Optical neural network signal sampling device

Abstract

The invention provides an optical neural network signal sampling device, comprising: m optical detectors used for converting m groups of optical signals generated by an optical neural network into corresponding analog current signals, m being an integer greater than or equal to 1; m amplifier groups, each of which is coupled to a corresponding optical detector and is used for converting the analog current signal into an amplified differential voltage signal; the analog-to-digital converter comprises m analog-to-digital conversion channels, and each analog-to-digital conversion channel is coupled to the corresponding amplifier group and is used for converting the amplified differential voltage signal into a digital electric signal; and the logic controller is coupled to the analog-to-digital converter and is used for performing linear processing on each digital electric signal to obtain calculation data of the optical neural network.

Description

Technical Field

[0001] This invention relates to the processing of optical neural network signals, and more specifically to an optical neural network signal sampling device. Background Technology

[0002] With the development of social technology, data centers and AI models are becoming increasingly autonomous, leading to a growing demand for neural network prediction calculations. Since electrons move at speeds far slower than the speed of light, the development of optical neural networks can significantly improve computational speed while reducing the size and energy consumption of related circuits. However, extracting the computational results of optical neural networks from optical signals remains a challenge for those skilled in the art. Summary of the Invention

[0003] In view of this, the present invention provides an optical neural network signal sampling device to at least partially solve the above-mentioned problems.

[0004] This invention provides an optical neural network signal sampling device, comprising: m photodetectors for converting m sets of optical signals generated by the optical neural network into corresponding analog current signals, where m is an integer greater than or equal to 1; m amplifier groups, each amplifier group being coupled to a corresponding photodetector and used to convert the analog current signals into amplified differential voltage signals; an analog-to-digital converter, including m analog-to-digital conversion channels, each analog-to-digital conversion channel being coupled to a corresponding amplifier group and used to convert the amplified differential voltage signals into digital electrical signals; and a logic controller, coupled to the analog-to-digital converter and used to perform linear processing on each digital electrical signal to obtain the computational data of the optical neural network.

[0005] According to an embodiment of the present invention, the logic controller is also coupled to the amplifier group and is also used to monitor the differential voltage signal of each analog-to-digital conversion channel and control the on / off state of the analog-to-digital conversion channel according to the monitoring result.

[0006] According to an embodiment of the present invention, the logic controller includes a register coupled to the analog-to-digital converter, and the register is used to control the on / off state of the analog-to-digital conversion channel based on monitoring results.

[0007] According to an embodiment of the present invention, each amplifier group includes: a transimpedance amplifier, coupled to a corresponding photodetector, for converting a corresponding analog current signal into an analog voltage signal, and amplifying the analog voltage signal through the feedback resistor of the transimpedance amplifier; and a differential amplifier, coupled to the transimpedance amplifier, for converting the amplified analog voltage signal into a differential voltage signal, and amplifying the differential voltage signal.

[0008] According to an embodiment of the present invention, the analog-to-digital converter includes k dual-channel analog-to-digital converters, each dual-channel analog-to-digital converter including two analog-to-digital conversion channels, each dual-channel analog-to-digital converter being coupled to two amplifier groups, where k is an integer greater than or equal to 1.

[0009] According to an embodiment of the present invention, m is 2 to the power of n, where n is an integer greater than or equal to 1; k is half of m, and the m amplifier groups correspond one-to-one with the m analog-to-digital conversion channels of the k dual-channel analog-to-digital converters.

[0010] According to an embodiment of the present invention, the sampling device further includes: a clock regulator coupled to a logic controller and used to generate a high-speed clock signal required by the analog-to-digital converter and a synchronization clock signal for the optical neural network.

[0011] According to an embodiment of the present invention, the sampling device further includes: a host computer, coupled to the logic controller, and used to visualize the calculation data obtained by the logic controller.

[0012] According to an embodiment of the present invention, the sampling device further includes: a memory module coupled between the logic controller and the host computer, the memory module being used to store the calculation data obtained by the logic controller, and the host computer being used to read the calculation data from the memory module and perform visualization display.

[0013] According to an embodiment of the present invention, the photodetector includes at least one of an avalanche photodiode, a photomultiplier tube, or a photodiode.

