoltaic module (400) may output a maximum value as the intensity of the output light signal (OOS) based on the transfer current and the threshold current of the surface light-emitting laser (VCSEL). For example, if the transfer current is less than the threshold current of the surface light-emitting laser (VCSEL), the surface light-emitting laser (VCSEL) may not oscillate. In this case, the intensity of the output light signal (OOS) may be '0'. If the transfer current is greater than the threshold current of the surface light-emitting laser (VCSEL), the surface light-emitting laser (VCSEL) may oscillate to output a maximum value as the intensity of the output light signal (OOS). For example, the intensity of the output light signal (OOS) may increase according to the intensity of the transfer current until it reaches a maximum value. After reaching a maximum value, the intensity of the output light signal (OOS) may remain constant as the maximum value.
[0086] In an exemplary embodiment, the photoelectric module (400) can output a maximum value as the intensity of the output optical signal (OOS) based on the intensity of the input optical signal (IOS) and the threshold current of the surface light-emitting laser (VCSEL).
[0087] For example, if the intensity of the input optical signal (IOS) is below a threshold value, the transfer current may be smaller than the threshold current of the surface light-emitting laser (VCSEL). In this case, the surface light-emitting laser (VCSEL) may not oscillate, and the intensity of the output optical signal (OOS) may be '0'. If the intensity of the input optical signal (IOS) exceeds the threshold value, the transfer current may be larger than the threshold current of the surface light-emitting laser (VCSEL). In this case, the surface light-emitting laser (VCSEL) may oscillate to output a maximum value determined by the resistor (R) as the intensity of the output optical signal (OOS). For example, the intensity of the output optical signal (OOS) may increase according to the intensity of the transfer current until it reaches a maximum value. After reaching a maximum value, the intensity of the output optical signal (OOS) may remain constant as the maximum value.
[0088] As described above, the photoelectric module (400) can output an output optical signal (OOS) having the characteristics of a sigmoid function among the activation functions based on the intensity of the transmission current or input optical signal (IOS). However, the scope of the present invention is not limited thereto, and the photoelectric module (300) can output an output optical signal (OOS) having the characteristics of various activation functions.
[0089] FIG. 8 illustrates an exemplary graph of an output light signal output from the photovoltaic module of FIG. 7. In FIG. 8, the horizontal axis represents the intensity of the input light signal (IOS), and the vertical axis represents the intensity of the output light signal (IOS).
[0090] Referring to FIGS. 7 and 8, the output light signal (OOS) output from the photoelectric module (400) may have the characteristics of a sigmoid function. That is, when the intensity of the input light signal (IOS) is below a first threshold value (TH1), the intensity of the output light signal (OOS) may be '0' or an approximation of '0'. When the intensity of the input light signal (IOS) exceeds the first threshold value (TH1) and is below the second threshold value (TH2), the intensity of the output light signal (OOS) may increase based on the intensity of the input light signal (IOS). When the intensity of the input light signal (IOS) exceeds the second threshold value (TH2), the intensity of the output light signal (OOS) may correspond to a maximum value. At least one of the first threshold value (TH1) and the second threshold value (TH2) may be related to the threshold current of the surface light emitting laser (VCSEL).
[0091] In FIG. 8, the intensity of the output optical signal (OOS) is shown as being related to the intensity of the input optical signal (IOS), but the scope of the invention is not limited thereto. For example, the intensity of the output optical signal (OOS) may be related to the intensity of the transmission current.
[0092] As described above, the graph of the output light signal (OOS) output from the photovoltaic module (400) may include two inflection points related to the threshold current of the surface light-emitting laser (VCSEL).
[0093] FIG. 9 illustrates an exemplary multi-channel photovoltaic module according to an embodiment of the present invention. Referring to FIG. 9, the multi-channel photovoltaic module (500) may include a first PCB substrate (510), a second PCB substrate (520), and a connector (530).
[0094] The first PCB substrate (510) may include a plurality of photodetectors. For example, the first PCB substrate (510) may include a plurality of photodiodes (PDs).
[0095] In an exemplary embodiment, a plurality of photodiodes (PDs) may be mounted on a first PCB substrate (510) in the form of a two-dimensional array. For example, a photodetector array composed of a plurality of photodiodes (PDs) may be placed in a first region (R1) on the first PCB substrate (510). The photodetector array may be in a 4x4 form, but the scope of the invention is not limited thereto, and the photodetector array may be implemented in various forms.
