Multifunctional automatic control method and device for semiconductor truncation

Abstract

The invention relates to the field of truncation control, discloses a multifunctional automatic control method and equipment for semiconductor truncation, and is used for solving the long-standing problem of quality control difficulty in the field of hard and brittle material processing. Comprising the following steps: collecting multi-physical field signals such as cutting force, acoustic emission, interface temperature and main shaft power in real time, and inputting dimensionless data into a tool wear dynamic model based on a fractional order differential equation and a cross-scale material response model based on damage mechanics in parallel; the tool health state trend and the material subsurface damage evolution trend are predicted in advance, the maximization of the material removal rate is taken as an optimization target, the two prediction trends do not exceed a safety threshold value as a constraint, online rolling optimization is carried out, and a process control instruction is generated. According to the method, the crossing from passive response to active predictive control is realized, and the problem that tool wear and material damage are difficult to cooperatively control in the machining process is effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cutting control, in particular to a multifunctional automatic control method and device for semiconductor cutting. BACKGROUND

[0002] In the field of semiconductor manufacturing, the cutting process of semiconductor materials is one of the key links, and its processing quality directly affects the performance and reliability of semiconductor devices. With the rapid development of semiconductor technology, the requirements for processing precision, efficiency and damage control of tools and materials during processing are increasingly stringent. In the manufacturing of integrated circuit chips, the cutting of silicon wafers and other semiconductor materials not only needs to ensure the flatness and smoothness of the cutting surface, but also needs to minimize the subsurface damage to the material during cutting to avoid affecting the subsequent circuit production and device performance. At the same time, in order to improve production efficiency and reduce cost, it is also crucial to prolong the service life of the cutting tool.

[0003] In the process of semiconductor cutting, complex physical phenomena and multiple factors interact. Cutting force, acoustic emission signal, interface temperature and spindle power and other physical signals can reflect the real-time state of the processing process, but these signals are affected by factors such as equipment size and processing conditions, making it difficult to directly assess the tool health status and material damage. The traditional processing control method is often based on experience or simple threshold control, which is difficult to consider the complex relationship between multiple factors, and cannot achieve precise control and optimization of the processing process. In terms of tool wear and material damage prediction, existing technologies are mostly based on traditional integer order models, which are difficult to accurately represent the memory and genetic characteristics of the diamond abrasive wear process, as well as the cross-scale relationship between macroscopic sensing signals and microscopic crack propagation. Therefore, the prediction accuracy of tool health status and material subsurface damage is low, and it cannot provide reliable basis for the optimization of the processing process. In terms of processing process control, existing technologies usually optimize a single processing index as the target, lack comprehensive consideration of tool health status and material damage, and are prone to ignore tool wear and material damage while pursuing processing efficiency, or sacrifice processing efficiency when excessively protecting tools and materials.

[0004] Therefore, we propose a multifunctional automatic control method and device for semiconductor cutting to solve the above problems. SUMMARY

[0005] The present application provides a multifunctional automatic control method and device for semiconductor cutting, which solves the long-standing quality control problem in the field of hard and brittle material processing.

[0006] The first aspect of the present application provides a multifunctional automatic control method for semiconductor cutting, comprising: collecting cutting force, acoustic emission signal, interface temperature and spindle power physical signals in real time during processing, and converting them into dimensionless state data using pre-calibrated equipment characteristic data; inputting the dimensionless state data into a tool wear dynamics model, which represents the memory and genetic characteristics of the diamond abrasive wear process, to predict the tool health state trend; inputting the dimensionless state data into a cross-scale material response model to obtain the material subsurface damage evolution trend; maximizing the material removal rate as the target, while the tool health state trend and the material subsurface damage evolution trend do not exceed the safety threshold as the constraint, performing online rolling optimization to generate process control instructions.

[0007] Optionally, in the first implementation manner of the first aspect of the present application, the method comprises: collecting cutting force signal, acoustic emission signal, interface temperature signal and spindle power signal in real time to generate original multi-physical field sensing data set; calling pre-stored equipment characteristic data set, which includes maximum allowed cutting force, spindle rated power, acoustic emission signal saturation value and tool material temperature resistance; multiplying the cutting force signal in the original multi-physical field sensing data set by the maximum allowed cutting force to generate dimensionless equivalent cutting force; multiplying the acoustic emission signal amplitude in the original multi-physical field sensing data set by the acoustic emission signal saturation value to generate dimensionless acoustic emission energy; multiplying the interface temperature signal in the original multi-physical field sensing data set by the tool material temperature resistance to generate dimensionless interface temperature; multiplying the spindle power signal in the original multi-physical field sensing data set by the spindle rated power to generate dimensionless power load; combining the dimensionless equivalent cutting force, dimensionless acoustic emission energy, dimensionless interface temperature and dimensionless power load to form a dimensionless state data set.

[0008] Optionally, in the second implementation manner of the first aspect of the present application, the method comprises: initializing the current state vector of the tool wear dynamics model with the dimensionless state data set at the current time as input, the state vector at least including the current dimensionless wear amount and its historical change rate; calling pre-calibrated model data set, which includes wear fractional order, wear coefficient and exponential data; using fractional order numerical algorithm to numerically solve the tool wear dynamics model for future control cycles to generate tool state vector sequence; analyzing the evolution trajectory of the tool state vector sequence to extract its change slope and curvature characteristics to generate tool wear prediction trend data.

[0009] Optionally, in the third implementation form of the first aspect of the present application, the method comprises: taking the dimensionless state data set at the current time as input, initializing a current state vector of the cross-scale material response model, the state vector at least containing a current dimensionless damage variable and its historical evolution rate; calling a pre-calibrated material property data set, the data set containing damage fractional order, crack propagation coefficient and sensitivity data; using a damage mechanics-based numerical algorithm, numerically solving the cross-scale material response model for a future number of control periods to generate a material state vector sequence; analyzing the evolution trajectory of the material state vector sequence, identifying its accelerated change characteristics, and generating subsurface damage prediction trend data.

[0010] Optionally, in the fourth implementation form of the first aspect of the present application, the method comprises: receiving the tool wear prediction trend data and the subsurface damage prediction trend data as constraint condition input; based on the current machining state, constructing a multi-objective optimization problem with the dimensionless material removal rate as the objective function, the tool wear prediction trend data not exceeding a preset wear safety threshold and the subsurface damage prediction trend data not exceeding a preset damage safety threshold as double constraint conditions; using a sequential quadratic programming algorithm to solve the multi-objective optimization problem in real time, obtaining a control data sequence for a future number of control periods that optimizes the objective function under the premise of meeting all constraint conditions, generating a preliminary optimized control sequence; extracting the data of the first control period in the preliminary optimized control sequence as the optimal control decision at the current time, and generating a process control instruction.

