Bonding hybrid speed reducer based on MEMS nanometer speed reducer and harmonic speed reducer

Abstract

The invention relates to the technical field of micro transmission, in particular to a bonding hybrid speed reducer based on an MEMS (Micro Electro Mechanical System) nano speed reducer and a harmonic speed reducer. Preparing a silicon wafer mixed bonding sample and an SOI wafer mixed bonding sample; cleaning and oxidizing the silicon wafer mixed bonding sample and the SOI wafer mixed bonding sample; carrying out photoresist coating, exposure and development on the silicon wafer mixed bonding sample and the SOI wafer mixed bonding sample which are subjected to cleaning and oxidation treatment to be bonded; performing hard mask preparation on the exposed and developed silicon wafer mixed bonding sample and the SOI wafer mixed bonding sample to obtain silicon structure etching; and performing silicon structure etching on the silicon wafer mixed bonding sample and the SOI wafer mixed bonding sample, and performing sacrificial layer release, surface functionalization, multi-layer structure bonding and driving integration to obtain a mixed bonding sample. According to the invention, hybrid bonding integration of a silicon wafer hybrid bonding sample and an SOI wafer is realized, and a harmonic reducer bonding hybrid method is established to design an AI algorithm aggregation and reinforcement learning system and a learning training module for a computing power system, an AI coprocessor and an intelligent waveform generation and storage system. The harmonic reducer bonding mixing method is formed through shape optimization, adaptive control system, anomaly detection, off-line training, on-line training, end-to-end integration and deployment.

Description

Technical Field

[0001] The present invention relates to the field of micro transmission technology, and specifically to a hybrid reduction device based on the bonding integration of a MEMS nano reducer and a traditional harmonic reducer, which is applicable to the high-precision and high-torque density transmission requirements in the fields of micro-robots, precision medical devices, aerospace, etc. Background Art

[0002] Traditional harmonic reducers rely on the elastic deformation of the flexspline to achieve a high reduction ratio, but they are relatively large in size and difficult to miniaturize; while MEMS reducers can be miniaturized through nanomachining, but are limited by material strength and structural design, suffering from problems such as low load-bearing capacity and short lifespan. Through heterogeneous material bonding and structural innovation, the present invention combines the advantages of both to propose a high-performance hybrid reduction solution.

[0003] With the growing demand for high-precision transmission systems in the fields of micro-robots, precision medical devices, aerospace equipment, and consumer electronics, the contradiction between traditional reduction technologies and miniaturization technologies has become increasingly prominent. The core challenge of the transmission system lies in how to achieve miniaturization (size < 10 mm 3 ) while also taking into account a high reduction ratio (> 100:1), high torque density (> 5 N·m / kg), and long lifespan (> 10 7 cycles). However, existing technologies are difficult to meet these stringent requirements due to material, process, and design limitations. The following problems exist:

[0004] Limitations of Traditional Reduction Technologies / MEMS Nano Reducers:

[0005] Silicon-based or silicon nitride (Si3N4) nano-gears (module 0.05 - 0.1) can achieve miniaturization (size < 1 mm 3 ) and high precision (tooth profile error < 10 nm), but are limited by material brittleness (the fracture toughness of silicon is only 0.8 MPa·m 1 / 2 ) and low load-bearing capacity (torque density < 0.1 N·m / kg), and cannot meet high-load scenarios such as the joints of surgical robots (requirement > 0.5 N·m) or industrial collaborative robots (requirement > 5 N·m). Severe friction and wear (friction coefficient between silicon - silicon > 0.3), and the lack of effective lubrication means result in a lifespan < 10 6 cycles.

[0006] Traditional harmonic reducers: Rely on the elastic deformation of the flexspline to achieve a high reduction ratio (single-stage 50 - 200:1), but are large in size (diameter > 10 mm), heavy ( > 10 g), and cannot be integrated into micro-devices (such as endoscopes, micro-satellite drive mechanisms). The processing cost of metal flexsplines is high (more than $500 per piece), and the coefficient of thermal expansion at high temperatures (CTE = 12×10 -6The backlash drift is caused by temperature rise (the error is >30% when the temperature rises by 50 °C).

[0007] Difficulties in heterogeneous material integration: Due to the mismatch of thermal expansion coefficients at the bonding interface between silicon-based MEMS and metal / ceramic materials (silicon CTE = 2.6×10 -6 / °C, steel CTE = 12×10 -6 / °C), stress concentration is likely to occur (the stress is >100 MPa when the temperature difference is 50 °C), resulting in interface peeling or crack propagation. Existing bonding technologies (such as Au-Si eutectic bonding, SiO2 direct bonding) have low strength (<5 MPa) and cannot withstand dynamic alternating loads (such as the high-frequency start-stop scenario of robots).