[0014] Based on the above, the optical neural network signal sampling device of this invention sequentially converts the optical signal generated by the optical neural network into an analog current signal, a differential voltage signal, and a digital electrical signal. Linear processing of the digital electrical signal yields the computational data of the optical neural network, achieving automatic quantization of the optical signal and improving the efficiency of optical neural network signal sampling. Simultaneously, amplification during the signal conversion process facilitates the removal of noise and signal interference, resulting in more accurate computational data, thus achieving high-precision sampling and processing of the optical neural network signal. Furthermore, this sampling device can flexibly adjust the number of photodetectors, amplifiers, and analog-to-digital converters according to the number of optical signal channels, thereby adapting to optical neural networks of different sizes. Attached Figure Description

[0015] The above and other objects, features and advantages of the present invention will become more apparent from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0016] Figure 1 A schematic diagram of a light neural network signal sampling device according to an embodiment of the present invention is shown.

[0017] Figure 2A schematic diagram of a light neural network signal sampling device according to another embodiment of the present invention is shown. Detailed Implementation

[0018] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0019] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0020] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0021] When using expressions such as "at least one of A, B, and C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). When using expressions such as "at least one of A, B, or C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, or C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0022] Figure 1 A schematic diagram of a light neural network signal sampling device according to an embodiment of the present invention is shown.

[0023] like Figure 1As shown, the optical neural network signal sampling device 100 may include m photodetectors 111 and 112. The m photodetectors 111 and 112 can be used to convert the m sets of optical signals generated by the optical neural network into corresponding analog current signals. Converting optical signals into electrical signals makes them easier to quantize and thus more intuitive to present. m is an integer greater than or equal to 1. Figure 1 The diagram schematically illustrates two photodetectors, but this does not imply a limitation on the number of photodetectors. The number of photodetectors, i.e., the value of m, depends on the number of input optical signals to the optical neural network. The number of input optical signals to the optical neural network is typically a power of 2, where n is an integer greater than or equal to 1. For example, m can be 2, 4, 8, etc.

[0024] Before the optical signal reaches photodetectors 111 and 112, it can be preprocessed to meet the minimum detection standards of the photodetectors. The fundamental wavelength of the optical signal can be generated by an external stable laser source, for example, with a wavelength of 1550 nm, or it can be adjusted as needed. The desired signal is then modulated onto the fundamental wavelength. This invention does not limit the generation or number of optical signals. The number of photodetectors 111 and 112 can be flexibly adjusted according to the number of optical signal groups.

[0025] In some embodiments, the photodetector may be at least one of an avalanche photodiode (APD), a photomultiplier tube, or a PIN photodiode.

[0026] Please continue reading. Figure 1 The optical neural network signal sampling device 100 may include m amplifier groups 121 and 122. Figure 1 In this diagram, the number of amplifier groups is for illustrative purposes only; its actual number depends on the number of photodetectors, which in turn depends on the number of optical signal channels. m amplifier groups 121 and 122 are coupled to corresponding photodetectors 111 and 112, respectively, and are used to convert analog current signals into amplified differential voltage signals. That is, amplifier group 121 converts the analog current signal generated by photodetector 111 into a differential voltage signal, and amplifier group 122 converts the analog current signal generated by photodetector 112 into a differential voltage signal. The signals of each channel do not interfere with each other. Through signal amplification processing, weak signals can be amplified and single-ended signals can be converted into differential signals, thereby effectively reducing the difficulty of signal reading and achieving high-quality sampling of weak signals.

[0027] In some embodiments, each amplifier group 121, 122 may include a trans-impedance amplifier (TIA) and a differential amplifier.

[0028] The transimpedance amplifier is coupled to the corresponding photodetectors 111 and 112, and is used to convert the corresponding analog current signal into an analog voltage signal. The analog voltage signal is then amplified through the feedback resistor of the transimpedance amplifier. The feedback resistor can suppress noise amplification, thereby allowing for better differentiation between the analog voltage signal and noise. The specific amplification factor of the transimpedance amplifier can be selected based on the required signal size and bandwidth. This invention does not impose a specific limitation on the signal amplification factor by adjusting the feedback resistor.

