A micro-led optical interconnection module

Abstract

The application discloses a micro-LED optical interconnection module, which comprises an emitting unit, a transmission unit, a receiving unit and a self-adaptive mutual feedback regulation module. The emitting unit emits a multi-wavelength optical signal through a Micro-LED array prepared on a wavelength-tunable epitaxial wafer; the transmission unit realizes wavelength division multiplexing transmission through a single optical fiber; the receiving unit realizes parallel receiving and separation of the multi-wavelength signal through a CMOS array cooperating with a pixel-level narrow-band band-pass filter film layer. The self-adaptive mutual feedback regulation module is introduced, the module constitutes a closed-loop control loop of "receiving end sensing-algorithm decision-emitting end execution", the emitting end, the transmission link and the receiving end form an intelligent whole, and the transmission quality, long-term stability and environmental adaptability of the system are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of optical interconnect technology, and more specifically to a Micro-LED optical interconnect module. Background Technology

[0002] With the rapid development of big data, artificial intelligence, and high-speed computing, the demand for high-speed interconnection between chips and boards is becoming increasingly urgent. Traditional electrical interconnection methods have inherent drawbacks such as bandwidth bottlenecks, electromagnetic interference (EMI), and high power consumption, making it difficult to meet the application requirements of ultra-high-speed and high-density interconnection. Optical interconnection technology, due to its advantages such as high bandwidth, low loss, and strong anti-interference capabilities, has become the core solution to the above problems.

[0003] In existing optical interconnect modules, the light source mostly uses laser diodes (LDs) or ordinary fixed-wavelength Micro-LED arrays. While LD light sources have the advantages of narrow linewidth and high modulation rate, they also suffer from high cost, require precise temperature control and complex driving circuits, and are difficult to integrate with silicon-based chips. Furthermore, they are not economically viable in short-distance optical interconnect scenarios. Although ordinary fixed-wavelength Micro-LED arrays have lower cost and higher integration, the fixed emission wavelength of each Micro-LED requires multiple independent light sources and multiple optical fibers to achieve multi-channel signal transmission. This results in increased module size, higher integration difficulty, and significantly higher transmission costs.

[0004] At the receiving end, existing solutions mostly use a single-channel detector in conjunction with wavelength demultiplexing devices (such as gratings and filter arrays) to achieve multi-wavelength signal separation. Such solutions have problems such as complex structure, large size, and poor compatibility with CMOS process, making it difficult to achieve high-density integration at the receiving end and restricting the miniaturization and low-cost development of optical interconnect modules.

[0005] More importantly, existing optical interconnect modules lack an effective closed-loop feedback mechanism between the transmitter, transmission link, and receiver, operating in an "open-loop" mode. In practical applications, due to time-varying characteristics such as dispersion, insertion loss fluctuations, and temperature drift in optical fiber transmission channels, as well as wavelength drift and optical power attenuation caused by aging of Micro-LED devices, multi-channel wavelength division multiplexing systems in open-loop mode cannot dynamically adapt to changes in channel characteristics. This leads to increased crosstalk between channels, decreased signal-to-noise ratio, and increased bit error rate, severely affecting the long-term stability and transmission reliability of the system. Furthermore, the power balance between multiple wavelength channels lacks automatic adjustment capability. When there are significant differences in received power between different channels, the signal quality of some channels deteriorates, and the overall transmission performance is limited by the worst-performing channel.

[0006] Therefore, developing a highly integrated, low-cost, multi-channel intelligent optical interconnect module with closed-loop adaptive feedback control capabilities and the ability to achieve coordination among the transmitter, transmission link, and receiver has become an urgent need in the field of optical interconnect technology. Summary of the Invention

[0007] To address the shortcomings of existing optical interconnect modules, such as low integration, high cost, difficulty in multi-channel transmission, and lack of closed-loop feedback control mechanisms, this invention provides a micro-LED optical interconnect module. Through the collaborative design of a wavelength-tunable Micro-LED transmitting array, wavelength division multiplexing transmission, CMOS filter receiving array, and adaptive mutual feedback control module, it achieves high-density, low-cost, high-speed transmission and accurate identification and extraction of multi-channel signals. Furthermore, through an algorithm-driven closed-loop feedback mechanism, the transmitter, transmission link, and receiver are integrated into a whole, enabling real-time adaptive optimization of system performance and significantly improving the system's transmission quality, stability, and environmental adaptability.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A micro-LED optical interconnect module includes a transmitting unit, a transmitting unit, a receiving unit, and an adaptive mutual feedback control module. The emitting unit includes a wavelength-tunable epitaxial wafer (DUT), a Micro-LED array, and a driving control circuit. The Micro-LED array is directly fabricated on the wafer of the same wavelength-tunable epitaxial wafer (DUT). The driving control circuit is electrically connected to the Micro-LED array and is used to control different Micro-LED chips in the Micro-LED array to emit light signals of different wavelengths through circuit modulation, and the wavelengths do not overlap.

[0009] Furthermore, the wavelength-tunable epitaxial wafer of the DUT is fabricated using III-V compound semiconductor materials. Its epitaxial structure includes a substrate, a buffer layer, an N-type confinement layer, a quantum well active region, a P-type confinement layer, and an ohmic contact layer. By adjusting the composition, thickness, or doping concentration of the quantum well active region, and in conjunction with the voltage / current modulation of the driving control circuit, the emission wavelength of the Micro-LED chip can be continuously or discretely tunable, with a tunable wavelength range of 400nm-1650nm.

[0010] Furthermore, in the Micro-LED array, the size of each Micro-LED chip is 1-100μm, the array density is 400-25 million / cm², and the emission wavelength interval of each chip is 1-100nm, ensuring that there is no crosstalk during the transmission of multi-wavelength signals and meeting the wavelength division multiplexing transmission requirements.

[0011] Furthermore, the driving control circuit is fabricated using CMOS driving technology and integrated with the Micro-LED array on the same wafer to achieve monolithic integration of the emitting unit; the driving control circuit controls the emission wavelength and optical power of each Micro-LED chip, with a modulation rate of up to 100Mbps-100Gbps.

[0012] Furthermore, the drive control circuit also integrates a feedback response submodule, which receives control commands from the adaptive mutual feedback control module and adjusts the drive voltage / current parameters of each Micro-LED chip in real time to achieve dynamic calibration of the emission wavelength and adaptive equalization of the emission power.

[0013] The transmission unit includes a single transmission optical fiber. The input end of the transmission optical fiber is coupled to the optical path output end of the transmitting unit to receive multi-wavelength mixed optical signals output by the transmitting unit. Through wavelength division multiplexing technology, the synchronous transmission of multi-wavelength optical signals within the same transmission optical fiber is achieved.

[0014] Furthermore, the transmission optical fiber is a single-mode or multi-mode optical fiber with a numerical aperture of 0.1-0.3, a core diameter of 2-100μm, and a transmission loss of ≤0.5dB / km, ensuring low-loss transmission of multi-wavelength optical signals during transmission without significant dispersion interference.

[0015] Furthermore, the transmission unit also includes a transmission channel feature monitoring interface, which is used in conjunction with the adaptive mutual feedback control module to obtain the channel state information of the transmission channel at the optical fiber output end, and to provide transmission link feature data for feedback control.

[0016] The receiving unit includes a CMOS array, a narrowband bandpass filter layer, and a signal processing circuit. The narrowband bandpass filter layer covers the photosensitive surface of the CMOS array, and the narrowband bandpass filter layer corresponds one-to-one with the pixels of the CMOS array. In the narrowband bandpass filter layer, the center wavelength of the filter film corresponding to different pixels corresponds one-to-one with the wavelengths emitted by the Micro-LED array in the transmitting unit, which is used to allow only the light signal of the corresponding wavelength to pass through and illuminate the corresponding CMOS pixel. The signal processing circuit is electrically connected to the CMOS array and is used to perform photoelectric conversion and signal extraction on the light signals received by each CMOS pixel to realize the separation and identification of signals of different bands.

[0017] Furthermore, the narrowband bandpass filter film is prepared by vacuum evaporation or ion beam sputtering, with a film thickness of 100-1000nm, a full width at half maximum (FWHM) of ≤10nm for each filter film, and a transmittance of ≥70%, ensuring high selective transmission of corresponding wavelength optical signals and effectively suppressing stray light and crosstalk of adjacent wavelength signals.

[0018] Furthermore, the CMOS array is fabricated using complementary metal-oxide-semiconductor technology, with a pixel size of 1-50μm. The array size is consistent with the Micro-LED array size of the transmitting unit, and each CMOS pixel corresponds one-to-one with the corresponding Micro-LED chip and narrowband filter film, realizing parallel reception and recognition of multi-channel signals.

[0019] Furthermore, the signal processing circuit includes a preamplifier, a filter circuit, an analog-to-digital converter circuit, and a signal demodulation circuit, which are used to convert the analog electrical signals output by the CMOS pixels into digital signals, and extract the original data signals corresponding to each wavelength through a demodulation algorithm to achieve the separation and output of multi-channel signals.