[0008] Lack of intelligence and adaptive ability: Traditional control relies on preset parameters and cannot compensate for environmental disturbances in real time (such as sudden changes in temperature and load), resulting in a decrease in transmission accuracy (backlash >1 μm). The lack of health status monitoring and life prediction capabilities is likely to cause sudden failures (such as the out-of-control of aerospace mechanisms due to fatigue fracture).

[0009] Technical empowerment requirements for AI and learning frameworks / Bottlenecks of traditional control algorithms

[0010] PID control: Relying on the linear model assumption, it is difficult to handle complex dynamic problems such as nonlinear friction and backlash hysteresis, and the dynamic error is >10%.

[0011] Open-loop control: Unable to sense load changes (such as when a surgical robot encounters a sudden change in tissue resistance), resulting in an output torque fluctuation of >20%.

[0012] Necessity of AI co-processors

[0013] Real-time adaptive control: Based on the dynamic parameter adjustment of reinforcement learning (RL), optimize the drive frequency, phase compensation, and lubrication strategy according to real-time sensor data (torque, temperature, vibration), and reduce the transmission error to ±10 nm. For example: In the deployment mechanism of a microsatellite, AI can autonomously adjust the reduction ratio to compensate for load fluctuations in a zero-gravity environment.

[0014] Multi-modal data fusion: Integrate multi-source data such as piezoresistive sensors (strain detection), infrared temperature measurement (thermal monitoring), and optical encoders (displacement feedback) to build a digital twin model and achieve full-life cycle state tracking.

[0015] Prognostics and health management (PHM): Use long short-term memory networks (LSTM) to analyze historical load spectra, predict the time of fatigue crack initiation (error <5%), and trigger preventive maintenance to avoid catastrophic failures (such as the stop of an artificial heart pump).

[0016] Core functions of the learning framework module / Federated learning and edge intelligence: On the premise of protecting data privacy, share local models of multiple devices (such as lubrication strategy optimization) through the federated learning framework to improve the generalization ability of the global model. Deploy lightweight AI models (such as TinyML) at the edge to achieve low-latency (<1ms) real-time control, and perform large-scale training and model iteration in the cloud.

[0017] Ethical and safety design: Data anonymization processing: The physiological data of patients in medical robots is only used for local model optimization after encryption to eliminate the risk of privacy leakage.

[0018] Fail-safe mechanism: The AI co-processor is built with a self-destruction instruction (such as cutting off the power when torque overload occurs) to ensure that no secondary damage will be caused when the system fails.

[0019] Innovation drive of MEMS technology combined with AI / Collaborative design of material-structure-algorithm heterogeneous materials

[0020] Gradient bonding: Through the SiC / Si3N4 transition layer and plasma-activated bonding technology, the interface strength is increased to >15MPa, and the thermal stress is reduced by 60%, solving the problem of silicon-metal bonding failure.

[0021] Intelligent lubrication and self-repair: Deposit molybdenum disulfide (MoS2) nanolubricating film (thickness 2nm, friction coefficient <0.05) by ALD, and combine AI to dynamically control the release of lubricant (such as activating perfluoropolyether encapsulated in microcapsules when the temperature >80°C).

[0022] Embedded sensing network: Integrate polysilicon nanowire piezoresistive sensors (sensitivity >100μV / με) inside the MEMS structure to monitor the tooth root stress distribution in real time, and optimize the driving strategy after the data is analyzed by the AI co-processor.

[0023] Technical responsibility and social value / Medical field: The transmission system of surgical robots needs to meet biocompatibility (such as silicon nitride coating), reliability in a sterile environment (lubricant is non-toxic), and AI ensures action accuracy (±10μm) to avoid the risk of accidental injury to nerves and blood vessels.

[0024] Aerospace field: Verify the reliability of the reducer in the extreme environment of -150°C to 200°C through AI-driven thermal-mechanical coupling simulation to avoid satellite failure due to material embrittlement.

[0025] Industrial sustainable development: AI optimizes energy consumption (transmission efficiency >90%), reduces carbon emissions of industrial robots; predictive maintenance reduces equipment downtime rate, contributing to green manufacturing. Summary of the invention

[0026] 1. A MEMS-based nano-reducer, characterized in that it includes;

[0027] Step S1: Prepare a silicon wafer hybrid bonding sample and an SOI wafer hybrid bonding sample;

[0028] Step S2: Clean and oxidize the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample respectively;

[0029] Step S3: Apply photoresist, expose and develop the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample to be bonded after cleaning and oxidation;

[0030] Step S4: Prepare a hard mask on the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample after exposure and development to obtain silicon structure etching;

[0031] Step S5: Release the sacrificial layer, functionalize the surface, bond the multi-layer structure, and integrate the drive on the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample to obtain a hybrid bonding sample;

[0032] 2. A bonding hybrid method based on a MEMS nano-harmonic reducer according to claim 1, wherein the hybrid bonding sample obtained in steps S1 to S5 includes MEMS nano-technology.

