Chip automatic conveying system of die bonder

Abstract

The application relates to the field of semiconductor packaging, and discloses a die bonder chip automatic conveying system for high-speed anti-falling transfer and precise placement of semiconductor bare chips. The system comprises a chip physical transfer mechanism with a vacuum suction nozzle and a die bonder main controller. According to the physical constraints such as the suction force of the nozzle and the material load, the system constructs a speed, acceleration and jerk constraint parameter set for preventing chip inertia slip and flying pieces; a dimensionless model is established in combination with the motion stroke, and the optimal smooth conveying trajectory is solved with the fastest handling rhythm and the smallest mechanical impact as the targets; further, the feedforward control amount for overcoming the nonlinear resistance of the guide rail and the periodic disturbance is constructed and iteratively corrected in combination with the instruction differential and the historical period handling position error, and an initial compensation table is established through no-load trial operation before the first operation. The application eliminates chip flying and end shaking in high-speed transfer, and improves the handling rhythm and placement accuracy of the equipment under different material working conditions.

Description

Technical Field

[0001] This invention relates to the field of semiconductor packaging, and more particularly to an automatic chip delivery system for a die bonder. Background Technology

[0002] Die bonders are one of the core pieces of equipment in the semiconductor packaging process, primarily used to precisely pick up chips from the blue film on wafers and place them at designated locations on the lead frame or substrate. As semiconductor devices evolve towards miniaturization and high density, chip sizes continue to shrink and pin pitches narrow, placing stringent demands on the delivery accuracy, production efficiency, and stability of the die bond process. Among these, the automated chip delivery process (i.e., the mechanical handling process where the nozzle picks up the bare chip and rapidly transfers it from the pick-up position to the bond position) is crucial for die bonders to achieve high throughput and high-precision placement of materials. Due to the increasing thinness and fragility of semiconductor dies, the physical clamping stability and mechanical movement smoothness of the transfer mechanism face severe challenges.

[0003] Existing chip transfer equipment typically employs a conventional acceleration / deceleration conveying mode with fixed operating parameters. However, due to the ineffective handling of the dynamic mechanical boundaries of material handling equipment, existing equipment exhibits significant physical limitations: Existing equipment does not mechanically couple its maximum handling acceleration with the vacuum suction holding force of the nozzle end face and the mass of the chip to be transported in real time. When switching to thin and brittle chips or accelerating the handling, if the mechanical acceleration and deceleration inertial force exceeds the static friction holding limit of the suction cup on the chip, it will directly cause serious material damage accidents such as chip end face displacement, deflection, or even centrifugal fling in the air (flying chip); if the equipment speed is blindly reduced in order to prevent flying chips, it will directly slow down the output cycle of the entire assembly line. The chip delivery mechanism is prone to mechanical resonance between the guide rail and the cantilever during start-up and stop, and long-stroke guide rails are often accompanied by nonlinear resistance and static-dynamic friction conversion. Traditional drive devices lack the ability to compensate for disturbances in the current torque based on the deviation of the previous measured transport trajectory, which often results in residual mechanical vibration and shaking when the nozzle carries the chip to the placement target point, causing sub-micron level positional offset at the moment of chip placement.

[0004] Therefore, we propose an automated chip delivery system for die bonders to solve the above problems. Summary of the Invention

[0005] This invention provides an automatic chip delivery system for die bonders, used for high-speed, anti-drop-off transfer and precise placement of bare semiconductor chips.