[0096] The second PCB substrate (520) may include a plurality of light sources. For example, the second PCB substrate (520) may include a plurality of surface light-emitting lasers (VCSELs).
[0097] In an exemplary embodiment, a plurality of surface light-emitting lasers (VCSELs) may be mounted on a second PCB substrate (520) in the form of a two-dimensional array. For example, a light source array composed of a plurality of surface light-emitting lasers (VCSELs) may be placed in a second region (R2) on the second PCB substrate (520). The light source array may be in a 4x4 shape, but the scope of the invention is not limited thereto, and the light source array may be implemented in various forms.
[0098] The connector (530) may be configured to connect the first PCB board (510) and the second PCB board (520). The connector (530) may be configured to transmit a signal between the first PCB board (510) and the second PCB board (520). For example, the connector (530) may transmit an electrical signal (e.g., current) output from an optical detector array placed on the first PCB board (510) to the second PCB board (520).
[0099] In an exemplary embodiment, the photodetector array and the light source array may be arranged such that the propagation direction of the input light signal (IOS) and the propagation direction of the output light signal (OOS) are the same. For example, the first PCB substrate (510) and the second PCB substrate (520) may be arranged in parallel. The photodetector array may be placed in a first region (R1) on the first PCB substrate (510), and the light source array may be placed in a second region (R2) on the second PCB substrate (520). However, the scope of the present invention is not limited thereto.
[0100] In an exemplary embodiment, electronic components such as transistors (TR) or operational amplifiers may be mounted on a first PCB substrate (510) or a second PCB substrate (520). For example, one or more transistors (TR) may be mounted on a first region (R1) on the first PCB substrate (510) or a second region (R2) on the second PCB substrate (520).
[0101] In an exemplary embodiment, a transistor (TR) and a surface light-emitting laser (VCSEL) may form a unit cell. The unit cell may be arranged in an array form on a second PCB substrate (520). For example, a unit cell array composed of a plurality of unit cells may be arranged in a second region (R2) on the second PCB substrate (520).
[0102] In an exemplary embodiment, a resistor (R) may be mounted on a second PCB substrate (520). For example, one or more resistors (R) may be mounted on a second region (R2) on the second PCB substrate (520).
[0103] In an exemplary embodiment, a transistor (TR), a resistor (R), and a surface light-emitting laser (VCSEL) may constitute a unit cell. The unit cell may be arranged in an array form on the second PCB substrate (520). For example, a unit cell array composed of a plurality of unit cells may be arranged in a second region (R2) on the second PCB substrate (520).
[0104] As described above, the multi-channel photovoltaic module (500) can be easily integrated by having a three-dimensional structure.
[0105] FIG. 10 exemplarily shows a multichannel photovoltaic module according to one embodiment of the present invention.
[0106] Referring to FIG. 10, a multi-channel photovoltaic module (600) can be implemented on a first PCB substrate (PCB1). The multi-channel photovoltaic module (600) may include a first PCB substrate (PCB1) on which a plurality of light detectors and a plurality of light sources are mounted.
[0107] In an exemplary embodiment, a plurality of photodiodes (PDs) and a plurality of surface light-emitting lasers (VCSELs) may be mounted on a first PCB substrate (PCB1) in the form of a two-dimensional array. For example, a photodetector array composed of a plurality of photodiodes (PDs) may be placed in a first region (R1) on the first PCB substrate (PCB1), and a light source array composed of a plurality of surface light-emitting lasers (VCSELs) may be placed in a second region (R2) on the first PCB substrate (PCB1). That is, the photodetector array and the light source array may be placed on different sides of the same substrate so that the input light signal (IOS) and the output light signal (OOS) travel in the same direction. The photodetector array and the light source array may be in a 4x4 configuration, but the scope of the present invention is not limited thereto, and the photodetector array and the light source array may be implemented in various forms.
[0108] In an exemplary embodiment, electronic components such as transistors (TR) or operational amplifiers may be mounted on a first PCB substrate (PCB1). For example, one or more transistors (TR) may be mounted on a first region (R1) or a second region (R2) on the first PCB substrate (PCB1).
[0109] In an exemplary embodiment, a transistor (TR) and a surface light-emitting laser (VCSEL) may form a unit cell. The unit cell may be arranged in an array form on a first PCB substrate (PCB1). For example, a unit cell array composed of a plurality of unit cells may be arranged in a second region (R2) on the first PCB substrate (PCB1).