[0011] Optionally, in the fifth implementation form of the first aspect of the present application, the dimensionless material removal rate is set as : , wherein, is a dimensionless feed speed instruction, is a dimensionless spindle power instruction, and k is a device coefficient.

[0012] Optionally, in the sixth implementation form of the first aspect of the present application, the process control instruction is mapped into a physical control instruction of a specific device, and the servo feed unit and the spindle drive unit are adjusted in real time, so that the machining process is actively controlled in advance, and the effects of prolonging tool life and inhibiting machining damage are achieved: the process control instruction is received as an input source, and contains a dimensionless feed speed instruction and a dimensionless spindle power instruction; a pre-stored device physical limit data set is called, including a maximum allowed feed speed of the servo feed unit, a maximum allowed rotating speed of the spindle drive unit, and a maximum output power of the spindle motor; the dimensionless feed speed instruction is multiplied by the maximum allowed feed speed to generate a physical feed speed control instruction; the dimensionless spindle power instruction is multiplied by the maximum output power to obtain an expected output power value, which is converted into a physical spindle rotating speed control instruction according to a spindle characteristic curve; and the physical feed speed control instruction and the physical spindle rotating speed control instruction are combined to generate an execution physical control instruction set.

[0013] Optionally, in the seventh implementation form of the first aspect of the present application, after a complete machining cycle is completed, surface quality detection data of a machined workpiece and actual tool wear measurement data are collected to generate a post-processing data set; the post-processing data set is compared and analyzed with corresponding prediction trend data to obtain a deviation between a predicted value and an actual value, and a model accuracy evaluation report is generated; based on the model accuracy evaluation report, a data adjustment algorithm based on sensitivity analysis is used to fine-tune data in the tool wear dynamics model and the cross-scale material response model, and an updated model data set is output.

[0014] The second aspect of the present application provides a multifunctional automatic control device for semiconductor cutting, which comprises a memory and at least one processor, the memory stores instructions; the at least one processor calls the instructions in the memory to make the multifunctional automatic control device for semiconductor cutting execute the multifunctional automatic control method for semiconductor cutting.

[0015] The mechanism of the present application is as follows: by converting the multi-physical field sensing signal into dimensionless state data, the limitation of device scale is broken, the memory genetic characteristics of diamond abrasive wear are described by using the fractional order tool wear dynamics model, the tool health state is predicted in advance, the correlation between macroscopic signals and microscopic crack propagation is established through the cross-scale material response model, the material subsurface damage evolution is predicted, and the two prediction trends are used as dynamic constraints to solve the optimal process parameters for maximizing the material removal rate in real time through online rolling optimization. Beneficial effects: The online and advanced prediction of the micro-crack initiation and propagation trend inside the material is realized, so that the process parameters can be actively adjusted to inhibit the macro-damage before it occurs, and the control of the subsurface damage of granite, precision ceramics and other materials reaches an unprecedented level, and the processing yield is significantly improved; The history cumulative effect and nonlinear evolution law of diamond abrasive wear are accurately described by using fractional differential equation, the health decline trend of the tool can be predicted in advance, so that intervention or tool replacement is carried out at the most appropriate time, the tool life utilization rate is maximized, and the production cost is greatly saved while ensuring the processing quality; The prediction trends of the two most critical constraints of tool health and material damage are directly integrated into the online rolling optimization with the goal of maximizing the material removal rate, the optimal process window can be dynamically found and always operated in, and the limit processing efficiency is realized under the premise of ensuring high quality and high reliability. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 An embodiment of the semiconductor cutting multifunctional automatic control method in the embodiment of the application is shown in the figure; Figure 2 Another embodiment of the semiconductor cutting multifunctional automatic control method in the embodiment of the application is shown in the figure; Figure 3 An embodiment of the semiconductor cutting multifunctional automatic control device in the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0017] The embodiment of the application provides a semiconductor cutting multifunctional automatic control method and device, and is used for solving the long-existing quality control problem in the field of hard and brittle material processing. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For the sake of understanding, the specific process of the embodiment of the application is described below, please refer to Figure 1 An embodiment of the semiconductor cutting multifunctional automatic control method in the embodiment of the application comprises: 101. Multi-physical field sensing data acquisition and dimensionless: real-time acquisition of cutting force, acoustic emission signal, interface temperature and spindle power physical signals in the machining process, and conversion into a set of dimensionless state data independent of equipment scale by using pre-calibrated equipment characteristic data; It can be understood that the execution subject of the present application can be a multifunctional automatic control device for semiconductor cutting, and can also be a terminal or a server, which is not limited here. The server is taken as an example for description of the embodiment of the present application.

[0019] It should be noted that in the semiconductor cutting process, special sensors and data processing units are required to realize multi-physical field sensing data acquisition and dimensionless. Sensor configuration and data acquisition: cutting force: piezoelectric force sensor (quartz sensor) is installed at the spindle bearing, sampling frequency is set to 50 kHz, three-direction cutting force (main cutting force Fx, feed force Fy, and back force Fz) is monitored in real time, and the measured cutting force signal is Fx=120N±15N (dynamic fluctuation value).

[0020] Acoustic emission signal: an acoustic emission sensor (piezoelectric ceramic type) with a resonant frequency of 340 kHz is installed at the tool tower position to collect high-frequency stress wave signals generated by crack propagation and abrasive wear, and the typical signal amplitude range is 65-90 dB.

[0021] Interface temperature: an infrared thermometer (wavelength range 8-14 μm) is used to non-contact measure the tool-workpiece contact area, the temperature sampling rate is 100 Hz, and the measured temperature range is room temperature to 220℃ (under finishing conditions).

[0022] Spindle power: the Hall current sensor built-in the spindle motor is used for indirect calculation, current and voltage signals are collected and converted into power value, and the typical power fluctuation range is 0.5-3.5 kW.

[0023] Dimensionless processing: The collected original physical signals need to be normalized by pre-calibrated equipment characteristic parameters: The cutting force is divided by the rated maximum cutting force of the equipment (200N) to obtain the dimensionless force value (Fx / 200=0.6); The acoustic emission signal amplitude is divided by the sensor range (100dB) to convert into a ratio (75dB / 100=0.75); The interface temperature is divided by the material tolerance threshold (silicon rod thermal damage threshold 250℃) to obtain the temperature coefficient (200℃ / 250=0.8); The spindle power is divided by the rated power (5kW) to obtain the power utilization rate (3kW / 5=0.6).

[0024] A set of dimensionless state data array is generated: [0.6, 0.75, 0.8, 0.6], which is independent of the scale of the equipment, for the subsequent parallel prediction of tool health status and material damage.