[0020] Furthermore, the receiving unit also includes a channel quality assessment submodule, which is electrically connected to the signal processing circuit and is used to collect the signal quality parameters of each wavelength channel in real time, and transmit the signal quality parameters to the adaptive mutual feedback control module as the input data source for feedback control.

[0021] The adaptive mutual feedback control module is communicatively connected to the feedback response submodule of the transmitting unit and the channel quality assessment submodule of the receiving unit, forming a closed-loop feedback control loop of "receiver perception → algorithm decision → transmitter execution". The adaptive mutual feedback control module includes a channel state perception engine, a channel matrix estimation and crosstalk decoupling algorithm unit, a multi-objective joint optimization decision unit, and a feedback command generation and transmission unit. The four subunits work together to realize the correlation and adaptive collaborative optimization of the transmitter, transmission link and receiver.

[0022] 1. Channel State Awareness Engine The channel state awareness engine receives signal quality parameters for each wavelength channel from the channel quality evaluation submodule of the receiving unit. These signal quality parameters include, but are not limited to, the received optical power P of each channel. r,i (where i is the channel number), Signal-to-noise ratio (SNR) i Bit error rate (BER) i Inter-channel crosstalk CT ij (Crosstalk contribution of channel j to channel i) and the received signal wavelength offset Δλ of each channel. i The channel state awareness engine preprocesses the above raw parameters, including data filtering and denoising (using Kalman filtering or sliding window mean filtering), normalization, and time series caching, to form a standardized channel state vector:

[0023] ; Where N is the total number of wavelength division multiplexing channels, and t is the current time step. The channel state vector S(t) simultaneously caches historical data from the most recent M time steps, forming a state-time series. It is used to support trend prediction and decision-making in subsequent algorithm units.

[0024] 2. Channel Matrix Estimation and Crosstalk Decoupling Algorithm Unit The channel matrix estimation and crosstalk decoupling algorithm unit constructs and updates the system's channel transmission matrix H in real time based on the channel state data output by the channel state awareness engine, thereby achieving an accurate characterization of the coupling relationship between each wavelength channel.

[0025] The channel transmission matrix H is defined as an N×N dimensional matrix, where the element h in the i-th row and j-th column is... ij The channel transmission matrix represents the response coefficient (including the direct transmission coefficient and crosstalk coefficient) of the optical signal in the j-th channel at the transmitting end in the i-th channel at the receiving end. The channel transmission matrix satisfies the received signal model: Y(t) = H(t) × X(t) + N(t); in This represents the received signal strength vector for each channel at the receiving end. Let N(t) be the transmitted signal strength vector of each channel at the transmitter, and let N(t) be the noise vector.

[0026] The channel matrix estimation employs either the adaptive minimum mean square error (MMSE) algorithm or the recursive least squares (RLS) algorithm for online estimation. Specifically, during the system initialization phase or the periodic calibration phase, the transmitter performs online estimation according to a preset training sequence X. train A known signal is transmitted in a channel-by-channel or orthogonal code division manner, and the receiver acquires the corresponding received signal Y. train The channel matrix is ​​estimated using the following formula:

[0027] Where σ 2 Let be the noise variance estimate, and I be the identity matrix. During normal system operation, a decision feedback method is adopted, using the decision data demodulated by the signal processing circuit as a reference, and the channel matrix is ​​recursively updated using the RLS algorithm:

[0028] ; Where K(t) is the RLS gain matrix. This is the estimated value of the transmitted signal for the decision feedback. By continuously updating the channel matrix, the system can track the time-varying changes in the characteristics of the transmission channel in real time.

[0029] Based on the estimated channel matrix The crosstalk decoupling algorithm calculates the crosstalk strength index for each channel: in This is the crosstalk strength index for channel i. A higher value indicates more severe crosstalk from adjacent channels. The crosstalk decoupling algorithm modifies the crosstalk strength index against a preset crosstalk threshold CI. th For CI i (t)>CI th The channel is marked as "crosstalk degraded channel" and the relevant information is transmitted to the multi-objective joint optimization decision unit.

[0030] 3. Multi-objective joint optimization decision-making unit The multi-objective joint optimization decision unit receives the channel matrix output by the channel matrix estimation and crosstalk decoupling algorithm unit. Crosstalk intensity index The system uses the channel state vector S(t) output by the channel state awareness engine and its historical time series as inputs to execute a multi-objective joint optimization algorithm to generate the optimal control parameters for the transmitter. The multi-objective joint optimization algorithm includes the following three cooperating optimization sub-algorithms:

[0031] (a) Wavelength dynamic calibration optimization sub-algorithm The wavelength dynamic calibration optimization sub-algorithm is used to compensate for the emission wavelength drift of Micro-LED chips caused by factors such as temperature changes and aging. The sub-algorithm is based on the wavelength offset Δλ of the received signal for each channel provided by the channel state awareness engine. i (t), and historical wavelength offset time series A prediction model based on Long Short-Term Memory (LSTM) networks is used to predict the wavelength drift trend at the next time step. Based on this, the wavelength calibration compensation amount δV is generated. λ,i (The adjustment amount corresponding to the driving voltage) minimizes the calibrated emission wavelength offset. The wavelength calibration objective function is:

[0032] in Let be the wavelength-voltage sensitivity coefficient of the i-th channel. This coefficient is obtained through initial calibration and continuously updated through online identification during operation. Here, "-0" does not represent a mathematical meaning, but rather indicates that the optimization objective is zero offset.

[0033] (b) Power Adaptive Equalization Optimization Sub-algorithm The power adaptive equalization optimization sub-algorithm is used to eliminate the received power imbalance caused by differences in transmission link loss, coupling efficiency, and filter film transmittance between wavelength channels. The sub-algorithm uses the consistency of the signal-to-noise ratio at the receiver of each channel as the optimization objective, and defines the power equalization objective function as follows:

[0034] Where δP t,i This represents the adjustment amount for the transmit power of the i-th channel. The power-to-signal-noise ratio sensitivity coefficient of the i-th channel (obtained through the channel matrix and noise level), SNR target The target signal-to-noise ratio is set for the system, and μ is a regularization coefficient (used to constrain the power adjustment amplitude and avoid over-adjustment causing system oscillation). The objective function is solved analytically to obtain the optimal power adjustment amount.

[0035] (c) Crosstalk suppression and channel spacing optimization sub-algorithm The crosstalk suppression and channel spacing optimization sub-algorithm is used to reduce inter-channel crosstalk by dynamically adjusting the transmission wavelength spacing of relevant channels for crosstalk-degraded channels. The sub-algorithm is based on the crosstalk coefficient h in the channel matrix. ij Construct the crosstalk cost function:

[0036] Where w ij For crosstalk weighting coefficients, w corresponds to the crosstalk degradation channel. ij A larger value is selected to enhance the optimization effect. Under the constraint of an adjustable wavelength spacing range, the sub-algorithm searches for the optimal wavelength allocation scheme using either a particle swarm optimization (PSO) algorithm or a genetic algorithm. This minimizes the crosstalk cost function while satisfying the constraint that the wavelengths of each channel do not overlap:

[0037] Where Δλ min The minimum wavelength spacing to ensure crosstalk-free transmission (determined by the full width at half maximum of the filter layer).

[0038] The multi-objective joint optimization decision unit integrates and coordinates the outputs of the above three sub-algorithms, and uses a weighted multi-objective optimization method to define the system's comprehensive performance index function: ; J λ J power J CTThese represent the objective function values ​​for wavelength calibration, power equalization, and crosstalk cost, respectively. ω1, ω2, and ω3 are the weighting coefficients for each objective, which can be dynamically adjusted according to the actual application scenario and priority of the system. The multi-objective joint optimization decision unit weighs the three objectives through Pareto optimality search or hierarchical optimization strategies, ultimately generating a set of transmitter control parameters. .

[0039] 4. Feedback instruction generation and transmission unit The feedback command generation and transmission unit encodes the control parameter set C(t) output by the multi-objective joint optimization decision unit into a standardized feedback command frame, which is then transmitted to the feedback response submodule of the transmitting unit via a feedback communication link. The feedback communication link can employ one of the following methods: (a) an electrical feedback channel, using low-speed electrical interconnects within the module to transmit the feedback command; (b) an optical feedback channel, using the reverse transmission direction of the transmission fiber or a reserved auxiliary fiber channel to transmit the feedback signal; or (c) a wireless feedback channel, using short-range wireless communication (such as Bluetooth Low Energy, UWB, etc.) to transmit the feedback command. The feedback command frame includes the control target channel number, wavelength adjustment amount, power adjustment amount, and priority identifier. The update period of the feedback command is configurable, ranging from 1μs to 100ms, and can be adjusted according to the system's real-time requirements.