[0110] In an exemplary embodiment, a resistor (R) may be mounted on a first PCB substrate (PCB1). For example, one or more resistors (R) may be mounted on a second region (R2) on the first PCB substrate (PCB1).
[0111] In an exemplary embodiment, a transistor (TR), a resistor (R), and a surface light-emitting laser (VCSEL) may constitute a unit cell. The unit cell may be arranged in an array form on a first PCB substrate (PCB1). For example, a unit cell array composed of a plurality of unit cells may be arranged in a second region (R2) on the first PCB substrate (PCB1).
[0112] FIG. 10 illustrates an exemplary multi-channel photovoltaic module according to one embodiment of the present invention. Referring to FIG. 11, the multi-channel photovoltaic module (700) may be implemented on a first PCB substrate (PCB1). The multi-channel photovoltaic module (600) may include a first PCB substrate (PCB1) on which a plurality of light detectors and a plurality of light sources are mounted.
[0113] In an exemplary embodiment, a plurality of photodiodes (PDs) and a plurality of surface light-emitting lasers (VCSELs) may be mounted on a first PCB substrate (PCB1) in the form of a two-dimensional array. For example, a photodetector array composed of a plurality of photodiodes (PDs) may be placed in a first region (R1) on the first PCB substrate (PCB1), and a light source array composed of a plurality of surface light-emitting lasers (VCSELs) may be placed in a second region (R2) on the first PCB substrate (PCB1). That is, the photodetector array and the light source array may be placed on the same surface of the same substrate so that the input light signal (IOS) and the output light signal (OOS) travel in opposite directions. The photodetector array and the light source array may be in the form of a 4x4 array, but the scope of the present invention is not limited thereto, and the photodetector array and the light source array may be implemented in various array forms.
[0114] In an exemplary embodiment, electronic components such as transistors (TR) or operational amplifiers may be mounted on a first PCB substrate (PCB1). For example, one or more transistors (TR) may be mounted on a first region (R1) or a second region (R2) on the first PCB substrate (PCB1).
[0115] In an exemplary embodiment, a resistor (R) may be mounted on a first PCB substrate (PCB1). For example, one or more resistors (R) may be mounted on a first region (R1) or a second region (R2) on the first PCB substrate (PCB1).
[0116] FIG. 12 illustrates an exemplary multi-channel photovoltaic module according to an embodiment of the present invention. Referring to FIG. 12, the multi-channel photovoltaic module (800) may be implemented as a unit cell array (UCA) on a first PCB substrate (PCB1). The unit cell array (UCA) may include a plurality of unit cells (UC). Each of the plurality of unit cells (UC) may include a photodiode (PD), a transistor (TR), and a surface light-emitting laser (VCSEL). In this case, the multi-channel photovoltaic module (800) may output an output light signal (OOS) having the characteristics of a Relu function among the activation functions based on an input light signal (IOS).
[0117] In an exemplary embodiment, each of the plurality of unit cells (UC) may include a photodiode (PD), a transistor (TR), a resistor (R), and a surface light-emitting laser (VCSEL). In this case, the multichannel photoelectric module (800) may output an output light signal (OOS) having the characteristics of a sigmoid function among the activation functions based on an input light signal (IOS).
[0118] In an exemplary embodiment, the unit cell array (UCA) may be in the form of a 6x6 array. However, the scope of the present invention is not limited thereto, and the unit cell array (UCA) may be in various array forms.
[0119] As described above, the multi-channel photoelectric module (800) can be implemented as a unit cell array on a single PCB substrate to process the process of outputting an output optical signal (OOS) from an input optical signal (IOS) at high speed. Additionally, since the multi-channel photoelectric module (800) is easy to integrate, it can be easy to implement an optical artificial neural network system.
[0120] In the embodiments described above, components according to the technical concept of the present invention have been described using terms such as first, second, third, etc. However, terms such as first, second, third, etc. are used to distinguish the components from one another and do not limit the present invention. For example, terms such as first, second, third, etc. do not imply a sequential order or any numerical meaning.