[0006] The first aspect of this invention provides an automatic chip conveying system for a die bonder, including a chip physical transfer mechanism, having a nozzle for holding a bare semiconductor chip in a vacuum adsorption at its end face, and a servo drive actuator for driving the nozzle to carry the chip in a reciprocating transfer motion between a pick-up position and a die bonder position; a die bonder main controller, communicatively connected to the chip physical transfer mechanism, for real-time control of the chip handling path and the power of the transfer equipment, wherein the die bonder main controller is internally configured with: a constraint parameter module, used to calculate speed threshold, acceleration threshold, and jerk threshold to prevent the bare chip from inertial slippage or falling off during high-speed displacement, based on the physical constraints and conveying task requirements of the system in material handling, forming a constraint parameter set, wherein the conveying task requirements include a motion stroke; and a motion model module, used to establish a dimensionless motion model based on the conveying time to be optimized and the motion stroke, to define Discretized dimensionless motion state; optimization solution module, used to achieve the fastest transfer cycle and minimum mechanical vibration impact, taking the cost function of the transport time and dimensionless accelerometer as the optimization objective, solves the optimization problem under the constraints of the parameter set, and obtains the optimal transport time and the corresponding optimal dimensionless accelerometer sequence; instruction acquisition module, used to calculate the discrete dimensionless position sequence based on the optimal dimensionless accelerometer sequence and the optimal transport time, and obtain the chip handling position movement instruction through inverse normalization mapping; cooperative control module, used to obtain the basic feedforward control quantity to overcome mechanical resistance based on the position movement instruction, and correct the basic feedforward control quantity based on historical position error to form an updated feedforward control quantity, and use the updated feedforward control quantity and feedback control quantity to co-drive the die bonder's actuator to make the nozzle accurately reach the die bond position, and complete the stable chip transport.

[0007] Optionally, in a first implementation of the first aspect of the present invention, the physical constraints of the system include the rated speed of the servo driver, the peak torque of the motor, the load inertia, the vacuum suction force of the nozzle, the current loop bandwidth, and the resonant frequency of the mechanical structure; the delivery task requirements also include the current chip quality; the constraint parameter module is specifically used for: The rated rotational speed of the servo driver is obtained as the speed threshold; A first acceleration upper limit is determined based on the peak torque of the motor and the load inertia, and a second acceleration upper limit is determined based on the vacuum suction force of the suction nozzle and the current chip mass. The smaller value between the first acceleration upper limit and the second acceleration upper limit is taken as the acceleration threshold. A first jerk upper limit is determined based on the current loop bandwidth and the acceleration threshold, and a second jerk upper limit is determined based on the mechanical structure resonant frequency. The smaller value between the first jerk upper limit and the second jerk upper limit is taken as the jerk threshold.

[0008] Optionally, in a second implementation of the first aspect of the present invention, the discretized dimensionless motion state includes a normalized time variable, a normalized position variable, a normalized velocity variable, a normalized acceleration variable, and a normalized jerk variable; the motion model module is specifically used for: The conveying time is used as the time normalization coefficient, and the motion stroke is used as the displacement normalization coefficient. The normalized time variable is defined as the ratio of the actual time to the time normalization coefficient, and the normalized position variable is defined as the difference between the actual displacement of the chip and the preset motion starting position divided by the displacement normalization coefficient. The normalized position variable is defined sequentially with respect to the normalized time variable to obtain the normalized velocity variable, the normalized acceleration variable, and the normalized jerk variable, respectively.

[0009] Optionally, in a third implementation of the first aspect of the present invention, the optimization solution module is specifically used for: The time interval corresponding to the delivery time is divided into multiple sub-intervals, and a quadratic programming problem is constructed with the dimensionless accelerometer constant in each sub-interval as the decision variable, which is the optimization problem. Within a preset time range, multiple candidate delivery times are selected. Under the constraints of the set of constraint parameters, the quadratic programming problem is solved for each candidate delivery time to obtain the corresponding cost function value. The candidate transport time with the minimum cost function value is taken as the optimal transport time, and the corresponding combination of dimensionless accelerometer constants is taken as the optimal dimensionless accelerometer sequence.