[0040] Furthermore, the adaptive mutual feedback control module also includes an anomaly detection and protection subunit, which monitors in real time whether the BER of each channel exceeds a preset safety threshold BER. safe When any channel BER is detected i (t)>BER safe When the channel quality returns to normal, the emergency control mode is triggered. In this mode, wavelength calibration and power boost are performed on the channel first, and the modulation rate of the channel is reduced to ensure transmission reliability. Once the channel quality returns to normal, the channel will automatically return to the normal control mode.

[0041] Furthermore, the algorithm running platform of the adaptive mutual feedback control module is a low-power embedded processor (such as a RISC-V core or ARM Cortex-M series) integrated into the receiver or an FPGA chip integrated into the signal processing circuit. The processor / FPGA runs all the above algorithms, and during the long-term operation of the system, it continuously trains and optimizes the parameters of the LSTM wavelength drift prediction model through accumulated historical channel data and control effect data, so as to realize the online self-learning and continuous evolution of the model.

[0042] 3.3 Beneficial Effects Compared with the prior art, the present invention has the following beneficial effects: 1. High integration and small size: The Micro-LED array of the transmitting unit is directly fabricated on the same wavelength-tunable epitaxial wafer of the DUT, and the driving control circuit is monolithically integrated with the Micro-LED array; the narrowband bandpass filter film of the receiving unit is integrated with the CMOS array, eliminating the need for additional wavelength demultiplexing devices, which greatly simplifies the module structure, improves the integration, reduces the module size, and is suitable for miniaturized and high-density interconnection scenarios.

[0043] 2. Low cost and simple fabrication: The use of wavelength-tunable epitaxial wafers (DUTs) eliminates the need to fabricate Micro-LED chips and wafers separately for different wavelengths, reducing chip fabrication costs. The receiver uses a CMOS array with a narrowband filter layer, replacing the traditional dedicated photodetector and complex demultiplexing structure. Furthermore, the CMOS process is mature and inexpensive, significantly reducing the overall fabrication cost and difficulty of the module.

[0044] 3. High efficiency of multi-channel transmission: By modulating the emission wavelength of the Micro-LED array through the driving control circuit, the synchronous output of multi-wavelength signals is realized. Combined with wavelength division multiplexing transmission of a single transmission fiber, high-speed transmission of multi-channel signals can be achieved without the need for multiple optical fibers, which greatly improves the transmission bandwidth and transmission efficiency, and effectively reduces the cost of the transmission link.

[0045] 4. Accurate signal recognition and strong anti-interference capability: The receiver adopts a pixel-level narrowband bandpass filter film layer that corresponds one-to-one with the transmitted wavelength, allowing only the corresponding wavelength of light signal to pass through, which can effectively suppress stray light and crosstalk of adjacent wavelength signals; combined with the parallel reception of the CMOS array and the demodulation optimization of the signal processing circuit, it can realize the accurate identification and extraction of signals of different bands, and improve the stability and reliability of signal transmission.

[0046] 5. Strong compatibility and good scalability: The driving control circuit of the transmitting unit adopts CMOS technology, which is compatible with existing silicon-based chip technology and is easy to integrate with other integrated circuits; the wavelength-tunable epitaxial wafer of the DUT supports adjustable wavelength range, and the emission wavelength and number of channels of the Micro-LED array can be flexibly adjusted according to actual transmission requirements, which is highly scalable.

[0047] 6. Closed-loop adaptive feedback ensures strong system robustness: A closed-loop control loop of "receiver sensing → algorithm decision → transmitter execution" is constructed through an adaptive mutual feedback control module. The channel matrix estimation and crosstalk decoupling algorithm can track the time-varying characteristics of the transmission channel in real time, and the multi-objective joint optimization algorithm can simultaneously achieve the three optimization objectives of wavelength calibration, power equalization, and crosstalk suppression. This enables the system to automatically compensate and adjust when faced with disturbances such as temperature drift, device aging, and changes in fiber bending loss, significantly improving the long-term operational stability and transmission reliability of the system under complex actual conditions.

[0048] 7. Algorithm-driven intelligent optimization capability: The adaptive mutual feedback control module introduces an LSTM neural network prediction model, which can predictively compensate for wavelength drift trends based on historical data, enabling early prediction and control. Compared with traditional threshold-triggered feedback control, it has a faster response, smoother control, and smaller system oscillations. At the same time, the online self-learning mechanism of the LSTM model enables the system to continuously accumulate experience during long-term operation, continuously improve performance optimization, and has self-evolution capability.

[0049] 8. Multi-layered protection and high reliability: The adaptive mutual feedback control module is equipped with an anomaly detection and protection sub-unit. When the system detects an abnormal increase in the channel bit error rate, it can automatically switch to emergency control mode to prioritize the availability of the communication link and avoid overall communication interruption due to local channel degradation, which greatly improves the system's fault tolerance and operational reliability. Attached Figure Description

[0050] Figure 1 This is an overall schematic diagram of the micro-LED optical interconnect module in an embodiment of the present invention; Figure 2 This is a schematic diagram of the transmitting unit in an embodiment of the present invention; Figure 3 This is a schematic diagram of the receiving unit in an embodiment of the present invention; Figure 4 This is a schematic diagram of the functional architecture and data flow of the adaptive mutual feedback control module in an embodiment of the present invention; Figure 5 This is a schematic diagram of the working process of the closed-loop feedback control loop in an embodiment of the present invention; Figure 6 This is a flowchart illustrating the logic of the channel matrix estimation and multi-objective joint optimization algorithm in this embodiment of the invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1 like Figure 1 As shown, this embodiment provides a micro-LED optical interconnect module, including a transmitting unit 1, a transmitting unit 2, a receiving unit 3, and an adaptive mutual feedback control module 4. The four work together to form a closed-loop intelligent optical interconnect system, realizing the transmission, reception, identification, and adaptive optimization of multi-channel optical signals.

[0053] Transmission unit like Figure 2 As shown, the emitting unit 1 includes a wavelength-tunable epitaxial wafer 11 (DUT), a Micro-LED array 12, a driving control circuit 13, and a feedback response submodule 14. The wavelength-tunable epitaxial wafer 11 is fabricated using GaN-based III-V compound semiconductor materials. Its epitaxial structure, from bottom to top, consists of a sapphire substrate, an AlN buffer layer, an N-type GaN confinement layer, an InGaN / GaN quantum well active region, a P-type GaN confinement layer, and a Ni / Au ohmic contact layer. By adjusting the In composition (range 0.1-0.5) and quantum well thickness (range 2-5 nm) of the InGaN quantum well active region, and in conjunction with voltage modulation by the driving control circuit 13, the emission wavelength of the Micro-LED chip can be discretely tunable, with a tunable wavelength range of 450 nm to 650 nm.

[0054] The Micro-LED array 12 is directly fabricated on the same wavelength-tunable epitaxial wafer 11 via photolithography, etching, and ohmic contact, eliminating the need for subsequent transfer processes and simplifying the fabrication process. The Micro-LED array 12 has an 8×8 array size, comprising 64 Micro-LED chips, each with a size of 10μm×10μm, resulting in an array density of 250,000 chips / cm², corresponding to a double spacing. The driving control circuit 13 is fabricated using CMOS driving technology and integrated with the Micro-LED array 12 on the same wafer, achieving monolithic integration of the emitting unit 1. The driving control circuit 13 controls each Micro-LED chip to emit light signals of different wavelengths. The initial wavelength spacing between each chip is 30nm or 2nm. Taking 30nm as an example, the emission wavelengths are 450nm, 480nm, 510nm, 540nm, 570nm, 600nm, 630nm, and 650nm (each wavelength corresponds to 8 chips), ensuring no crosstalk between the multi-wavelength signals and a modulation rate of 1Gbps.

[0055] The feedback response submodule 14 is integrated into the drive control circuit 13 and is fabricated synchronously with the drive control circuit 13 using CMOS technology. It includes a feedback command decoder and a parameter register set. The feedback command decoder receives standardized feedback command frames from the adaptive mutual feedback control module 4 and parses the wavelength adjustment amount δV within them. λ,i Power adjustment amount δP t,i The parameters are assigned priority identifiers and then written to the parameter register of the corresponding channel. The drive control circuit 13 reads the latest parameters from the parameter register in each modulation cycle and superimposes them onto the reference drive voltage / current to achieve real-time dynamic adjustment of the emission wavelength and optical power of each Micro-LED chip.

[0056] Transmission unit The transmission unit 2 includes a single multimode transmission fiber 21. The transmission fiber 21 is made of PMMA material with a numerical aperture of 0.2, a core diameter of 62.5 μm, a cladding diameter of 125 μm, and a transmission loss of ≤0.3 dB / km. The input end of the transmission fiber 21 is coupled to the optical output end of the transmitting unit 1 through a microlens, with a coupling efficiency of ≥85%. It is used to receive 64 mixed optical signals of different wavelengths output by the transmitting unit 1. Through wavelength division multiplexing technology, the 64 optical signals are synchronously transmitted within the same transmission fiber 21 without the need for multiple optical fibers, which greatly reduces the cost of the transmission link.