[0033] 3. The MEMS technology according to claim 2, characterized in that the microelectronic system implantation system of the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample according to claim 1;

[0034] 4. The microelectronic implantation system according to claim 3, characterized in that the MEMS technology in claims 1 to 3 is based on an AI (Artificial Intelligence) co-processor.

[0035] 5. The AI (Artificial Intelligence) co-processor according to claim 4, which includes self-developed AI chips and multi-composite modules for building and integration.

[0036] 6. A harmonic reducer bonding hybrid reduction device, characterized in that it includes

[0037] Step 1: Identify, calculate and verify the computing power system, convert the digital signal processor (DSP) for input and output verification, and identify a high-precision control digital signal processor for the waveform. The high-precision control digital signal processor performs identification and calculation through data information to obtain a high-precision control system for the waveform.

[0038] Step 2: Perform identification and calculation on the data information to obtain a high-precision control system for the waveform, and store the programmable waveform;

[0039] Step 3: Program and identify the data information of the programmable waveform storage to calculate the intelligent waveform generation and storage system;

[0040] Step 4: Dynamically adjust the waveform of the intelligent waveform generation and storage system design through reinforcement learning to adjust and calculate and identify digital data information;

[0041] Step 5: Aggregate the adjusted calculated and identified digital data information through AI algorithms, and calculate and identify and verify the digital parameters of the data information through reinforcement learning and algorithm clustering for the learning and training module;

[0042] Step 6: Identify and verify the DSP system of the learning and training module to calculate the digital parameters of the data information to obtain an intelligent waveform generator based on the learning and training module;

[0043] Step 7: Use the intelligent waveform generator based on the learning and training module as the learning and training module, characterized in that the intelligent waveform generator of the learning and training module in Steps 1 to 6 is used for verification and identification of digital waveform optimization of calculation data information, adaptive control system, anomaly detection, offline training and online training, end-to-end integration, and deployment to form a harmonic reducer bonding hybrid method. Description of the Drawings

[0044] Figure 1 A harmonic reducer bonding hybrid method based on MEMS nano-reducer and harmonic reducer in an embodiment of the present invention.

[0045] Figure 2 Silicon wafer hybrid bonding samples and SOI wafer hybrid bonding samples in an embodiment of the present invention are implanted with MEMS technology.

[0046] Figure 3 In an embodiment of the present invention, MEMS technology and an AI (Artificial Intelligence) co-processor are integrated and respectively include self-developed AI chips and multi-composite modules for building and integration.

[0047] Figure 4 A harmonic reducer bonding hybrid method process in an embodiment of the present invention. Detailed Embodiment

[0048] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically describes the specific embodiments of the present invention with reference to the drawings;

[0049] It should be noted that in the embodiments of the present invention, without conflict, including self-developed AI chips and multi-composite modules, is another patent implementation for building an embodiment, and the meanings of "including, having, containing, generating, obtaining, identifying, verifying, calculating", etc. are non-limiting.

[0050] As Figure 1 shown, an embodiment of the present invention provides a bonding hybrid method based on a MEMS nano-reducer and a harmonic reducer, including:

[0051] Step S1, preparing a silicon wafer hybrid bonding sample and a SOI wafer hybrid bonding sample;

[0052] Step S2, respectively cleaning and oxidizing the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample;

[0053] Step S3, coating photoresist, exposing and developing the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample to be bonded after cleaning and oxidizing;

[0054] Step S4, preparing a hard mask on the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample after exposure and development to obtain silicon structure etching;

[0055] Step S5, performing sacrificial layer release, surface functionalization, multi-layer structure bonding, and drive integration on the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample obtained by silicon structure etching to obtain a hybrid bonding sample;

[0056] Compared with the prior art, in the embodiment of the present invention, the combination of the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding in the present invention combines MEMS technology and advanced coating technology, which can promote the development of robot reducers towards smaller, stronger, and more intelligent directions.

[0057] In the embodiment of the present invention: Step S1, preparing a silicon wafer hybrid bonding sample and a SOI wafer hybrid bonding sample; Step S2, respectively cleaning and oxidizing the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample;

[0058] In the embodiment of the present invention: Step S3, coating photoresist, exposing and developing the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample to be bonded after cleaning and oxidizing;

[0059] In the embodiment of the present invention: Step S4, preparing a hard mask on the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample after exposure and development to obtain silicon structure etching; Step S5, performing sacrificial layer release, surface functionalization, multi-layer structure bonding, and drive integration on the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample obtained by silicon structure etching to obtain a hybrid bonding sample;

[0060] In the embodiment of the present invention: A bonding hybrid method based on a MEMS nano and harmonic reducer, characterized in that the hybrid bonding sample obtained in Steps S1 to S5 includes MEMS nano technology.