[0121] The description above describes specific examples for implementing the present invention. The present invention will include not only the embodiments described above, but also embodiments that can be easily modified or simply changed. Furthermore, the present invention will include technologies that can be easily modified and implemented in the future using the embodiments described above. Explanation of the symbols
[0122] 100: Optical artificial neural network system 200, 300, 400: Photovoltaic module 500, 600, 700, 800: Multichannel Photovoltaic Module
Claims
Claim 1 A photoelectric module comprising: a photodetector that receives input light and converts the input light to generate a first current; an electronic device that amplifies the first current to generate a second current; and a light source device that converts the second current to generate output light having characteristics in which a non-linear calculation of an optical artificial neural network is performed, wherein the output light has characteristics in which an activation function is applied, and the light source generates the output light based on the second current and the threshold current of the light source device, wherein the photodetector includes a photodiode, the electronic device includes at least one of a transistor and an operational amplifier, and the light source device includes a surface light emitting laser (VCSEL). Claim 2 delete Claim 3 In claim 1, the activation function is a Relu function, and when the second current is greater than the threshold current, the light source element is a photovoltaic module in which the intensity of the output light is proportional to the intensity of the second current. Claim 4 A photoelectric module according to claim 1, further comprising a resistor determining the maximum value of the intensity of the output light, wherein the activation function is a sigmoid function, and when the second current is greater than the threshold current, the light source element oscillates such that the output light has an intensity corresponding to the maximum value. Claim 5 delete Claim 6 In claim 1, the photoelectric module comprises a unit cell array composed of a plurality of unit cells, and each of the plurality of unit cells comprises the photodetector, the electronic element, and the light source element. Claim 7 In claim 6, a photoelectric module wherein each of the plurality of unit cells further comprises a resistor that determines the maximum value of the intensity of the output light. Claim 8 In claim 1, the photoelectric module comprises: a photodetector array in which the photodetectors are arranged in an array form; and a light source array in which the light source elements are arranged in an array form. Claim 9 In claim 8, the photodetector array and the light source array are arranged such that the input light and the output light travel in the same direction in a photoelectric module. Claim 10 In claim 9, the photoelectric module comprises: a first PCB substrate including the photodetector array; a second PCB substrate including the light source array and the electronic element; and a connector connecting the first PCB substrate and the second PCB substrate. Claim 11 In claim 9, the photoelectric optical module wherein the photodetector array is disposed on a first surface of a PCB substrate and the light source array is disposed on a second surface of the PCB substrate. Claim 12 In claim 8, the photodetector array and the light source array are arranged such that the input light and the output light travel in opposite directions to each other, forming a photoelectric photovoltaic module. Claim 13 In claim 12, the photodetector array and the light source array are a photoelectric optical module disposed on a first surface of a PCB substrate. Claim 14 An optical artificial neural network system comprising: a linear calculator that receives input light containing input data and performs a linear calculation on the input data to generate a first processed light; and a nonlinear calculator that performs a nonlinear calculation on the data contained in the first processed light to generate an output light having characteristics to which an activation function is applied, wherein the nonlinear calculator comprises: a photodetector that converts the first processed light to generate a first current; an electronic device that amplifies the first current to generate a second current; and a light source device that converts the second current to generate the output light having characteristics to which the activation function is applied, wherein the light source device generates the output light based on the second current and the threshold current of the light source, wherein the photodetector comprises a photodiode, the electronic device comprises at least one of a transistor and an operational amplifier, and the light source device comprises a surface light emitting laser (VCSEL). Claim 15 delete Claim 16 In claim 14, the activation function is a ReLU function, and when the second current is greater than the threshold current, the light source element oscillates such that the intensity of the output light is proportional to the intensity of the second current, in an optical artificial neural network system. Claim 17 In claim 14, the nonlinear calculator further includes a resistor determining the maximum value of the intensity of the output light, the activation function is a sigmoid function, and when the second current is greater than the threshold current, the light source element oscillates such that the output light has an intensity corresponding to the maximum value, an optical artificial neural network system. Claim 18 delete Claim 19 In claim 14, the nonlinear calculator comprises a unit cell array composed of a plurality of unit cells, and each of the plurality of unit cells comprises an optical artificial neural network system including a photodetector, an electronic device, and a light source device. Claim 20 In claim 14, the nonlinear calculator comprises: a photodetector array in which the photodetectors are arranged in an array form; and a light source array in which the light source elements are arranged in an array form, an optical artificial neural network system.
Citation Information
Patent Citations
Device and method for optical neural networks
KR1020190015732A
Optical artificial neural network system
KR1020230122539A
GaAS-based optoelectronic neurons
US5204521A
Artificial neuron with switched-capacitor synapses using analog storage of synaptic weights
US5343555A