[0057] Receiving unit like Figure 3 As shown, the receiving unit 3 includes a CMOS array 31, a narrowband bandpass filter layer 32, a signal processing circuit 33, and a channel quality evaluation submodule 34. The CMOS array 31 is fabricated using a 0.18μm CMOS process, with an array size identical to the Micro-LED array 12, consisting of 8×8 pixels (64 pixels in total). Each pixel is 10μm×10μm in size, and the pixel pitch is 10μm. The narrowband bandpass filter layer 32 is fabricated on the photosensitive surface of the CMOS array 31 using a vacuum evaporation process. The layer thickness is 500nm. In the narrowband bandpass filter layer 32, the center wavelength of the filter film corresponding to each CMOS pixel corresponds one-to-one with the eight wavelengths emitted by the Micro-LED array 12 in the transmitting unit 1. Each filter film has a full width at half maximum (FWHM) of 5nm and a transmittance ≥75%, ensuring that only the corresponding wavelength of light signal is allowed to pass through and illuminate the corresponding CMOS pixel, effectively suppressing stray light and crosstalk between adjacent wavelength signals.

[0058] The signal processing circuit 33 is electrically connected to the CMOS array 31 via metal wires. The signal processing circuit 33 includes a preamplifier (gain of 60dB), an RC filter circuit (cutoff frequency of 2GHz), a 12-bit analog-to-digital converter circuit (sampling rate of 5GSps), and an FPGA signal demodulation circuit. The CMOS array 31 converts the received optical signal into an analog electrical signal. After being amplified by the preamplifier, filtered by the RC filter circuit, and converted into a digital signal by the analog-to-digital converter circuit, the FPGA signal demodulation circuit extracts the original data signal corresponding to each wavelength through a demodulation algorithm, realizing the separation, identification, and output of 64 different band signals.

[0059] The channel quality assessment submodule 34 is integrated into the FPGA chip of the signal processing circuit 33, and acquires the signal quality parameters of the following 8 wavelength channels (N=8) in real time: the received optical power P of each channel. r,i (Obtained through photocurrent amplitude of CMOS pixels), Signal-to-noise ratio (SNR) i(Calculated by the ratio of the demodulated signal power to the noise floor), Bit Error Rate (BER) i (Obtained through bit error statistics of known training sequences, or estimated through the number of correction bits in the forward error correction code), and the wavelength offset Δλ of the received signal. i (Whether the transmitted wavelength deviates from the center wavelength of the filter diaphragm is inferred by comparing the signal leakage ratio of adjacent filter channels). Specifically, the wavelength offset Δλ i The estimation method is as follows: For channel i, detect the leakage signal intensity L on the CMOS pixels corresponding to adjacent channels i−1 and i+1. i,i−1 and L i,i+1 When the emitted wavelength is precisely aligned with the center of the filter diaphragm, the leakage signal strength on both sides should be approximately symmetrical. If asymmetry occurs, it indicates a wavelength offset, which can be estimated using the following approximate formula:

[0060] ; FWHM i Determination: The full width at half maximum (FWHM) of each filter film was obtained by actual measurement using a spectrophotometer after fabrication. Its nominal value was 5 nm, and the actual deviation was controlled within ±0.3 nm. In the formula, FWHM... i The calculation was performed using the nominal value of 5nm.

[0061] The channel quality assessment submodule 34 packages and outputs the above parameters to the adaptive mutual feedback control module 4 at a fixed period (10μs in this embodiment).

[0062] It should be noted that the specific circuit implementation of the channel quality assessment module of the present invention can adopt FPGA signal quality monitoring circuits known in the art, such as calculating the bit error rate through conventional circuits such as comparators and counters, and calculating the received optical power through ADC sampling, etc.

[0063] Adaptive mutual feedback control module like Figure 4 and Figure 5 As shown, the algorithm running carrier of the adaptive mutual feedback control module 4 is a dedicated hardware acceleration logic integrated in the receiver FPGA chip (sharing the same FPGA chip as the signal processing circuit 33, and using an independent logic partition). Its functional architecture includes a channel state perception engine 41, a channel matrix estimation and crosstalk decoupling algorithm unit 42, a multi-objective joint optimization decision unit 43, a feedback instruction generation and transmission unit 44, and an anomaly detection and protection subunit 45.

[0064] The channel state awareness engine 41 receives signal quality parameters for each wavelength channel output from the channel quality evaluation submodule 34 of the receiving unit. These signal quality parameters include, but are not limited to, the received optical power P of each channel. r,i(where i is the channel number), Signal-to-noise ratio (SNR) i Bit error rate (BER) i Inter-channel crosstalk CT ij (Crosstalk contribution of channel j to channel i) and the received signal wavelength offset Δλ of each channel. i The channel state awareness engine preprocesses the above raw parameters, including data filtering and denoising (using Kalman filtering or sliding window mean filtering), normalization, and time series caching, to form a standardized channel state vector:

[0065] ; Where N is the total number of wavelength division multiplexing channels, and t is the current time step. The channel state vector S(t) simultaneously caches historical data from the most recent M time steps, forming a state-time series. It is used to support trend prediction and decision-making in subsequent algorithm units.

[0066] Specifically, in this embodiment, the channel state perception engine 41 receives the signal quality parameters of 8 channels output by the channel quality evaluation submodule 34, performs sliding window mean filtering on the original data (the window length is 16 sampling periods), normalizes it to the range of [0,1], and forms a 32-dimensional channel state vector (4 parameters per channel × 8 channels). At the same time, it caches the historical state vectors of the most recent 128 time steps for use by subsequent algorithms.

[0067] like Figure 6 As shown, the channel matrix estimation and crosstalk decoupling algorithm unit 42 constructs and updates the channel transmission matrix H of the system in real time based on the channel state data output by the channel state awareness engine 41, thereby achieving an accurate characterization of the coupling relationship between each wavelength channel.

[0068] The channel transmission matrix H is defined as an N×N dimensional matrix, where the element h in the i-th row and j-th column is... ij The channel transmission matrix represents the response coefficient (including the direct transmission coefficient and crosstalk coefficient) of the optical signal in the j-th channel at the transmitting end in the i-th channel at the receiving end. The channel transmission matrix satisfies the received signal model: Y(t) = H(t) × X(t) + N(t); in This represents the received signal strength vector for each channel at the receiving end. Let N(t) be the transmitted signal strength vector of each channel at the transmitter, and let N(t) be the noise vector.

[0069] The channel matrix estimation employs either the adaptive minimum mean square error (MMSE) algorithm or the recursive least squares (RLS) algorithm for online estimation. Specifically, during the system initialization phase or the periodic calibration phase, the transmitter performs online estimation according to a preset training sequence X. trainA known signal is transmitted in a channel-by-channel or orthogonal code division manner, and the receiver acquires the corresponding received signal Y. train The channel matrix is ​​estimated using the following formula:

[0070] Where σ 2 Let be the noise variance estimate, and I be the identity matrix. During normal system operation, a decision feedback method is adopted, using the decision data demodulated by the signal processing circuit as a reference, and the channel matrix is ​​recursively updated using the RLS algorithm:

[0071] ; Where K(t) is the RLS gain matrix. This is the estimated value of the transmitted signal for the decision feedback. By continuously updating the channel matrix, the system can track the time-varying changes in the characteristics of the transmission channel in real time.

[0072] Based on the estimated channel matrix The crosstalk decoupling algorithm calculates the crosstalk strength index for each channel: in This is the crosstalk strength index for channel i. A higher value indicates more severe crosstalk from adjacent channels. The crosstalk decoupling algorithm modifies the crosstalk strength index against a preset crosstalk threshold CI. th For CI i (t)>CI th The channel is marked as "crosstalk degraded channel" and the relevant information is transmitted to the multi-objective joint optimization decision unit.

[0073] Specifically, in this embodiment, the channel matrix estimation and crosstalk decoupling algorithm unit 42 controls the transmitter to transmit known signals channel by channel according to a preset orthogonal training sequence (an 8×8 Hadamard matrix is ​​used as the training code in this embodiment) during system power-on initialization. The receiver collects the corresponding 64 received signals and estimates the initial 8×8 channel transmission matrix using the MMSE algorithm. During normal operation, a training frame is inserted every 1000 data frames. The training frame data is then processed using the RLS algorithm (with a forgetting factor λ). RLS =0.998) Recursively update the channel matrix. Simultaneously, calculate the crosstalk intensity index (CI) for each channel based on the updated channel matrix. i In this embodiment, a crosstalk threshold CI is set. th =0.05 (meaning the crosstalk power does not exceed 5% of the pass signal power).