[0061] In an embodiment of the present invention: The MEMS technology has a microelectronic system implantation system with a silicon wafer hybrid bonding sample and a SOI wafer hybrid bonding sample;

[0062] In an embodiment of the present invention: The microelectronic implantation system contains an MEMS technology-based AI (Artificial Intelligence) coprocessor.

[0063] In an embodiment of the present invention: The AI (Artificial Intelligence) coprocessor includes a self-developed AI chip and a multi-composite module for building and integration.

[0064] In an embodiment of the present invention: A harmonic reducer bonding hybrid method; Step 1, identify and calculate the computing power system to verify the digital signal processor (DSP), convert it for input and output verification, and identify a high-precision control digital signal processor for the waveform. The high-precision control digital signal processor performs identification and calculation through data information to obtain a high-precision control system for the waveform.

[0065] In an embodiment of the present invention: Step 3, perform data information programming on the programmable waveform storage to identify and calculate to obtain an intelligent waveform generation and storage system;

[0066] Step 4, design the intelligent waveform generation and storage system to perform reinforcement learning to dynamically adjust the waveform and adjust and calculate to identify digital data information;

[0067] In an embodiment of the present invention: Step 5, perform aggregation on the adjusted calculated identification digital data information through an AI algorithm, perform calculation and identification verification on the data information digital parameters through reinforcement learning and algorithm clustering, and perform programming for the learning and training module;

[0068] Step 6, perform identification and verification on the learning and training module for the DSP system to calculate the data information parameters digitally to obtain an intelligent waveform generator based on the learning and training module;

[0069] In an embodiment of the present invention: Step 7, perform the learning and training module on the intelligent waveform generator based on the learning and training module. It is characterized in that the intelligent waveform generator of the learning and training module in Steps 1 to 6 performs verification and identification calculation on the data information digital waveform optimization, adaptive control system, anomaly detection, offline training, online training, end-to-end integration, and deployment to form a harmonic reducer bonding hybrid method.

Claims

1. A MEMS-based nano reducer, characterized in that: include; Step S1, preparing a silicon wafer hybrid bonding sample and an SOI wafer hybrid bonding sample; Step S2, cleaning and oxidizing the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample respectively; Step S3, coating the cleaned and oxidized silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample to be bonded with photoresist, exposing and developing; Step S4, preparing a hard mask on the exposed and developed silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample to obtain silicon structure etching; Step S5, subjecting the silicon wafer hybrid bonding sample and the SOI wafer hybrid bonding sample to silicon structure etching for sacrificial layer release, surface functionalization, multi-layer structure bonding, and driver integration to obtain a hybrid bonding sample.

2. A hybrid bonding method based on MEMS nano and harmonic reducer according to claim 1, characterized in that The hybrid bonding sample obtained in steps S1 to S5 includes MEMS nanotechnology.

3. The MEMS technology according to claim 2, characterized in that The microelectronic system implantation system described in the silicon wafer hybrid bonding sample and SOI wafer hybrid bonding sample claimed in claim 1.

4. The microelectronic implant system according to claim 3, characterized in that: The MEMS technology in claims 1 to 3 is based on an AI (artificial intelligence) coprocessor.

5. The AI ​​(artificial intelligence) coprocessor according to claim 4 includes independently developed AI chips and multiple complex modules for construction and integration.

6. A harmonic reducer bonded hybrid reduction device, characterized in that: include Step 1: The computing system is identified, calculated and verified by the digital signal processor (DSP) for input and output verification and identification to obtain a high-precision control digital signal processor of the waveform. The high-precision control digital signal processor uses data information to identify and calculate the waveform to obtain a high-precision control system. Step 2, identifying and calculating the data information to obtain a high-precision control system of the waveform, and performing programmable waveform storage; Step 3, performing data information programming recognition calculation on the programmable waveform storage to obtain an intelligent waveform generation and storage system; Step 4: Perform reinforcement learning on the intelligent waveform generation and storage system design to dynamically adjust the waveform to adjust, calculate and identify digital data information; Step 5: aggregate the adjusted calculation and recognition digital data information through AI algorithm, and perform calculation and recognition verification data information digital parameter programming through reinforcement learning and algorithm clustering to perform learning and training module; Step 6: Identify and verify the learning and training module, and the DSP system calculates the data information parameter digitally to obtain an intelligent waveform generator based on the learning and training module; Step 7, the learning and training module-based intelligent waveform generator is subjected to a learning and training module, characterized in that the learning and training module intelligent waveform generator in steps 1 to 6 performs verification and recognition calculation data information digital waveform optimization, adaptive control system, anomaly detection, offline training, online training, end-to-end integration and deployment to form a harmonic reducer bonding hybrid method.

Images