[0025] 102、Tool health status parallel prediction: input the dimensionless state data into the tool wear dynamics model based on fractional differential equation, which characterizes the memory and genetic characteristics of the diamond abrasive wear process, and predicts the tool health status trend in the future period of time; It should be noted that in the tool health status parallel prediction of silicon rod diamond wire cutting machine, the wear dynamics model based on fractional differential equation is the core of capturing the memory effect (the influence of history on the current) and genetic characteristics (inheritance of state evolution) in the diamond abrasive wear process: Model input and initialization: receive a set of dimensionless state data from the previous step (101): [cutting force coefficient = 0.62, acoustic emission intensity = 0.78, interface temperature coefficient = 0.85, spindle power coefficient = 0.59].

[0026] The model is preset with a fractional order of α = 0.85 (representing the memory decay rate), and the pre-trained parameter weights are loaded, which are calibrated by historical diamond wire wear data.

[0027] Fractional order dynamics prediction process: the model takes the input sequence as a time series segment (the last 50 sampling periods), calculates the dependence of the current state on the historical state through the fractional differential operator, and weights and superimposes the dimensionless value 0.78 of the acoustic emission signal with the acoustic emission data of the previous 10 periods, the weight is determined by the fractional order α = 0.85, which reflects the "memory effect" of wear.

[0028] The model predicts the wear evolution trend in the next 5 minutes through the state transfer function (fractional integral equation) across time scales: the output result is that the health index decreases from the current 0.75 to 0.68, and the decay rate gradually accelerates with time (genetic characteristics).

[0029] Output and threshold comparison: generate a tool health status curve (time series) and compare it with the safety threshold (lower limit of health index 0.60) in real time. If the predicted trend touches the threshold, an early warning (diamond wire remaining life less than 15 minutes) is triggered.

[0030] This process more accurately describes the nonlinear process of wear accumulation through the non-local characteristics of fractional differential equation, which is superior to the integer order model.

[0031] 103、Material subsurface damage parallel prediction: simultaneously input the dimensionless state data into a cross-scale material response model built based on damage mechanics, which establishes the correlation between macroscopic sensing signals and microscopic crack propagation, and predicts the material subsurface damage evolution trend in the future period of time; It should be noted that in the silicon rod diamond wire cutting process, material subsurface damage parallel prediction is realized through a cross-scale material response model, which correlates macroscopic sensing signals with microscopic crack propagation behavior based on damage mechanics principles: Model input and initialization: receive dimensionless state data from step 101: [cutting force coefficient = 0.62, acoustic emission intensity = 0.78, interface temperature coefficient = 0.85, spindle power coefficient = 0.59].

[0032] The model assumes that the material is single crystal silicon (100 crystal orientation) and loads pre-calibrated damage parameters, including crack initiation threshold (equivalent plastic strain threshold 0.0015) and crack propagation rate coefficient (exponent parameter m = 3.2 in Paris formula).

[0033] Cross-scale damage prediction process: the model maps macroscopic dimensionless data to microscopic scale, cutting force coefficient 0.62 corresponds to microscopic stress intensity factor amplitude ΔK = 12 MPa·m 1 / 2 , acoustic emission intensity 0.78 corresponds to dislocation density increment 1.2×10 14 m -2 , interface temperature coefficient 0.85 leads to local thermal stress peak value 85 MPa.

[0034] Based on the damage accumulation rule, the crack propagation depth in the next 5 minutes is calculated: the subsurface crack is predicted to propagate at a rate of 0.15 μm per minute from the current depth of 2.1 μm, with a final depth of 2.85 μm (under the assumption of linear accumulation).

[0035] Meanwhile, the damage distribution characteristics are predicted: the crack mainly propagates along the 〈110〉 crystal orientation, and the maximum crack width is predicted to be 0.45 μm.

[0036] Generate material subsurface damage evolution curve (time-depth / severity) and compare with safety threshold (crack depth upper limit 3.0 μm) in real time. If the predicted trend approaches the threshold, trigger an early warning (subsurface damage is expected to exceed the limit in 8 minutes).

[0037] 104、Multi-objective constrained online optimization: maximize material removal rate while constraining tool health state trend and material subsurface damage evolution trend not to exceed safety threshold, perform online rolling optimization, and generate a set of optimized process control instructions; It should be noted that in the silicon rod diamond wire cutting processing, the online optimization under multi-objective constraints is realized by a rolling optimization algorithm, which takes maximizing material removal rate as the core objective, while strictly following the safety threshold constraints of tool health state and material subsurface damage: Receive the tool health state prediction trend (health index from 0.72 to 0.68 in the next 5 minutes) from step 102 and the material subsurface damage evolution trend (crack depth from 2.5 μm to 2.9 μm) from step 103.

[0038] Set the safety constraint threshold: the lower limit of tool health index is 0.60, and the upper limit of subsurface crack depth is 3.0 μm.

[0039] Optimization objective: maximize material removal rate (calculation benchmark is the current value 120 mm 3 / s).

[0040] Online rolling optimization process: the optimization variable is the process control parameter (feed speed, spindle speed, cutting depth), and the initial parameter group is [feed speed = 0.045 mm / rev, spindle speed = 1200 rpm, cutting depth = 0.15 mm].

[0041] Using multi-objective optimization algorithm (NSGA-II or gradient descent method), calculate the feasibility of parameter combination in each 30-second rolling window: First, check the constraint condition: if the predicted tool health index (0.68) is higher than the threshold (0.60) and the crack depth (2.9 μm) is lower than the threshold (3.0 μm), the current parameter is feasible; Take the maximum material removal rate as the goal, iteratively adjust the parameters (increase the feed speed to 0.048 mm / rev and the spindle speed to 1250 rpm), and re-evaluate the constraint conditions.

[0042] If parameter adjustment leads to predicted value out of limit (feed speed exceeds 0.050 mm / rev, health index may fall below 0.60), the algorithm backtracks to the safe parameter interval.

[0043] Optimization instruction generation: output a set of optimized process control instructions: [feed speed = 0.047 mm / rev, spindle speed = 1230 rpm, cutting depth = 0.15 mm], the predicted material removal rate is increased to 128 mm³ / s, while the tool health index is maintained above 0.65 and the crack depth is controlled within 2.85 μm. Instructions are sent to the lower controller through real-time communication protocol (OPCUA), and the next round of rolling optimization is triggered.