[0010] Optionally, in a fourth implementation of the first aspect of the present invention, when the optimization solution module selects multiple candidate delivery times within a preset time range, it is specifically used for: The theoretical shortest delivery time is calculated based on the motion stroke, the speed threshold, and the acceleration threshold, and is taken as the minimum feasible value. The interval between the minimum feasible value and the preset maximum feasible value is divided into equal parts with a fixed time step to obtain multiple candidate delivery times, wherein the fixed time step is an integer multiple of the servo cycle of the actuator.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, the cost function employs a calculation formula that includes a dimensional calibration factor: ; in, The cost function value; The candidate delivery time to be substituted; For exercise journey; This represents the total number of sub-intervals; Let be the dimensionless accelerometer constant for the k-th subinterval. Square it; This is a time-weighted coefficient, which has the dimension of the reciprocal of time to eliminate the dimension of time. This is the jerk energy weighting coefficient.

[0012] Optionally, in a sixth implementation of the first aspect of the present invention, the instruction obtaining module is specifically used for: Based on the optimal dimensionless acceleration sequence, the normalized acceleration, normalized velocity, and normalized position at the endpoint of each sub-interval are calculated one by one using the numerical integration method to obtain the discrete dimensionless position sequence. Multiply the values ​​in the discrete dimensionless position sequence by the motion distance and combine them with the preset motion starting position to convert them into an actual position sequence; The actual position sequence is combined with the actual time point deduced from the optimal delivery time to form a time-position curve as the position movement command.

[0013] Optionally, in the seventh implementation of the first aspect of the present invention, when the cooperative control module corrects the basic feedforward control quantity based on the historical position error to form the updated feedforward control quantity, it is specifically used for: The position motion command is differentiated to obtain the corresponding velocity command value and acceleration command value; Based on the speed command value, the acceleration command value, and the system load parameters, the basic feedforward torque is calculated as the basic feedforward control quantity; The historical position error is multiplied by a preset learning gain to obtain the correction compensation amount; The corrected compensation amount is superimposed on the basic feedforward torque to form an updated feedforward torque, which is then used as the updated feedforward control amount.

[0014] Optionally, in an eighth implementation of the first aspect of the present invention, an error initialization module is further included, used for: Before the first chip delivery, an unloaded test run was performed, and the initial deviation sequence between the actual position and the theoretical position was recorded during the unloaded test run. The initial deviation sequence is used as the historical position error during the first transport; and at the end of each subsequent transport cycle, the deviation between the actual position of the current cycle and the theoretical position corresponding to the position movement command is updated as the historical position error for the next transport cycle.

[0015] Optionally, in the ninth implementation of the first aspect of the present invention, the error initialization module, when recording the initial deviation sequence between the actual position and the theoretical position during the no-load trial run, is specifically used for: During the no-load test run, multiple different constant speed segment operation operations were performed, and the initial deviation within each constant speed segment was recorded; The feedforward torque correction value is calculated based on the speed value of each constant speed segment and the corresponding initial deviation. A correction lookup table is constructed that includes the mapping relationship between the speed value and the feedforward torque correction value. The initial deviation sequence is then output based on the correction lookup table.

[0016] Beneficial effects: It constructs a safe and non-detachable boundary for adsorption, eliminates chip flying and slippage during high-speed chip handling, and transforms the mechanical relationship between the vacuum adsorption holding force of the suction nozzle and the chip quality into an acceleration threshold constraint of the equipment. This ensures that the mechanical transfer speed is always kept within the absolute safety zone of material anti-slip, solving the industry problem of chip deflection and flying caused by inertial impact of traditional conveying equipment. The system optimizes smooth handling throughout the entire process, significantly improving the material conveying throughput: Under the constraints of anti-drop and anti-resonance, the system automatically generates a new conveying route that balances the shortest time and the lowest impact, minimizing the reciprocating transfer time of the suction nozzle between the crystal picking position and the crystal solidification position, and effectively reducing the mechanical stress loss of the guide rail and structural components. By combining historical handling deviations for dynamic torque compensation, this system accurately overcomes the nonlinear frictional resistance and elastic deformation interference of the mechanical actuator; in particular, it completely eliminates the residual vibration at the end of the nozzle when carrying material to the die bonding point, so that the chip is in an absolutely stable and static state at the moment of placement, ensuring the sub-micron level mounting and positioning accuracy of the chip. By automatically generating initial compensation data through a simple trial run, the machine can quickly and adaptively adjust the drive torque for bare wafers of different quality specifications, eliminating the need for a large amount of manual trial and error debugging due to changes in material specifications, and improving the intelligence and operational level of the assembly line. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of one embodiment of the automatic chip delivery system for die bonders in this invention.