[0074] The multi-objective joint optimization decision unit 43 receives the channel matrix output by the channel matrix estimation and crosstalk decoupling algorithm unit 42. Crosstalk intensity index The system uses the channel state vector S(t) output by the channel state awareness engine and its historical time series as inputs to execute a multi-objective joint optimization algorithm to generate the optimal control parameters for the transmitter. The multi-objective joint optimization algorithm includes the following three cooperating optimization sub-algorithms:

[0075] (a) Wavelength dynamic calibration optimization sub-algorithm The wavelength dynamic calibration optimization sub-algorithm is used to compensate for the emission wavelength drift of Micro-LED chips caused by factors such as temperature changes and aging. The sub-algorithm is based on the wavelength offset Δλ of the received signal for each channel provided by the channel state awareness engine. i (t), and historical wavelength offset time series A prediction model based on Long Short-Term Memory (LSTM) networks is used to predict the wavelength drift trend at the next time step. Based on this, the wavelength calibration compensation amount δV is generated. λ,i (The adjustment amount corresponding to the driving voltage) minimizes the calibrated emission wavelength offset. The wavelength calibration objective function is:

[0076] in Let be the wavelength-voltage sensitivity coefficient of the i-th channel. This coefficient is obtained through initial calibration and continuously updated through online identification during operation. Here, "-0" does not represent a mathematical meaning, but rather indicates that the optimization objective is zero offset.

[0077] (b) Power Adaptive Equalization Optimization Sub-algorithm The power adaptive equalization optimization sub-algorithm is used to eliminate the received power imbalance caused by differences in transmission link loss, coupling efficiency, and filter film transmittance between wavelength channels. The sub-algorithm uses the consistency of the signal-to-noise ratio at the receiver of each channel as the optimization objective, and defines the power equalization objective function as follows:

[0078] Where δP t,i This represents the adjustment amount for the transmit power of the i-th channel. The power-to-signal-noise ratio sensitivity coefficient of the i-th channel (obtained through the channel matrix and noise level), SNR target The target signal-to-noise ratio is set for the system, and μ is a regularization coefficient (used to constrain the power adjustment amplitude and avoid over-adjustment causing system oscillation). The objective function is solved analytically to obtain the optimal power adjustment amount.

[0079] (c) Crosstalk suppression and channel spacing optimization sub-algorithm The crosstalk suppression and channel spacing optimization sub-algorithm is used to reduce inter-channel crosstalk by dynamically adjusting the transmission wavelength spacing of relevant channels for crosstalk-degraded channels. The sub-algorithm is based on the crosstalk coefficient h in the channel matrix. ij Construct the crosstalk cost function:

[0080] Where w ij For crosstalk weighting coefficients, w corresponds to the crosstalk degradation channel. ij A larger value is selected to enhance the optimization effect. Under the constraint of an adjustable wavelength spacing range, the sub-algorithm searches for the optimal wavelength allocation scheme using either a particle swarm optimization (PSO) algorithm or a genetic algorithm. This minimizes the crosstalk cost function while satisfying the constraint that the wavelengths of each channel do not overlap:

[0081] Where Δλ min The minimum wavelength spacing to ensure crosstalk-free transmission (determined by the full width at half maximum of the filter layer).

[0082] The multi-objective joint optimization decision unit integrates and coordinates the outputs of the above three sub-algorithms, and uses a weighted multi-objective optimization method to define the system's comprehensive performance index function: ; J λ J power J CT These represent the objective function values ​​for wavelength calibration, power equalization, and crosstalk cost, respectively. ω1, ω2, and ω3 are the weighting coefficients for each objective, which can be dynamically adjusted according to the actual application scenario and priority of the system. The multi-objective joint optimization decision unit weighs the three objectives through Pareto optimality search or hierarchical optimization strategies, ultimately generating a set of transmitter control parameters. .

[0083] Specifically, the multi-objective joint optimization decision unit 43 executes the following joint optimization process: First, the wavelength dynamic calibration optimization sub-algorithm, based on 128 steps of historical wavelength shift data, uses a lightweight LSTM network (2-layer LSTM, with a hidden layer dimension of 16) to predict the wavelength drift trend of each channel in the next step, and calculates the wavelength calibration voltage adjustment amount δV. λ,i To ensure control stability, the adjustment step size is constrained to no more than ±0.1V / step. Secondly, the power adaptive equalization optimization sub-algorithm adjusts the current SNR of each channel relative to the target SNR (set in this embodiment). target The deviation of 20dB is used to obtain the power adjustment amount δP through analytical solution. t,iThe regularization coefficient μ = 0.01. Finally, for CI... i For channels with crosstalk degradation >0.05, the crosstalk suppression and channel spacing optimization sub-algorithm is activated. Within the wavelength fine-tuning range of ±5nm, a simplified particle swarm optimization algorithm (20 particles and 10 iterations) is used to search for the optimal wavelength fine-tuning scheme to minimize the crosstalk cost function.

[0084] The principle for selecting the regularization coefficient μ is: μ=0.01 is the critical damping value determined based on system stability analysis. Under this value, the overshoot of the step response of the closed-loop system with power adjustment is less than 5%, and the adjustment time is less than 10 feedback cycles.

[0085] The specific structure of the LSTM prediction model is as follows: Input layer (16 dimensions) → First LSTM layer (16 hidden units, return sequence = true) → Dropout layer (dropout rate 0.2) → Second LSTM layer (16 hidden units, return sequence = false) → Fully connected layer (output dimension 1). The activation function is tanh, and the recurrent activation function is sigmoid. Training method: 128 historical wavelength offset sequences {Δλ... t-127 ,...,Δλ(t)} are used as input features, with the actual measured value Δλ t+1 Training samples were constructed using labels. The loss function was mean squared error (MSE), the optimizer was Adam (learning rate 0.001), the batch size was 32, and the training epochs were 100. The model was continuously fine-tuned online during system operation, with an incremental update performed every 100 new samples received.

[0086] The outputs of the three sub-algorithms are weighted and fused to generate the final set of control parameters. In this embodiment, the weight coefficients are set to ω1=0.4, ω2=0.35, and ω3=0.25, with wavelength calibration having the highest weight because wavelength drift is the primary factor affecting system performance.

[0087] The feedback command generation and transmission unit 44 encodes the control parameter set C(t) output by the multi-objective joint optimization decision unit 43 into a standardized feedback command frame, and transmits it to the feedback response submodule of the transmitting unit through a feedback communication link. The feedback communication link can adopt one of the following methods: (a) an electrical feedback channel, using low-speed electrical interconnects within the module to transmit feedback commands; (b) an optical feedback channel, using the reverse transmission direction of the transmission optical fiber or a reserved auxiliary optical fiber channel to transmit feedback signals; (c) a wireless feedback channel, using short-range wireless communication (such as Bluetooth Low Energy, UWB, etc.) to transmit feedback commands. The feedback command frame includes the control target channel number, wavelength adjustment amount, power adjustment amount, and priority identifier. The update period of the feedback command is configurable, ranging from 1μs to 100ms, and can be adjusted according to the system's real-time requirements.

[0088] Specifically, the feedback command generation and transmission unit 44 encodes the final control parameter set into a 64-bit feedback command frame (containing a 3-bit channel number + a 12-bit wavelength adjustment + a 12-bit power adjustment + a 2-bit priority + a check bit), and transmits it via the module's internal low-speed I / O. 2 The feedback response submodule 14 at the transmitter transmits data via the C bus (clock frequency 400kHz). The feedback command update cycle is 100μs (i.e., the feedback control bandwidth is 10kHz), which meets the real-time tracking requirements for slow-changing processes such as temperature drift and device aging.

[0089] The anomaly detection and protection subunit monitors in real time whether the BER of each channel exceeds the preset safety threshold BER. safe When any channel BER is detected i (t)>BER safe When the channel quality returns to normal, the emergency control mode is triggered. In this mode, wavelength calibration and power boost are performed on the channel first, and the modulation rate of the channel is reduced to ensure transmission reliability. Once the channel quality returns to normal, the channel will automatically return to the normal control mode.

[0090] Specifically, the anomaly detection and protection subunit 45 continuously monitors the BER of each channel. In this embodiment, a safety threshold BER is set. safe =10 −6 When any channel BER is detected i >10 −6 When this occurs, an emergency control mode is triggered: the feedback priority of this channel is increased to the highest, the wavelength calibration and power adjustment step size is increased to 3 times that of the normal mode to accelerate convergence, and a deceleration command is sent to the signal processing circuit 33 to reduce the modulation rate of this channel from 1Gbps to 500Mbps, until the BER of this channel is below 10 for 100 consecutive evaluation cycles. −9 Afterwards, it automatically restores the normal control mode and the original modulation rate.

[0091] It should be noted that before performing calculations, each algorithm unit in this invention pre-normalizes the physical quantities involved. For example, parameters such as the received optical power, signal-to-noise ratio, bit error rate, wavelength offset, transmit power adjustment, and drive voltage adjustment of each channel are divided by their respective reference values ​​(which are determined during the system initialization and calibration phase) to convert them into dimensionless normalized quantities before being substituted into the objective functions and constraints for calculation. The parameter values ​​and units listed in the formulas in the specification are only used to illustrate the physical meaning and typical value range. In the actual algorithm implementation, the normalized dimensionless values ​​are used in the calculations.