[0044] 105、Instruction mapping and adaptive execution: mapping the optimized process control instructions into physical control instructions for the specific equipment, and adjusting the servo feed unit and spindle drive unit in real time, so as to achieve proactive control of the machining process, and achieve the effects of prolonging tool life and suppressing machining damage at the same time; It should be noted that in the silicon rod diamond wire cutting machine, the "instruction mapping and adaptive execution" step is to convert the optimized process control instructions into physical actions executable by the equipment, and to adjust the actuator in real time, so as to finally achieve precise control of the machining process: Instruction receiving and analysis: the control system receives the optimized process control instructions from step 104: [feed speed = 0.048 mm / rev, spindle speed = 1250 rpm, cooling liquid flow = 1.8 L / min]. These instructions are usually transmitted to the equipment main controller (PLC or special motion control card) through real-time industrial Ethernet (EtherCAT).

[0045] Instruction mapping to physical equipment: the controller maps the optimized instructions into specific physical control signals: feed speed: mapped to the control voltage of the servo feed unit (0.048 mm / rev→analog voltage output 3.8V), driving the ball screw servo motor to feed at the set speed.

[0046] Spindle speed: mapped to the digital set value of the spindle drive unit (1250 rpm→output a specific duty cycle through PNM signal), control the speed of the spindle motor, with an error control within ±5 RPM.

[0047] Cooling parameters: mapped to the opening instruction of the proportional valve (1.8 L / min→valve opening 60%), control the cooling liquid flow through the analog output module (±10V).

[0048] Adaptive execution and real-time adjustment: during the execution of the instructions, the system realizes adaptive adjustment through closed-loop feedback: the grating ruler of the servo feed unit monitors the actual feed position in real time (resolution 0.1 μm), compares it with the target value, and dynamically adjusts the torque of the servo motor through PID algorithm (when the feed resistance increases, the motor torque is instantaneously increased to 110% of the rated value to prevent stalling).

[0049] The spindle drive unit checks the speed in real time through the encoder feedback (sampling frequency 1 kHz), and if the speed fluctuation (due to load change, the speed drops to 1230 RPM) is found, the drive current is immediately adjusted to 12.5 A to ensure that the speed is stable within the range of 1250±10 RPM.

[0050] The system continuously monitors the diamond wire tension (through a tension sensor, range 0-120N), ensuring that the tension is stable at the preset value (28N). If the tension fluctuates more than ±2N, it is quickly compensated through the torque mode of the take-up and pay-off servo motor (response time < 50ms).

[0051] The physical parameters after execution (actual feed speed, spindle power, tension value) are collected in real time and fed back to the dimensionless module of step 101, forming a closed loop control: the actual material removal rate reaches 128mm 3 / s, the tool health index is maintained at 0.65 or above, and the subsurface crack depth is stabilized at 2.85μm, synchronously achieving the goals of efficiency improvement and damage suppression.

[0052] In the embodiment of the present application, a plurality of physical signals are collected in real time and converted into dimensionless data, breaking through the limitations of traditional single physical quantity monitoring, providing a basis for comprehensive and accurate evaluation of the processing state, and the dimensionless processing makes the data universal, facilitating comparative analysis under different equipment and working conditions; the tool wear dynamics model constructed based on the fractional differential equation fully considers the memory and genetic characteristics of diamond abrasive wear, can more accurately describe the nonlinear process of wear accumulation, and is more accurate in prediction than the integer order model, can provide early warning of tool life, avoid unexpected downtime, and improve production efficiency; the cross-scale material response model constructed based on damage mechanics successfully establishes the correlation between macroscopic sensing signals and microscopic crack propagation, can predict the evolution trend of material subsurface damage, provides key basis for controlling processing quality and preventing material failure, and ensures product quality stability; while maximizing material removal rate, the tool health state and material subsurface damage are strictly constrained, the process control parameters are adjusted in real time through a rolling optimization algorithm, and the balance between processing efficiency and equipment and material protection is achieved, improving overall production benefit; the optimization instructions are accurately mapped to physical actions of the equipment, and the execution mechanism is adjusted in real time through closed-loop feedback, ensuring that the processing process strictly follows the optimization instructions, effectively responding to dynamic changes in the processing process, further prolonging tool life, suppressing processing damage, and improving processing precision and stability.

[0053] Please refer to Figure 2 Another embodiment of the semiconductor cutting multifunctional automatic control method in the embodiment of the present application includes: 201, multi-physical field sensing data acquisition and dimensionless: real-time acquisition of cutting force, acoustic emission signal, interface temperature and spindle power physical signals in the processing process, and conversion of them into a group of dimensionless state data independent of the scale of the equipment by using pre-calibrated equipment characteristic data; Specifically, the cutting force signal is collected in real time by a piezoelectric force sensor installed on the main shaft of the device, the acoustic emission signal of material fracture is collected by an acoustic emission sensor coupled in the machining area, the tool-workpiece interface temperature signal is collected by a thermocouple embedded in the tool head, and the real-time power signal is collected by the main shaft driver to generate the original multi-physical field sensing data set; the device characteristic data set pre-stored in the control system is called, the device characteristic data set includes the maximum allowed cutting force, the rated power of the main shaft, the acoustic emission signal saturation value and the tool material temperature resistance; the cutting force signal in the original multi-physical field sensing data set is multiplied by the maximum allowed cutting force to generate a dimensionless equivalent cutting force; the acoustic emission signal amplitude in the original multi-physical field sensing data set is multiplied by the acoustic emission signal saturation value to generate a dimensionless acoustic emission energy; the interface temperature signal in the original multi-physical field sensing data set is multiplied by the tool material temperature resistance to generate a dimensionless interface temperature; the main shaft power signal in the original multi-physical field sensing data set is multiplied by the rated power of the main shaft to generate a dimensionless power load; the dimensionless equivalent cutting force, the dimensionless acoustic emission energy, the dimensionless interface temperature and the dimensionless power load are combined to form a set of dimensionless state data sets independent of the size of the device.

[0054] It should be noted that in the semiconductor wafer cutting process, taking a diamond wire saw device as an example, the steps of multi-physical field sensing data acquisition and dimensionless are as follows: Multi-physical field sensing data acquisition: a piezoelectric three-way force sensor (9257B multi-component cutting force gauge) installed on the main shaft is used to collect the cutting force signal in real time, and the current cutting force is measured to be 128N. The material micro-fracture signal is collected by the acoustic emission sensor (R15 type) coupled in the machining area, and the acoustic emission signal amplitude is measured to be 78dB. The tool-workpiece interface temperature signal is collected by the K-type thermocouple embedded in the tool head, and the interface temperature is measured to be 625℃.

[0055] The real-time power signal is collected by the power monitoring unit of the main shaft driver, and the main shaft power is measured to be 3.2kW.