[0018] Figure 2 This is a schematic diagram of another embodiment of the die bonder chip automatic delivery system in this invention.

[0019] Figure 3 This is a flowchart illustrating the processing steps of the automatic chip delivery system for the die bonder in an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of one embodiment of the automatic chip delivery equipment for die bonding machines in this invention. Detailed Implementation

[0021] This invention provides an automated die bonding system for high-speed, anti-drop-off transfer and precise placement of bare semiconductor dies. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the die bonder's automatic chip delivery system in this invention includes a chip physical transfer mechanism, having a suction nozzle for vacuum adsorption of bare semiconductor chips, and a servo drive actuator for driving the suction nozzle to reciprocate between the chip pick-up position and the die bonder position; a die bonder main controller, communicatively connected to the chip physical transfer mechanism, and the die bonder main controller is internally configured with: 101. Constraint Parameter Module: This module is used to calculate the speed threshold, acceleration threshold, and jerk threshold based on the servo driver current loop bandwidth, mechanical structure resonant frequency, and nozzle vacuum suction force of the die bonder, combined with the current chip mass and motion stroke, thus forming a constraint parameter set.

[0023] It is understood that the executing entity of this invention can be a die bonder chip automatic delivery device, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0024] It should be noted that during chip delivery operations on a certain type of die bonder, the system acquires basic hardware evaluation parameters and current operating conditions. The acquired data is as follows: servo driver current loop bandwidth 2000Hz, mechanical structure resonant frequency 150Hz, nozzle vacuum suction force 0.08N, static friction coefficient between nozzle and chip surface 0.4, current chip mass 0.001kg, current travel distance 0.03m, and maximum physical speed limit of the equipment guide rail 1.5m / s.

[0025] The chip maintains its relative position during translation using the static friction of the suction nozzle. The maximum static friction force is the product of the vacuum suction force and the coefficient of static friction, which is 0.032 N. According to Newton's second law, dividing this force by the chip's mass yields a critical acceleration of 32 m / s². 2 To ensure stable operation and allow for control margin, the system cuts off to 30 m / s. 2 The acceleration threshold is .

[0026] The electrical response at 2000Hz is extremely fast, but the mechanical resonant frequency of 150Hz limits the maximum jerk to prevent the excitation of mechanical resonance. The system uses an internal suppression mapping table to calculate that the maximum permissible jerk under the 150Hz limit is 3000 m / s². 3 Compare the two values ​​and take the smaller one to determine 3000 m / s. 3 To add a jerk threshold.

[0027] Finally, the velocity threshold is derived. Within a short stroke of 0.03m, assuming the motion consists entirely of acceleration and deceleration, the theoretically achievable maximum speed is equal to the square root of the product of the acceleration threshold and the stroke. (30m / s²) 2 Multiplying by 0.03m gives 0.9, and its square root is approximately 0.94m / s, which is below the physical speed limit of 1.5m / s. To ensure smooth operation, the system sets a speed threshold of 0.9m / s.

[0028] The system obtained a velocity threshold of 0.9 m / s and an acceleration threshold of 30 m / s². 2 3000m / s² accelerometer threshold 3 The set of constraint parameters.

[0029] 102. Motion Model Module: This module is used to establish a dimensionless motion model. The model uses the total transport time as the time normalization coefficient and the motion stroke as the displacement normalization coefficient. It defines the mapping relationship between normalized time variables, normalized position variables, normalized velocity variables, normalized acceleration variables, and normalized jerk variables, thus forming a definition scheme for the dimensionless motion state.

[0030] It should be noted that the system specifies the current travel distance as 0.03m. To find the optimal time, the system sets a candidate value for the total transport time for evaluation; here, 0.1s is selected as the current total time being evaluated. The system sets the displacement normalization factor to 0.03m and the time normalization factor to 0.1s.