[0092] The system workflow is as follows: like Figure 5As shown, in this embodiment, the workflow of the micro-LED optical interconnect module is as follows: (1) System power-on initialization: The drive control circuit 13 configures the drive voltage of each Micro-LED chip according to the initial parameters, so that each chip emits light signals of preset wavelength; the adaptive mutual feedback control module 4 performs training sequence transmission and initial channel matrix estimation.

[0093] (2) Normal data transmission: The drive control circuit 13 controls each Micro-LED chip in the Micro-LED array 12 to emit light signals of corresponding wavelengths. The light signals carry data information. After the light signals of each wavelength converge, they are coupled into the transmission optical fiber 21 through a microlens. Synchronous transmission within the same optical fiber is achieved through wavelength division multiplexing. After being transmitted to the receiving end, the light signals illuminate the photosensitive surface of the CMOS array 31. The narrow bandpass filter film 32 only allows the light signals of the corresponding wavelengths to pass through and be received by the corresponding CMOS pixels. The CMOS pixels convert the light signals into electrical signals. After being amplified, filtered, converted from analog to digital and demodulated by the signal processing circuit 33, the original data signals corresponding to each band are extracted.

[0094] (3) Real-time assessment of channel quality: The channel quality assessment submodule 34 collects parameters such as received optical power, SNR, BER, and wavelength offset of each channel in real time, and outputs the assessment results every 10μs.

[0095] (4) Closed-loop feedback control: The channel state perception engine 41 preprocesses and buffers the received parameters; the channel matrix estimation and crosstalk decoupling algorithm unit 42 updates the channel matrix and calculates the crosstalk index using the training frame data; the multi-objective joint optimization decision unit 43 performs joint optimization of wavelength calibration, power equalization and crosstalk suppression to generate a set of control parameters; the feedback command generation and transmission unit 44 encodes and transmits the control parameters to the transmitter.

[0096] (5) Dynamic adjustment of transmitter parameters: The feedback response submodule 14 parses the received feedback instructions and updates the driving parameters of each channel. The drive control circuit 13 adjusts the emission wavelength and optical power of each Micro-LED chip accordingly.

[0097] (6) Iterative loop: The above steps 2 to 5 are continuously executed in a loop to form a continuous closed-loop optimization, and the system performance is constantly approaching the optimal state.

[0098] Example 2 The difference between this embodiment and Embodiment 1 lies in the algorithm implementation of the multi-objective joint optimization decision unit 43 in the adaptive mutual feedback control module 4. In this embodiment, the wavelength dynamic calibration optimization sub-algorithm does not use the LSTM prediction model, but instead adopts the classical feedback control method based on an incremental PID controller. Specifically, for the wavelength offset Δλ of each channel... i(t), defining the deviation e i (t)=−Δλ i (t), the output of the incremental PID controller is:

[0099] ; Where K P =0.5, K I =0.05, K D =0.02 is the PID parameter, which only represents the numerical value and does not mean that it is dimensionless. Its dimension can be set according to actual needs. Compared with the LSTM method in Example 1, the PID method in this example has lower computational complexity and less FPGA resource consumption, and is suitable for cost-sensitive applications or applications with a small number of channels. However, its prediction ability is weaker than the LSTM method when facing complex nonlinear wavelength drift modes.

[0100] The remaining parts of this embodiment, including the structure and parameters of the transmitting unit, the transmitting unit, and the receiving unit, are the same as those in Embodiment 1, and will not be described again.

[0101] Example 3 The difference between this embodiment and Embodiment 1 lies in the implementation method of the feedback communication link. In this embodiment, the feedback communication link does not use an electrical feedback channel (I... 2 Instead of using a C-bus, an optical feedback channel is employed, transmitting the feedback signal via the reverse transmission direction of the transmission fiber 21. Specifically, an 850nm VCSEL laser for feedback is added at the receiving end to modulate the feedback command frame onto an 850nm optical carrier, which is then transmitted in the reverse direction to the transmitting end via the transmission fiber 21. At the transmitting end, an 850nm photodetector is added to receive the feedback optical signal and demodulate the feedback command frame, which is then passed to the feedback response submodule 14. Since the wavelength of the feedback signal (850nm) does not overlap with the wavelength of the data signal (450nm-650nm), data transmission and feedback transmission can occur simultaneously in both directions within the same optical fiber without interference. This scheme is suitable for long-distance optical interconnection scenarios where there is no electrical connection between the transmitting and receiving ends.

[0102] The rest of this embodiment is the same as that of Embodiment 1, and will not be repeated here.

[0103] Additional notes: In this invention, "DUT" is an abbreviation for "Dynamic Undulation Tuning," referring to an epitaxial wafer technology that pre-defines multi-region differentiated quantum well structures during the epitaxial growth stage and achieves precise wavelength control during the operation stage by fine-tuning the driving voltage / current. Its wavelength tuning comprises two levels: during the epitaxial stage, large-interval wavelength allocation between channels is achieved through selected-area epitaxy; during the operation stage, dynamic calibration from ±2nm to ±5nm is achieved through modulation of driving parameters.

[0104] The preferred method for achieving multi-wavelength emission from the same wafer in this invention is selective epitaxy (SAE): before MOCVD epitaxial growth, SiO2 dielectric masks with different aperture widths are deposited on the wafer surface using photolithography. Utilizing the mask's effect on the migration and redistribution of precursor molecules, a systematic difference in the In composition and thickness of the InGaN quantum wells in different regions is created, thereby obtaining multiple emission wavelengths in a single epitaxial growth. This process is a mature technology in the III-V semiconductor field. An alternative method is multiple patterned etching and regeneration: through N cycles of "epitaxy-photolithography-etching," quantum well layers with different compositions are prepared in different regions of the wafer, and finally, a P-type layer is grown uniformly to complete the epitaxy.

[0105] PID parameters (K) P =0.5, K I =0.05, K D =0.02) Based on the Ziegler-Nichols critical proportional method, the closed-loop phase margin is greater than 45° and the step response overshoot is less than 10%. Those skilled in the art can readjust it according to the actual device characteristics.

[0106] The weighting coefficients (ω1=0.4, ω2=0.35, ω3=0.25) were determined based on the sensitivity analysis of each index to the system bit error rate: wavelength offset has the greatest impact and therefore has the highest weight, followed by power imbalance, and crosstalk is relatively controllable when the channel spacing is greater than 2 times FWHM, so it has the lowest weight.

[0107] On a 200MHz FPGA, the total latency of each algorithm within a normal feedback cycle is approximately 7μs, and in the worst-case scenario (simultaneous triggering of RLS update and particle swarm optimization), it is approximately 42μs, all within a 100μs feedback cycle. For resource-constrained platforms, timing feasibility can be ensured by increasing the feedback cycle to 1ms or replacing LSTM with PID.

[0108] In summary, to verify the technical effects of the present invention, the following experiments are provided for verification: I. Verification Objective (1) Single optical fiber multi-wavelength wavelength division multiplexing synchronous transmission capability; (2) The dynamic calibration effect of the adaptive mutual feedback control module on wavelength drift; (3) The effect of the adaptive mutual feedback control module on suppressing crosstalk between channels; (4) The effect of power adaptive equalization on improving the consistency of signal-to-noise ratio in each channel; (5) The effect of anomaly detection and protection mechanisms on the robustness of the system; (6) The overall stability of the system under long-term continuous operation and variable temperature environment.

[0109] II. Experimental Platform Setup 2.1 Sample under test Two sets of micro-LED optical interconnect module samples were prepared. The experimental group sample was a complete system, including a transmitting unit, a transmitting unit, a receiving unit, and an adaptive mutual feedback control module. Each sub-module was fully integrated according to the technical specifications. The control group sample had the same structure as the experimental group, but the adaptive mutual feedback control module was in the off state (i.e., open-loop operation mode). The Micro-LED chips at the transmitting end operated with fixed initial parameters and did not receive any feedback control commands.

[0110] 2.2 Parameter Settings The micro-LED array is configured with 8 channels, with designed center wavelengths of 450nm, 465nm, 480nm, 495nm, 510nm, 525nm, 540nm, and 555nm for each channel (a 15nm channel spacing was selected for experiments to test more challenging dense wavelength division multiplexing conditions). The single-chip size is 10μm, and the array density is 250,000 cells / cm². Multimode fiber with a core diameter of 62.5μm and a numerical aperture of 0.2 is used for transmission, with transmission lengths set at 1m, 10m, 100m, and 1km. Amplitude modulation is used for driving, with a modulation rate of 1Gbps / channel.