[0056] These raw data together constitute the original multi-physical field sensing data set: {cutting force: 128N, acoustic emission: 78dB, interface temperature: 625℃, main shaft power: 3.2kW}.

[0057] Calling the device characteristic data set: the characteristic parameters of the device are called from the pre-stored data in the control system: the maximum allowed cutting force is 200N; the acoustic emission signal saturation value is 100dB; the tool material temperature resistance is 800℃; the rated power of the main shaft is 5kW; Dimensionless calculation: Dimensionless equivalent cutting force: the cutting force signal is multiplied by the maximum allowed cutting force, 128N / 200N=0.64; Dimensionless acoustic emission energy: acoustic emission amplitude divided by saturation value, 78 dB / 100 dB = 0.78; Dimensionless interface temperature: interface temperature signal divided by tolerance temperature, 625 °C / 800 °C = 0.781; Dimensionless power load: main shaft power signal divided by rated power, 3.2 kW / 5 kW = 0.64; Generate dimensionless state data set: combine the above calculation results to obtain a set of dimensionless state data independent of equipment scale: [0.64, 0.78, 0.781, 0.64]. This data set can be used for tool health state and material damage prediction in subsequent steps.

[0058] 202、Tool health state parallel prediction: input the dimensionless state data into the tool wear dynamics model constructed based on fractional order differential equation, which characterizes the memory and genetic characteristics of the diamond abrasive wear process, and predict the tool health state trend in the future period of time; Specifically, the dimensionless state data set at the current time is taken as input to initialize the current state vector of the tool wear dynamics model, which at least includes the current dimensionless wear amount and its historical change rate; a pre-labeled model data set is called, which includes the fractional order order representing the memory strength of the wear process, the wear coefficient, and the index data related to the cutting force and temperature; a fractional order numerical algorithm is used to numerically solve the tool wear dynamics model for a future number of control periods, generating a tool state vector sequence in the future period of time, which includes a series of dimensionless wear amount prediction values arranged in chronological order; analyze the evolution trajectory of the tool state vector sequence, extract its change slope and curvature characteristics, and generate tool wear prediction trend data representing the tool health state; output the tool wear prediction trend data as one of the original inputs for multi-objective constraint optimization in the subsequent step, used to constrain the optimization process to prevent excessive tool wear.

[0059] It should be noted that in the semiconductor wafer cutting process, taking a diamond wire saw device as an example, the steps of tool health state parallel prediction are as follows: Model initialization and data input: take the dimensionless state data set at the current time as input, which includes the dimensionless equivalent cutting force (0.64), the dimensionless acoustic emission energy (0.78), the dimensionless interface temperature (0.781), and the dimensionless power load (0.64). Input these data into the tool wear dynamics model constructed based on fractional order differential equation to initialize the current state vector of the model. The state vector includes the current dimensionless wear amount (0.35) and its historical change rate (0.02).

[0060] Call pre-calibrated model parameters: Call pre-calibrated model dataset from control system, which includes: Fractional order representing memory strength of wear process: 0.85; Wear coefficient: 1.2; Exponent related to cutting force: 1.5; Exponent related to temperature: 1.8; These parameters are calibrated from historical machining data, which are used to accurately describe the memory and genetic characteristics of diamond abrasive wear process.

[0061] Numerical solution and prediction sequence generation: Adopt fractional order numerical algorithm (fractional order Adams method) to numerically solve the tool wear dynamics model for the next 5 control cycles (10 seconds each cycle). After solving, generate a tool state vector sequence, which contains a series of dimensionless wear amount prediction values arranged in chronological order: T+10s: wear amount 0.38; T+20s: wear amount 0.42; T+30s: wear amount 0.47; T+40s: wear amount 0.53; T+50s: wear amount 0.60; Trend analysis feature extraction: Analyze the evolution trajectory of the tool state vector sequence and extract its change slope and curvature features. The calculated change slope is 0.045 / cycle (showing an accelerating upward trend), and the curvature radius is 0.12 (indicating that the trend is gradually steep). According to these features, generate tool wear prediction trend data representing the tool health state, and mark that the wear amount will approach the safety threshold (0.65) at T+50s.

[0062] Output prediction trend data: Output tool wear prediction trend data, including future wear amount sequence, slope and curvature features. These data are used as raw input for subsequent multi-objective constrained optimization, which is used to prevent tool over-wear and trigger control instruction adjustment when wear amount exceeds 0.60.

[0063] 203、Material subsurface damage parallel prediction: At the same time, input the dimensionless state data into the cross-scale material response model based on damage mechanics, which establishes the correlation between macroscopic sensing signals and microscopic crack propagation, and predicts the evolution trend of material subsurface damage in the future period; Specifically, using the dimensionless state dataset at the current moment as input, the current state vector of the cross-scale material response model is initialized. The state vector at least includes the current dimensionless damage variables and their historical evolution rates. A pre-calibrated material property dataset is called, which includes fractional order representing the material damage evolution rate, crack propagation coefficient, and sensitivity data related to stress state. A numerical algorithm based on damage mechanics is used to numerically solve the cross-scale material response model for several future control cycles, generating a material state vector sequence over a future period. This sequence contains a series of predicted values ​​of dimensionless damage variables arranged in chronological order. The evolution trajectory of the material state vector sequence is analyzed to identify its accelerated change characteristics, generating subsurface damage prediction trend data representing the degree of internal damage in the material. The subsurface damage prediction trend data is output as another original input for multi-objective constrained optimization in subsequent steps, used to constrain the optimization process to prevent processing damage to the material.

[0064] It should be noted that, in semiconductor wafer dicing, taking a diamond wire saw for cutting silicon wafers as an example, the specific steps for parallel prediction of subsurface material damage are as follows: Model initialization and data input: The current dimensionless state dataset is used as input, which includes dimensionless equivalent cutting force (0.64), dimensionless acoustic emission energy (0.78), dimensionless interface temperature (0.781), and dimensionless power load (0.64). This data is input into a multi-scale material response model based on damage mechanics to initialize the model's current state vector. The state vector contains the current dimensionless damage variables (0.42) and their historical evolution rate (0.05 / cycle).

[0065] Retrieve pre-calibrated material parameters: Retrieve a pre-calibrated material property dataset from the control system. These parameters include: fractional order characterizing the material damage evolution rate: 0.75; crack propagation factor: 1.8. Sensitivity index related to cutting force: 1.6; Sensitivity index related to temperature: 1.9; These parameters were calibrated using historical processing data and offline experiments (scanning electron microscopy (SEM) and transmission electron microscopy (TEM) observation of subsurface damage in processed samples) to establish the correlation between macroscopic sensing signals and microscopic crack propagation.