[0029] A differential amplifier is coupled to a transimpedance amplifier and is used to convert amplified analog voltage signals into differential voltage signals, and then amplify these differential voltage signals. A differential amplifier can convert single-ended signals into differential signals, increasing signal immunity and reducing signal loss during high-speed propagation.

[0030] The signal generated after photoelectric conversion is a current signal with a relatively small amplitude and a certain degree of noise. A transimpedance amplifier is used for the first stage of amplification, converting the electrical signal into a voltage signal that the ADC can sample. This amplification process also helps suppress noise to some extent. Depending on the system's signal amplitude requirements, the signal bandwidth and amplification factor can be adjusted by designing the feedback resistor. After the first stage of amplification, the signal enters a differential amplifier for the second stage of amplification. This is because the ADC supports differential input, and differential signals ensure signal quality during high-speed transmission. Through these two stages of signal amplifiers, weak signals can be further amplified, and single-ended signals can be converted to differential signals, effectively reducing the difficulty of signal reading and achieving high-quality sampling of weak signals.

[0031] Please continue reading. Figure 1 The optical neural network signal sampling device 100 may include an analog-to-digital converter 130. The analog-to-digital converter 130 includes m analog-to-digital conversion channels A and B. Each analog-to-digital conversion channel A or B is coupled to a corresponding amplifier group 121 or 122 and is used to convert the amplified differential voltage signal into a digital electrical signal. The number of analog-to-digital conversion channels A and B is consistent with the number of optical signal channels to ensure that each analog-to-digital conversion channel A or B processes each corresponding optical signal without interfering with each other. The analog-to-digital conversion channels A and B can be integrated into a multi-channel analog-to-digital converter, or they can be separated into different single-channel analog-to-digital converters, or a combination of a single-channel analog-to-digital conversion chip and a multi-channel analog-to-digital converter, as long as each analog-to-digital conversion channel independently processes each optical signal.

[0032] In some embodiments, analog-to-digital converters A and B may include k dual-channel analog-to-digital converters, each including two analog-to-digital conversion channels, and each dual-channel analog-to-digital converter coupled to two amplifier groups 121 and 122, where k is an integer greater than or equal to 1. For example, when two channels of optical signals need to be processed, a single dual-channel analog-to-digital converter can be used. This reduces the number of analog-to-digital converters while maintaining the same number of channels.

[0033] In some embodiments, the optical neural network signal sampling device 100 may employ dual-channel analog-to-digital converters (ADCs) throughout, with the number of ADCs k being half the number of optical signal channels m. Employing dual-channel ADCs throughout reduces the number of converters and makes control easier and more convenient.

[0034] Please continue reading. Figure 1 The optical neural network signal sampling device 100 may include a logic controller 140. The logic controller 140 is coupled to the analog-to-digital converter 130 and is used to perform linear processing on each digital electrical signal to obtain the computational data for the optical neural network. The logic controller 140 may be, for example, a field-programmable gate array (FPGA). The FPGA can perform linear processing on the sampled digital signals using a maximum value algorithm, using the data with the largest sample value in each group as the computational data for the optical neural network to ensure the accuracy of the computational data.

[0035] In some embodiments, the logic controller 140 may also be coupled to amplifier groups 121, 122. The logic controller 140 may also be used to monitor the differential voltage signal input to each analog-to-digital conversion channel A, B, and control the on / off state of the analog-to-digital conversion channels A, B based on the monitoring results.

[0036] As an example, communication between the logic controller 140 and the analog-to-digital converter 130 can be achieved via the SPI protocol, continuously monitoring the differential voltage signals of analog-to-digital conversion channels A and B. If no data is read from analog-to-digital conversion channel A or B for a prolonged period, it is assumed that there is no signal input to that channel, and the logic controller 140 can shut down the channel to reduce the power consumption of the analog-to-digital converter. If the logic controller 140 detects that the channel A or B continuously receives signals, such as non-long zero data, it is determined that there is a signal in the channel, and the logic controller 140 can turn on the channel to continue the analog-to-digital conversion. In this way, the power consumption of the analog-to-digital converter can be reduced while ensuring its normal operation.