[0111] 2.3 Testing Instruments A spectrum analyzer (resolution ≤ 0.1 nm) is used to accurately measure the emission wavelength and spectral morphology of each channel. A high-speed digital communication analyzer is used to measure the bit error rate (BER). An optical power meter (sensitivity ≤ -60 dBm) is used to measure the received optical power of each channel. A high-speed oscilloscope (bandwidth ≥ 10 GHz) is used to observe the modulation signal waveform. A temperature-controlled chamber (temperature range -20℃ to +80℃, temperature control accuracy ±0.5℃) is used to simulate a variable temperature environment. A data acquisition system is used to record parameters such as signal-to-noise ratio, bit error rate, optical power, and wavelength offset of each channel in real time.

[0112] III. Experiment 1: Verification of Multi-wavelength Wavelength Division Multiplexing Transmission Capability of a Single Fiber 3.1 Experimental Objective Verify whether a Micro-LED array based on a wavelength-tunable epitaxial wafer (DUT) can achieve 8-channel wavelength division multiplexing synchronous transmission in a single optical fiber, and compare its transmission capacity with that of a traditional single-wavelength single-channel transmission scheme.

[0113] 3.2 Experimental Procedure The first step involved activating all 8 channels of the experimental Micro-LED array. Each channel simultaneously transmitted pseudo-random binary sequence (PRBS-31) test data at a rate of 1 Gbps. The optical signals were coupled and injected into a single multimode fiber (1 m in length). The receiving end CMOS array, in conjunction with a narrowband bandpass filter, synchronously received the optical signals from all 8 channels. The received optical power, signal-to-noise ratio, and bit error rate (BER) of each channel were recorded. The second step, as a control, involved transmitting a single-wavelength signal through the same fiber at the same modulation rate using a single Micro-LED chip, recording the same parameters. The third step involved repeating the above two sets of experiments by successively changing the fiber length to 10 m, 100 m, and 1 km. The fourth step involved gradually increasing the number of simultaneously operating channels (from 1 channel to 8 channels), recording the changes in the total aggregate bandwidth and the BER of each channel at each step.

[0114] 3.3 Evaluation Indicators The evaluation metrics are the system aggregate transmission bandwidth (the sum of the rates of each channel) and the bit error rate of each channel.

[0115] IV. Experiment 2: Verification of Adaptive Wavelength Drift Dynamic Calibration Effect 4.1 Experimental Objective The study aimed to verify whether the adaptive feedback control module could maintain the wavelength stability of each channel by calibrating the wavelength shift in real time through closed-loop feedback when temperature changes cause the emission wavelength to drift.

[0116] 4.2 Experimental Procedure The experimental and control group samples were placed in a temperature-controlled chamber with an initial temperature set at 25℃. Data acquisition began after the system started and ran stably for 5 minutes. The experiment was then conducted according to the following temperature change curve: the temperature was increased from 25℃ to 65℃ at a rate of 2℃ / min, held at 65℃ for 10 minutes, then decreased to -10℃ at a rate of 2℃ / min, held at -10℃ for 10 minutes, and finally increased back to 25℃. Throughout the temperature cycle, the actual emission wavelengths of 8 channels from both the experimental and control groups were simultaneously acquired every 10 seconds using a spectrometer, and the offset of each channel wavelength relative to the designed center wavelength was calculated. The adaptive feedback control module of the experimental group was enabled, while that of the control group was disabled.

[0117] 4.3 Evaluation Indicators The evaluation metrics are the maximum value, mean value, and standard deviation of the wavelength offset for each channel.

[0118] V. Experiment 3: Verification of Inter-channel Crosstalk Suppression Effect 5.1 Experimental Objective Verify the effect of the channel matrix estimation and crosstalk decoupling algorithm in the adaptive mutual feedback control module on suppressing spectral crosstalk between adjacent channels.

[0119] 5.2 Experimental Procedure The first step is static crosstalk measurement. Channels 1 through 8 of the transmitter are turned on individually, while the remaining 7 channels are turned off. At the receiver, the response signal strength of the CMOS pixels corresponding to the target channel and all non-target channels is measured to construct an 8×8 channel response matrix as the initial crosstalk baseline.

[0120] The second step is dynamic crosstalk measurement. All eight channels are turned on simultaneously, and each channel sends a different PRBS test sequence. The crosstalk intensity index CI value (the ratio of the sum of leakage power of non-target channels to the direct power of the target channel) of each channel is recorded under two states: feedback control off (control group) and feedback control on (experimental group).

[0121] The third step is the degradation condition test. The emission wavelength of channel 4 is artificially shifted by +3nm (simulating a severe wavelength drift situation), and the changes in crosstalk and bit error rate of channels 3, 4, and 5 in the experimental and control groups are observed under this perturbation. The response time of the experimental group from detecting crosstalk anomalies to completing wavelength spacing optimization adjustment is recorded.

[0122] 5.3 Evaluation Indicators The evaluation indicators are the crosstalk intensity index (CI) and crosstalk suppression ratio (CI of control group / CI of experimental group) for each channel.

[0123] VI. Experiment 4: Verification of the Power Adaptive Equalization Effect 6.1 Experimental Objective Verify whether the power equalization algorithm of the adaptive mutual feedback control module can automatically adjust the transmit power of each channel when the transmission loss of each channel is inconsistent, so that the signal-to-noise ratio of each channel at the receiver tends to be consistent.

[0124] 6.2 Experimental Procedure The first step involved artificially introducing differential attenuation between channels at the fiber optic input end: by adjusting the coupling alignment offset between the Micro-LED chip and the fiber end face of each channel, the coupling loss of the eight channels was made to exhibit a gradient distribution, specifically with additional losses of 0dB, 0.5dB, 1.0dB, 1.5dB, 2.0dB, 2.5dB, 3.0dB, and 3.5dB for channels 1 to 8, respectively. The second step involved measuring the signal-to-noise ratio (SNR) at the receiver of each of the eight channels in both the control group (fixed transmit power, feedback off) and the experimental group (feedback on, power adaptive equalization algorithm operating). The third step involved recording the convergence process of the power equalization algorithm in the experimental group, i.e., the time and number of iterations from when the system detected power imbalance to when the SNR of each channel reached a steady state.

[0125] 6.3 Evaluation Indicators The evaluation metrics are the maximum, minimum, mean, and standard deviation of the signal-to-noise ratio at the 8-channel receiver.

[0126] VII. Experiment 5: Robustness Verification of Anomaly Detection and Protection Mechanisms 7.1 Experimental Objective The verification aims to determine whether the anomaly detection and protection subunit can promptly trigger the emergency control mode and quickly restore channel quality when a sudden performance degradation occurs in the channel, thereby improving the system's fault tolerance.

[0127] 7.2 Experimental Procedure The first step, under normal system operation, simulates a sudden channel degradation event by suddenly inserting a 3dB optical attenuator into the 5th optical path. Record the time from the moment the attenuator is inserted until the experimental group's anomaly detection subunit detects that the bit error rate of the 5th channel exceeds the safety threshold (set to 1×10⁻⁶). -6 The test included recording the time of the emergency control mode, the time it took for the bit error rate of channel 5 to recover below the threshold, and the time it took for the system to return to the normal control mode. The control group recorded only the change in the bit error rate of channel 5 under the same disturbance, without any control measures. The second step involved a more stringent test: a 6dB attenuator was inserted into the optical path of channel 5, and the above test was repeated to observe whether the system could achieve channel recovery by increasing the transmit power and reducing the modulation rate under emergency control mode. The third step involved a multi-channel simultaneous fault test: 3dB attenuators were simultaneously inserted into channels 3, 5, and 7 to test whether the system could handle multi-channel anomalies in parallel.

[0128] 7.3 Evaluation Indicators The evaluation metrics are anomaly detection delay time, emergency control trigger delay time, channel recovery time, and bit error rate level after recovery.

[0129] 8. Experiment 6: Comprehensive Stability Verification During Long-Term Continuous Operation 8.1 Experimental Objective The test verifies whether the adaptive feedback control module can continuously maintain the system's wavelength stability, power balance, and low bit error rate under long-term continuous operation, thus demonstrating the long-term effectiveness of closed-loop control.

[0130] 8.2 Experimental Procedure Both the experimental and control groups operated continuously at full capacity for 72 hours with all 8 channels. During operation, the ambient temperature cyclically varied as follows: each 8-hour cycle consisted of a sinusoidal temperature change between 20°C and 50°C, simulating diurnal temperature fluctuations in real-world applications. The data acquisition system recorded the transmit wavelength, receive optical power, signal-to-noise ratio, bit error rate, and crosstalk parameters for each channel every minute. Statistical analysis was performed on all collected data after 72 hours.

[0131] 8.3 Evaluation Indicators The evaluation indicators are the time series statistical characteristics (mean, maximum, and standard deviation) of wavelength offset of each channel within 72 hours, the time series statistical characteristics of bit error rate of each channel, the time evolution trend of the standard deviation of signal-to-noise ratio of 8 channels, and the overall level of the system aggregate bit error rate.