[0066] Numerical solution and damage prediction sequence generation: Adopting a numerical algorithm based on damage mechanics (considering the deformation behavior and damage mechanism of the material under dynamic load), the cross-scale material response model is numerically solved for the next 5 control periods (10 seconds each period). After solving, a material state vector sequence is generated, which contains a series of dimensionless damage variable prediction values arranged in chronological order: T+10s: damage variable 0.45; T+20s: damage variable 0.49; T+30s: damage variable 0.54; T+40s: damage variable 0.60; T+50s: damage variable 0.67; Trend analysis and feature identification: Analyze the evolution trajectory of the material state vector sequence and identify its accelerated change characteristics. The calculation results show that the change slope increases from 0.04 / period to 0.07 / period (showing an accelerating upward trend), and the curvature radius is 0.15 (indicating that the trend is gradually steep). According to these characteristics, sub-surface damage prediction trend data representing the degree of internal damage of the material are generated and marked as the damage growth accelerates significantly after T+40s, and the damage variable will exceed the safety threshold (0.65) at T+50s.

[0067] Output prediction trend data: Output the sub-surface damage prediction trend data, which includes the future damage sequence, slope, and curvature characteristics. These data serve as another original input for subsequent multi-objective constrained optimization, which is used to constrain the optimization process to prevent material processing damage, and trigger control instructions to adjust to suppress crack propagation when the damage variable exceeds 0.60.

[0068] 204、Online optimization under multi-objective constraints: Perform online rolling optimization to maximize material removal rate while constraining the tool health state trend and material sub-surface damage evolution trend not to exceed the safety threshold, and generate a set of optimized process control instructions; Specifically, the tool wear prediction trend data and the sub-surface damage prediction trend data are received as constraint input; based on the current processing state, a multi-objective optimization problem is constructed with the dimensionless material removal rate as the objective function, and the tool wear prediction trend data not exceeding the preset wear safety threshold and the sub-surface damage prediction trend data not exceeding the preset damage safety threshold as double constraint conditions; a sequential quadratic programming algorithm is used to solve the multi-objective optimization problem in real time, and under the premise of meeting all constraint conditions, the control data sequence of the future number of control periods that optimizes the objective function is calculated, generating a preliminary optimized control sequence; the data of the first control period in the preliminary optimized control sequence is extracted as the optimal control decision at the current time, and a set of instantaneously optimized process control instructions is generated; It should be noted that in the cutting process of semiconductor wafers, taking a diamond wire saw device cutting silicon wafers as an example, the steps of realizing online optimization under multi-objective constraints are as follows: Input and Constraint Reception: Receive the predicted trend data from steps 202 and 203: Tool wear prediction trend data: The dimensionless wear prediction sequence for the next 5 control cycles (10 seconds per cycle) is [0.38, 0.42, 0.47, 0.53, 0.60]. The preset wear safety threshold is 0.65.

[0069] Subsurface damage prediction trend data: The dimensionless damage variable prediction sequence for the next 5 control periods is [0.45, 0.49, 0.54, 0.60, 0.67]. The preset damage safety threshold is 0.65.

[0070] These data serve as the input constraints for the optimization problem.

[0071] Optimization problem formulation: Objective function: to maximize the dimensionless material removal rate ( With the target being the removal rate, which is positively correlated with feed rate and spindle power, it can be expressed as: ,in This is a dimensionless feed rate command. This is a dimensionless spindle power command, where k is the equipment coefficient (k=1.2 in this example).

[0072] Constraints: The predicted tool wear value shall not exceed the wear safety threshold (0.65); the predicted subsurface damage value shall not exceed the damage safety threshold (0.65).

[0073] Decision variables: dimensionless feed rate command f and dimensionless spindle power command P for the future control cycle.

[0074] Solving Sequential Quadratic Programming (SQP): The SQP algorithm is used to solve the above-mentioned constrained nonlinear optimization problem in real time. Under the premise of satisfying all constraints, the algorithm calculates the optimal control sequence for the next 5 control cycles (50 seconds).

[0075] Preliminary optimization of control sequences (partial examples):

[0076] The solution ensures that all predicted wear and damage amounts are strictly below the safety threshold (0.65), while the objective function... It reaches its maximum within the constraints.

[0077] Generate real-time optimized control commands: Extract the data from the first control cycle (T+10s) from the preliminary optimized control sequence as the optimal decision for the current moment: Dimensionless feed rate command: 0.75; Dimensionless spindle power command: 0.70 205、Instruction mapping and adaptive execution: mapping the optimized process control instructions to the physical control instructions of the specific equipment, and adjusting the servo feed unit and the spindle drive unit in real time, so as to realize the proactive control of the machining process, and achieve the effects of prolonging the tool life and inhibiting the machining damage; Specifically, the instant optimized process control instructions are received as the input source, and the instructions at least include dimensionless feed speed instructions and dimensionless spindle power instructions; the pre-stored equipment physical limit data set in the system is called, and the data set includes the maximum allowed feed speed of the servo feed unit, the maximum allowed rotating speed of the spindle drive unit, and the maximum output power of the spindle motor; the dimensionless feed speed instructions are multiplied by the maximum allowed feed speed to generate specific physical feed speed control instructions; the dimensionless spindle power instructions are multiplied by the maximum output power to obtain the expected output power value, which is converted into specific physical spindle rotating speed control instructions according to the spindle characteristic curve; the physical feed speed control instructions and the physical spindle rotating speed control instructions are combined to generate the final executable physical control instruction set; and the executable physical control instruction set is issued to the execution controllers of the servo feed unit and the spindle drive unit in real time, so as to realize the dual effects of prolonging the diamond tool life and inhibiting the stone or ceramic material machining edge collapse in the actual machining process by adjusting the machining data in advance.

[0078] It should be noted that in the semiconductor wafer cutting process, taking a diamond wire saw equipment cutting a 300mm silicon wafer as an example, the steps of realizing instruction mapping and adaptive execution are as follows: Receiving optimized instructions: The instant optimized process control instructions are received from step 204, and the instructions include: dimensionless feed speed instructions: 0.75; dimensionless spindle power instructions: 0.70; Calling equipment physical limit parameters: the following parameters are called from the pre-stored equipment physical limit data set in the control system: the maximum allowed feed speed of the servo feed unit: 50mm / min; the maximum allowed rotating speed of the spindle drive unit: 3000rpm; the maximum output power of the spindle motor: 5kW; Instruction mapping calculation: physical feed speed control instruction calculation: multiplying the dimensionless feed speed instructions by the maximum allowed feed speed: physical feed speed = 0.75 x 50 = 37.5mm / min; Physical spindle rotating speed control instruction calculation: multiplying the dimensionless spindle power instructions by the maximum output power to obtain the expected output power value: expected output power = 0.70 x 5 = 3.5kW; According to the spindle characteristic curve (assuming a linear relationship) of the equipment, the expected output power is converted into a specific spindle rotating speed: =2100rpm; Generate executable physical control instruction set: combine the above calculation results to generate the final executable physical control instruction set: physical feed speed control instruction: 37.5 mm / min; physical spindle speed control instruction: 2100 rpm; Instruction issuing and execution: real-time issue the executable physical control instruction set to the execution controllers of the servo feed unit and spindle drive unit of the equipment. The servo system immediately adjusts the feed speed and spindle speed according to the new instructions to realize proactive control of the machining process.