[0037] In some embodiments, the logic controller 140 may include a register. The register is coupled to the analog-to-digital converter 130 and is used to control the on / off state of the analog-to-digital conversion channel based on monitoring results. For example, when no data is detected in the channel, the channel can be turned off by writing to the register controlling the channel switch. When a signal input is detected, the channel can be turned on by writing to the register controlling the channel switch, achieving adaptive channel switching and more effectively saving energy.

[0038] In some embodiments, the optical neural network signal sampling device 100 may further include a clock conditioner. The clock conditioner is coupled to the logic controller 140 and is used to generate a high-speed clock signal required by the analog-to-digital converter 130 and a synchronization clock signal for the optical neural network. The clock conditioner can communicate with the logic controller 140 via the SPI protocol. The high-speed clock signal ensures high accuracy and high speed in the analog-to-digital conversion process, while the synchronization clock signal ensures timing consistency among the various components within the network.

[0039] In some embodiments, the optical neural network signal sampling device 100 may further include a host computer coupled to the logic controller 140, and used to visualize the computational data obtained by the logic controller 140. Through the graphical interface of the host computer, users can intuitively obtain the computational data obtained by the optical neural network signal sampling device 100, which facilitates subsequent analysis and decision-making.

[0040] In some embodiments, the optical neural network signal sampling device 100 may further include a memory module coupled between the logic controller 140 and the host computer. The memory module stores the computational data obtained by the logic controller 140, and the host computer reads the computational data from the memory module and displays it visually. The memory module can store the computational data output by the logic controller 140, providing a buffer for data access for the host computer. This helps to reduce the data processing pressure on the logic controller 140 and allows the host computer to read the data at any time when needed. The communication interface in the optical neural network signal sampling device 100 that communicates with the host computer includes, but is not limited to, USB 3.0, SFP optical module, PCIe, and UART. The relevant protocols can be edited through the FPGA to realize data exchange between the host computer and the circuit board.

[0041] Based on the above embodiments, the following describes an optical neural network signal sampling device 200 comprising four photodetectors (APDs), four transimpedance amplifiers (TIAs), four differential amplifiers, and two dual-channel analog-to-digital converters (ADCs), and in conjunction with... Figure 2 The optical neural network signal sampling device 200 will be further described. It is important to emphasize that... Figure 2 The optical neural network signal sampling device 200 shown is merely an example and does not constitute a limitation of the present invention. Figure 2 A schematic diagram of a light neural network signal sampling device according to another embodiment of the present invention is shown.

[0042] like Figure 2 As shown, the optical neural network signal sampling device 200 includes four photodetectors (APDs), four transimpedance amplifiers (TIAs), four differential amplifiers, two dual-channel analog-to-digital converters (ADCs), an FPGA, a memory module, a communication interface, and a host computer. The optical neural network signal sampling device 200 can support up to four signal transmissions. That is, the number of input optical signal channels can be four or less.

[0043] The optical neural network signal sampling device 200 can include processes such as initial preparation, photoelectric conversion, secondary amplification of electrical signals, analog-to-digital conversion, signal generation, storage, and retrieval. When the system is powered on, the optical neural network signal sampling device 200 completes an initial reset. Its registers can be configured via the Vitis project to ensure normal operation. For example, data can be written to relevant registers via the FPGA to configure the ADC's clock input, rate adjustment, and power-on reset initialization operations. After register configuration, the optical neural network signal sampling device 200 can process the optical neural network calculation results (optical signals). In the secondary signal amplification stage, after the optical signals from the four channels are converted into electrical signals by the four photodetectors (APDs), they enter the circuit system through the SMA (Subminiature Version A) interface. They are amplified by four transimpedance amplifiers (TIAs) and four differential amplifiers, respectively, and then enter the analog-to-digital conversion stage. The amplified four differential voltage signals are transmitted to the four analog-to-digital conversion channels of the two dual-channel analog-to-digital converters (ADCs) for sampling and conversion into digital signals. During ADC operation, by detecting changes in voltage signals in each channel, the switching of relevant ADC channels can be controlled, thereby controlling ADC power consumption. Finally, the data is linearly processed by the FPGA and stored in the memory module (DDR). The host computer then reads and visualizes the data through a communication interface.