[0132] IX. Data Processing and Result Analysis Methods All raw data collected in the experiments were standardized using the 3σ criterion: first, obvious outliers were removed; then, the mean, standard deviation, maximum, minimum, and 95% confidence intervals were calculated for the valid data. The performance differences between the experimental and control groups were analyzed using a paired-samples t-test, with a significance level set at α=0.05. A p-value less than 0.05 was considered statistically significant.

[0133] 10. Design of the Experimental Results Summary Table The final experimental report should include the following summary comparison table, systematically presenting the core quantitative results of the six experimental groups: In summary, the above experiments verified the technical effects of the present invention. This demonstrates that the Micro-LED optical interconnect module described in this invention has broad application prospects and value.

Claims

1. A micro-LED optical interconnect module, comprising a transmitting unit (1), a transmitting unit (2), and a receiving unit (3), which are sequentially optically connected, characterized in that, It also includes an adaptive mutual feedback control module (4): The transmitting unit (1) includes a wavelength-tunable epitaxial wafer (11), a Micro-LED array (12), a driving control circuit (13), and a feedback response submodule (14); the feedback response submodule (14) is integrated into the driving control circuit (13) and is used to receive the control instructions of the adaptive mutual feedback control module (4) and adjust the driving parameters of each Micro-LED chip in real time. The transmission unit (2) includes a single transmission optical fiber (21), the input end of which is coupled to the optical path output end of the transmitting unit (1) for receiving the multi-wavelength mixed optical signal output by the transmitting unit (1). The receiving unit (3) includes a CMOS array (31), a narrowband bandpass filter film (32), a signal processing circuit (33), and a channel quality assessment submodule (34); the channel quality assessment submodule (34) is electrically connected to the signal processing circuit (33) and is used to collect the signal quality parameters of each wavelength channel in real time. The adaptive mutual feedback control module (4) is connected to the feedback response submodule (14) of the transmitting unit (1) and the channel quality assessment submodule (34) of the receiving unit (3) to form a closed-loop feedback control loop of receiving end perception-algorithm decision-transmitting end execution. It is used to generate control instructions through adaptive algorithm according to the signal quality parameters of each wavelength channel of the receiving end, and dynamically adjust the emission wavelength and / or emission power of each Micro-LED chip of the transmitting end.

2. The micro-LED optical interconnect module according to claim 1, characterized in that: The adaptive mutual feedback control module (4) includes a channel state perception engine (41), a channel matrix estimation and crosstalk decoupling algorithm unit (42), a multi-objective joint optimization decision unit (43), and a feedback command generation and transmission unit (44). The channel state perception engine (41) preprocesses the signal quality parameters of each wavelength channel to form a standardized channel state vector. The channel matrix estimation and crosstalk decoupling algorithm unit (42) constructs and updates the channel transmission matrix in real time based on the channel state data and calculates the crosstalk intensity index of each channel. The multi-objective joint optimization decision unit (43) generates a transmitter control parameter set through a multi-objective optimization algorithm based on the channel transmission matrix and the channel state vector. The feedback command generation and transmission unit (44) encodes and transmits the control parameter set to the transmitter.

3. The micro-LED optical interconnect module according to claim 2, characterized in that: The channel transmission matrix H is an N×N dimensional matrix, where N is the total number of wavelength division multiplexing channels, and the matrix elements h ij The response coefficient of the optical signal in the j-th channel at the transmitting end is represented by the response coefficient in the i-th channel at the receiving end. The channel matrix estimation is performed initially based on the training sequence using the minimum mean square error algorithm during system initialization, and recursively updated online using the recursive least squares algorithm during system operation. The crosstalk intensity index CI for each channel is also mentioned. i Defined as the ratio of the sum of crosstalk signal power received by the channel from the non-target channel to the direct signal power of the target channel; the multi-objective joint optimization decision unit (43) includes three cooperating optimization sub-algorithms: wavelength dynamic calibration optimization sub-algorithm, power adaptive equalization optimization sub-algorithm, and crosstalk suppression and channel spacing optimization sub-algorithm. The wavelength dynamic calibration optimization sub-algorithm is used to generate wavelength calibration compensation based on the wavelength offset of each channel and its historical time series. The power adaptive equalization optimization sub-algorithm is used to generate the adjustment amount of the transmit power of each channel based on the deviation between the signal-to-noise ratio of each channel and the target signal-to-noise ratio. The crosstalk suppression and channel spacing optimization sub-algorithm is used to search for the optimal wavelength allocation scheme within the adjustable wavelength spacing range for channels with excessive crosstalk in order to minimize crosstalk.

4. The micro-LED optical interconnect module according to claim 3, characterized in that: The wavelength dynamic calibration optimization sub-algorithm employs a prediction model based on a long short-term memory network. It predicts the wavelength drift trend for the next time step based on historical wavelength offset data from the most recent M time steps, and generates a predictive wavelength calibration compensation amount accordingly. The long short-term memory network continuously updates its parameters online using accumulated historical data during system operation. Alternatively, the wavelength dynamic calibration optimization sub-algorithm employs an incremental PID controller, using the wavelength offset of each channel as feedback deviation, and generates the wavelength calibration compensation amount through proportional, integral, and derivative control.

5. The micro-LED optical interconnect module according to claim 2, characterized in that: The adaptive mutual feedback control module (4) further includes an anomaly detection and protection subunit (45). The anomaly detection and protection subunit (45) monitors the bit error rate of each channel in real time. When the bit error rate of any channel exceeds the preset safety threshold, an emergency control mode is triggered. In the emergency control mode, the feedback priority and control step size of the channel are increased and the modulation rate of the channel is reduced. After the channel quality returns to normal, the normal control mode is automatically restored. The signal quality parameters collected by the channel quality evaluation subunit (34) include the received optical power, signal-to-noise ratio, bit error rate and received signal wavelength offset of each channel. The received signal wavelength offset is obtained by estimating the asymmetry of the leakage signal intensity of the optical signal of the target channel on the corresponding CMOS pixel of the adjacent filter channel.

6. The micro-LED optical interconnect module according to claim 1, characterized in that: The feedback communication link between the adaptive mutual feedback control module (4) and the transmitter can be any one of the following: the internal electrical feedback channel of the module, the reverse optical feedback channel of the transmission optical fiber, or the short-range wireless feedback channel.

7. The micro-LED optical interconnect module according to claim 1, characterized in that: The Micro-LED array (12) is directly fabricated on the wafer of the same DUT wavelength tunable epitaxial wafer (11). The driving control circuit (13) is electrically connected to the Micro-LED array (12) and is used to control the different Micro-LED chips in the Micro-LED array (12) to emit light signals of different wavelengths through circuit modulation, and the wavelengths do not overlap. The DUT wavelength tunable epitaxial wafer (11) is fabricated using III-V compound semiconductor materials. Its epitaxial structure includes a substrate, a buffer layer, an N-type confinement layer, a quantum well active region, a P-type confinement layer and an ohmic contact layer. The tunable wavelength range is 400nm-1650nm. In the Micro-LED array (12), the size of each Micro-LED chip is 1-100μm, the array density is 400-25 million / cm², and the emission wavelength interval of each chip is 1-100nm.

8. The micro-LED optical interconnect module according to claim 1, characterized in that: The driving control circuit (13) is fabricated using CMOS driving technology and integrated on the same wafer as the Micro-LED array (12). It controls the emission wavelength and optical power of each Micro-LED chip through circuit modulation, with a modulation rate of 100Mbps-100Gbps. The signal processing circuit (33) includes a preamplifier, a filter circuit, an analog-to-digital converter circuit, and a signal demodulation circuit, which are used to convert analog electrical signals into digital signals and demodulate and extract the original data.

9. The micro-LED optical interconnect module according to claim 1, characterized in that: The transmission optical fiber (21) is a single-mode fiber or a multimode fiber with a numerical aperture of 0.1-0.3, a core diameter of 2-100μm, and a transmission loss of ≤0.5dB / km.

10. The micro-LED optical interconnect module according to claim 1, characterized in that: The narrowband bandpass filter film (32) is covered on the photosensitive surface of the CMOS array (31), and the narrowband bandpass filter film (32) corresponds one-to-one with the pixels of the CMOS array (31); in the narrowband bandpass filter film (32), the center wavelength of the filter film corresponding to different pixels corresponds one-to-one with the wavelength emitted by the Micro-LED array (12) in the emitting unit (1); the signal processing circuit (33) is electrically connected to the CMOS array (31) and is used to perform photoelectric conversion and signal extraction on the optical signals received by each CMOS pixel; the narrowband bandpass filter film (32) is prepared by vacuum evaporation or ion beam sputtering process, the film thickness is 100-1000nm, the half width at half maximum of each filter film is ≤10nm, and the transmittance is ≥70%; the array size of the CMOS array (31) is consistent with the size of the Micro-LED array (12) in the emitting unit (1), the pixel size is 1-50μm, and each CMOS pixel corresponds one-to-one with the corresponding Micro-LED chip and narrowband filter film.

Images