[0079] 206、After completing the complete machining cycle, collect the surface quality detection data of the machined workpiece and the actual tool wear measurement data, generate the post-validation data set; compare the post-validation data set with the corresponding prediction trend data, calculate the deviation between the predicted value and the actual value, and generate a model accuracy evaluation report; based on the model accuracy evaluation report, use a data adjustment algorithm based on sensitivity analysis to fine-tune specific data in the tool wear dynamics model and the cross-scale material response model, and output the updated model data set; feed the updated model data set back to the prediction model for prediction calculation in the subsequent machining cycle, realizing self-evolution and precision improvement of the model data; It should be noted that after completing the complete cutting machining cycle of a 300mm silicon wafer, the model self-evolution and precision improvement is realized by the following steps: Post-validation data set collection: use a white light interferometer to measure the subsurface damage depth of the machined wafer, and measure the actual maximum damage depth of 18.2 μm. Measure the wear width of the diamond wire saw by electron microscope, and measure the actual wear amount of 125 μm. Combine the process record to generate the post-validation data set: the actual dimensionless damage variable is 0.58, and the actual dimensionless wear amount is 0.62.

[0080] Comparison and analysis of predicted value and actual value: retrieve the prediction trend data in this machining cycle: the predicted damage variable at the end of machining is 0.60, and the predicted wear amount is 0.60. Compare the predicted value with the actual measurement value, and calculate the damage prediction deviation of +0.02 (relative error 3.4%) and the wear prediction deviation of -0.02 (relative error 3.2%).

[0081] Generate model accuracy evaluation report: based on the comparison results, generate the accuracy evaluation report, which indicates that the tool wear model is slightly conservative at the end of machining (the predicted value is 3.2% lower than the actual value), and the material damage model is slightly aggressive (the predicted value is 3.4% higher than the actual value). The report marks the prediction deviation of the model under high temperature and high load working conditions.

[0082] Data adjustment based on sensitivity analysis: using sensitivity analysis algorithm, identify the model parameters that have the greatest impact on the prediction bias: wear coefficient in tool wear model (current value 1.2) and temperature-related sensitivity index in material damage model (current value 1.9). According to the bias direction and data relevance, the wear coefficient is fine-tuned from 1.2 to 1.18, and the temperature sensitivity index is fine-tuned from 1.9 to 1.87.

[0083] Model dataset update and feedback: the updated model parameters are repackaged into a new model dataset, replacing the original parameter version in the system. The updated dataset will be used for prediction calculation in the next processing cycle, realizing the self-evolution of model parameters.

[0084] Through this process, the control system continuously optimizes the prediction model using actual processing data, further improving the prediction accuracy of the next wafer cutting processing cycle, forming a benign self-evolution cycle.

[0085] In the embodiment of the application, the cutting force, acoustic emission signal, interface temperature and spindle power and other multi-physical field signals are collected in real time, and the characteristic data of the pre-calibration equipment are converted into dimensionless state data, breaking through the limitations of traditional single physical quantity monitoring. The dimensionless data is independent of the scale of the equipment, enhancing the universality and comparability of the data, facilitating data integration and analysis under different equipment and working conditions, and providing a reliable data foundation for subsequent accurate prediction and optimization; A tool wear dynamics model based on fractional differential equation and a cross-scale material response model based on damage mechanics are constructed to respectively and in parallel predict the tool health state and the material subsurface damage evolution trend. The fractional differential equation can better represent the memory and genetic characteristics of the wear process, and the damage mechanics model establishes the correlation between the macroscopic sensing signal and the microscopic crack propagation, improving the accuracy and foresight of the prediction, helping to discover potential problems in advance and avoid excessive tool wear and material processing damage. With the goal of maximizing material removal rate, while keeping the tool health state trend and material subsurface damage evolution trend within the safety threshold as dual constraints, online rolling optimization is performed to maximize the material removal rate while ensuring processing quality and tool life, improving processing efficiency and economic benefits, and realizing multi-objective collaborative optimization of the processing process. The optimized process control instructions are mapped to the physical control instructions of the specific equipment, and the servo feed unit and spindle drive unit are adjusted in real time, realizing proactive control of the processing process, and enabling timely adjustment of processing parameters based on real-time prediction results, thereby achieving the effects of extending tool life and suppressing processing damage, and improving the stability and reliability of the processing. After completing the processing cycle, actual data is collected to generate post-validation data sets, compared with predicted trend data, calculate the deviation and generate model accuracy evaluation report, based on sensitivity analysis fine-tune model parameters, realize the self-evolution of model data, through the actual processing data constantly optimize the model, make the model can adapt to different working conditions and material characteristics, continuously improve the prediction accuracy, form a virtuous cycle, provide more accurate prediction and optimization support for subsequent processing, further improve the processing quality and efficiency.

[0086] Figure 3 The semiconductor cutting multifunctional automatic control device 300 can have a large difference due to different configurations or performances, and can include one or more central processing units (CPUs) 310 (for example, one or more processors) and a memory 320, one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and the storage medium 330 can be temporary storage or persistent storage. The programs stored in the storage medium 330 can include one or more modules (not shown in the figure), each of which can include a series of instruction operations in the semiconductor cutting multifunctional automatic control device 300. Further, the processor 310 can be configured to communicate with the storage medium 330 and execute a series of instruction operations in the storage medium 330 on the semiconductor cutting multifunctional automatic control device 300.

[0087] The semiconductor cutting multifunctional automatic control device 300 can also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, MacOS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that the semiconductor cutting multifunctional automatic control device 300 can also include other components, such as a bus, an input / output interface, a user interface, a display, a storage medium, etc. Figure 3 The semiconductor cutting multifunctional automatic control device structure shown does not constitute a limitation on the semiconductor cutting multifunctional automatic control device, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0088] The present application also provides a semiconductor cutting multifunctional automatic control device, which includes a memory and a processor, the memory stores computer readable instructions, and the computer readable instructions are executed by the processor to make the processor execute the steps of the semiconductor cutting multifunctional automatic control method in each of the above embodiments.