[0044] In the optical neural network signal sampling device 200, the FPGA can have multiple functions: it can configure chip-related registers to ensure the normal startup and operation of the chip; it can provide a unified clock source to ensure signal synchronization; it can perform linear processing on the digital signals acquired by the ADC to ensure the accuracy of the acquired signals as much as possible, and store the processed data into the corresponding memory module; it can also transmit data with the host computer for result comparison.

[0045] Based on the above embodiments, the optical signal generated by the optical neural network is sequentially converted into an analog current signal, a differential voltage signal, and a digital electrical signal. Linear processing of the digital electrical signal yields the computational data for the optical neural network, achieving automatic quantization of the optical signal and improving the efficiency of optical neural network signal sampling. By controlling the ADC enable register, the switching state can be adaptively controlled according to the signal state, effectively reducing circuit board power consumption. A two-stage signal amplifier amplifies weak signals and converts single-ended signals to differential signals, effectively reducing the difficulty of signal reading and achieving high-quality sampling of weak signals. Furthermore, the number of photodetectors, amplifiers, and analog-to-digital converters can be flexibly adjusted according to the number of optical signal channels to adapt to optical neural networks of different scales. In summary, the multi-channel adaptive optical neural network signal sampling device provided by this invention features high signal sampling accuracy, high speed, and low power consumption, and can be used as a readout circuit for high-speed optical neural network computation results.

[0046] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A signal sampling device for an optical neural network, characterized in that, include: m photodetectors are used to convert m sets of optical signals generated by the optical neural network into corresponding analog current signals, where m is an integer greater than or equal to 1; m amplifier groups, each of which is coupled to a corresponding photodetector and is used to convert the analog current signal into an amplified differential voltage signal; An analog-to-digital converter includes m analog-to-digital conversion channels, each of which is coupled to a corresponding amplifier group and is used to convert the amplified differential voltage signal into a digital electrical signal. A logic controller, coupled to the analog-to-digital converter, is used to perform linear processing on each of the digital electrical signals to obtain the computational data of the optical neural network.

2. The sampling device according to claim 1, characterized in that, The logic controller is also coupled to the amplifier group and is also used to monitor the differential voltage signal input to each of the analog-to-digital conversion channels, and control the on / off state of the analog-to-digital conversion channels according to the monitoring results.

3. The sampling device according to claim 2, characterized in that, The logic controller includes a register coupled to the analog-to-digital converter, and the register is used to control the on / off state of the analog-to-digital conversion channel based on the monitoring results.

4. The sampling device according to claim 1, characterized in that, Each of the amplifier groups includes: A transimpedance amplifier, coupled to the corresponding photodetector, is used to convert the corresponding analog current signal into an analog voltage signal, and to amplify the analog voltage signal through the feedback resistor of the transimpedance amplifier. A differential amplifier is coupled to the transimpedance amplifier and is used to convert the amplified analog voltage signal into a differential voltage signal and amplify the differential voltage signal.

5. The sampling device according to claim 1, characterized in that, The analog-to-digital converter includes k dual-channel analog-to-digital converters, each of which includes two analog-to-digital conversion channels, and each of which is coupled to two amplifier groups, where k is an integer greater than or equal to 1.

6. The sampling device according to claim 5, characterized in that, m is 2 to the power of n, where n is an integer greater than or equal to 1; k is half the value of m, and the m amplifier groups correspond one-to-one with the m analog-to-digital conversion channels of the k dual-channel analog-to-digital converters.

7. The sampling device according to claim 1, characterized in that, Also includes: A clock regulator, coupled to the logic controller, is used to generate the high-speed clock signal required by the analog-to-digital converter and the synchronization clock signal of the optical neural network.

8. The sampling device according to claim 1, characterized in that, Also includes: The host computer is coupled to the logic controller and is used to visualize the calculation data obtained by the logic controller.

9. The sampling device according to claim 8, characterized in that, Also includes: A memory module is coupled between the logic controller and the host computer. The memory module is used to store the calculation data obtained by the logic controller, and the host computer is used to read the calculation data from the memory module and perform visualization display.

10. The sampling device according to claim 1, characterized in that, The photodetector includes at least one of an avalanche photodiode, a photomultiplier tube, or a photodiode.

Images