[0089] The application further provides a computer readable storage medium, which can be a nonvolatile computer readable storage medium or a volatile computer readable storage medium, and the computer readable storage medium stores instructions, and the instructions make a computer execute the steps of the multifunctional automatic control method for semiconductor truncation when the instructions are run on the computer.

[0090] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0091] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the application or the whole or part of the technical solutions that make essential contributions to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0092] The above-described embodiments are only used to illustrate the technical solutions of the application, rather than limit the application; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A multifunctional automated control method for semiconductor cutoff, characterized by, The semiconductor cutting-off multifunctional automatic control method comprises: Real-time acquisition of cutting force, acoustic emission signal, interface temperature and spindle power physical signals in the processing process, and conversion of the same into dimensionless state data by using pre-calibrated equipment characteristic data; Inputting the dimensionless state data into a tool wear dynamics model, the model representing the memory and genetic characteristics of the diamond abrasive wear process, and predicting a tool health state trend; Inputting the dimensionless state data into a cross-scale material response model to obtain a material subsurface damage evolution trend; Maximizing material removal rate as an objective, while taking the tool health state trend and the material subsurface damage evolution trend not exceeding a safety threshold as a constraint, performing online rolling optimization to generate a process control instruction.

2. The multi-functional automated control method for semiconductor cutoff according to claim 1, characterized by, It comprises: Real-time acquisition of cutting force signal, acoustic emission signal, interface temperature signal and spindle power signal to generate an original multi-physical field sensing data set; Calling a pre-stored equipment characteristic data set, the equipment characteristic data set including maximum allowable cutting force, spindle rated power, acoustic emission signal saturation value and tool material temperature resistance; Taking the cutting force signal in the original multi-physical field sensing data set and the maximum allowable cutting force to generate a dimensionless equivalent cutting force; Taking the acoustic emission signal amplitude in the original multi-physical field sensing data set and the acoustic emission signal saturation value to generate a dimensionless acoustic emission energy; Taking the interface temperature signal in the original multi-physical field sensing data set and the tool material temperature resistance to generate a dimensionless interface temperature; Taking the spindle power signal in the original multi-physical field sensing data set and the spindle rated power to generate a dimensionless power load; Combining the dimensionless equivalent cutting force, the dimensionless acoustic emission energy, the dimensionless interface temperature and the dimensionless power load to form a dimensionless state data set.

3. The multi-functional automated control method for semiconductor cutoff according to claim 2, characterized by, It comprises: Taking the dimensionless state data set at the current time as input to initialize the current state vector of the tool wear dynamics model, the state vector at least including the current dimensionless wear amount and its historical change rate; Calling a pre-calibrated model data set, the model data set including wear fractional order, wear coefficient and exponential data; Using a fractional order numerical algorithm to numerically solve the tool wear dynamics model for a plurality of future control periods to generate a tool state vector sequence; Analyzing the evolution trajectory of the tool state vector sequence to extract its change slope and curvature characteristics to generate tool wear prediction trend data.

4. The multi-functional automated control method for semiconductor cutoff according to claim 3, characterized by, It comprises: Taking the dimensionless state data set at the current time as input to initialize the current state vector of the cross-scale material response model, the state vector at least including the current dimensionless damage variable and its historical evolution rate; Calling a pre-calibrated material characteristic data set, the data set including damage fractional order, crack propagation coefficient and sensitivity data; Using a damage mechanics-based numerical algorithm to numerically solve the cross-scale material response model for a plurality of future control periods to generate a material state vector sequence; Analyzing the evolution trajectory of the material state vector sequence to identify its accelerated change characteristics to generate subsurface damage prediction trend data.

5. The multi-functional automated control method for semiconductor cutoff according to claim 4, characterized by, The method comprises: receiving the tool wear prediction trend data and the subsurface damage prediction trend data as constraint input; based on the current machining state, constructing a multi-objective optimization problem with the dimensionless material removal rate as the objective function, and the tool wear prediction trend data not exceeding the preset wear safety threshold and the subsurface damage prediction trend data not exceeding the preset damage safety threshold as double constraint conditions; using a sequential quadratic programming algorithm to solve the multi-objective optimization problem in real time, and obtaining control data sequences of future control periods that make the objective function optimal under the premise of meeting all constraint conditions, to generate a preliminary optimized control sequence; extracting the data of the control periods in the preliminary optimized control sequence as optimal control decisions at the current time, and generating process control instructions.

6. The multi-functional automated control method for semiconductor cutoff according to claim 5, wherein The dimensionless material removal rate is set as the objective function, and the dimensionless material removal rate is set as : ； wherein, is the dimensionless feed speed command, is the dimensionless spindle power command, k is the equipment coefficient.

7. The multi-functional automated control method for semiconductor cutoff according to claim 5, wherein The method also comprises mapping the process control instructions to physical control instructions of specific equipment, and adjusting the servo feed unit and the spindle drive unit in real time, so as to realize proactive control of the machining process, and simultaneously achieve the effects of prolonging tool life and suppressing machining damage: receiving the process control instructions as input sources, including dimensionless feed speed instructions and dimensionless spindle power instructions; calling a pre-stored equipment physical limit data set, including the maximum allowed feed speed of the servo feed unit, the maximum allowed speed of the spindle drive unit, and the maximum output power of the spindle motor; multiplying the dimensionless feed speed instructions by the maximum allowed feed speed to generate physical feed speed control instructions; multiplying the dimensionless spindle power instructions by the maximum output power to obtain an expected output power value, and then converting it to physical spindle speed control instructions according to the spindle characteristic curve; combining the physical feed speed control instructions and the physical spindle speed control instructions to generate an execution physical control instruction set.

8. The multi-functional automated control method for semiconductor cutoff according to claim 1, characterized by, The method also comprises: after completing a complete machining cycle, collecting surface quality detection data of the machined workpiece and actual tool wear measurement data to generate a post-validation data set; comparing and analyzing the post-validation data set and the corresponding prediction trend data to obtain the deviation between the predicted value and the actual value, and generating a model accuracy evaluation report; based on the model accuracy evaluation report, using a data adjustment algorithm based on sensitivity analysis to fine-tune the data in the tool wear dynamics model and the cross-scale material response model, and outputting an updated model data set.

9. A multifunctional automated control apparatus for semiconductor cutoff, characterized by comprising: The semiconductor cutting multifunctional automatic control device comprises a memory and at least one processor, and the memory stores instructions; the at least one processor calls the instructions in the memory to enable the semiconductor cutting multifunctional automatic control device to perform the semiconductor cutting multifunctional automatic control method according to any one of claims 1-8.

Images