[0031] The system defines the conversion relationship between the actual physical state and the dimensionless state. Taking the intermediate moment of 0.02s as an example: Time mapping: The actual time 0.02s divided by the total time 0.1s yields a normalized time of 0.2. Position mapping: Assuming the actual displacement at this moment is 0.00125m, dividing it by the travel distance of 0.03m yields a normalized position of approximately 0.0416.

[0032] Speed ​​mapping: Assuming the actual speed at this moment is 0.3m / s, multiply it by the total time of 0.1s and divide it by the distance traveled of 0.03m to obtain a normalized speed of 1.0.

[0033] Acceleration mapping: Assume the actual acceleration at this moment reaches 30 m / s². 2 Multiplying it by the square of the total time (0.01) and then dividing by the distance traveled 0.03m, we get the normalized acceleration as 10.0.

[0034] Acceleration mapping: Assume the actual acceleration at this moment is 1500 m / s². 3 Multiplying it by the cube of the total time (0.001) and then dividing by the distance traveled 0.03m, we get the normalized jerk as 50.0.

[0035] Based on the above calculation rules, a numerical proportional model was established inside the die bonder controller, and the mapping relationship was established.

[0036] 103. The optimization solution module is used to discretize the dimensionless motion state definition scheme at the cost of the weighted sum of the total transport time and the square integral of the jerk. It constructs a quadratic programming problem with the jerk constant in each segment as the decision variable. The quadratic programming problem is solved by scanning the candidate values ​​of the total transport time to obtain the optimal total transport time and the corresponding jerk constant sequence.

[0037] The system defines the candidate time scan range as 0.08s to 0.15s, with a step size of 0.01s. The dimensionless time interval is divided into 10 discrete segments, each representing one-tenth of the period. The system sets the dimensionless accelerometer within these 10 discrete segments as 10 independent decision variables.

[0038] When the system scans to a candidate time of 0.1s, it performs boundary constraint transformation according to the mapping rules in step 102: the dimensionless accelerometer limit is equal to the physical threshold (3000) multiplied by the cube of time and then divided by the travel distance, resulting in 100; the dimensionless accelerometer limit is equal to the physical threshold (30) multiplied by the square of time and then divided by the travel distance, resulting in 10; the dimensionless velocity limit is transformed into 3.

[0039] The system inputs a cost function, calculated as a weighted sum of the sum of squares of the decision variables multiplied by 0.1s (time cost weight), the sum of squares of the decision variables, and the weight of the jerk cost, into the quadratic programming solver. Under the premise of not exceeding the aforementioned boundaries (i.e., dimensionless jerk not exceeding 10, jerk not exceeding 100), the solver outputs a sequence of 10 decision variables that minimizes the cost: 50, 50, -50, -50, 0, 0, -50, -50, 50, 50. The system detects that the maximum dimensionless jerk in this sequence is 50 (less than the boundary 100), and the maximum dimensionless jerk obtained through integration is 10 (not exceeding the boundary 10), satisfying all physical constraints.

[0040] After comparing the entire range, the shorter time (0.08s) is deemed unsolvable or will cause severe mechanical impact due to the overly strict upper limit of dimensionless acceleration. The system determines 0.1s as the optimal total transport time with the lowest cost among the feasible solutions and outputs the above sequence as the corresponding optimal jerk constant sequence.

[0041] 104. The instruction acquisition module is used to perform integral recursion based on the jerk constant sequence to obtain a discrete dimensionless position sequence, and then obtain the actual time-position curve as the position motion instruction through inverse normalization mapping.

[0042] It should be noted that the actual motion distance is 0.03m, the optimal total delivery time is 0.1s, and the dimensionless time step is 0.1. The optimized dimensionless acceleration sequence is: 50, 50, -50, -50, 0, 0, -50, -50, 50, 50.

[0043] The system performs dimensionless integration. In the first segment (dimensionless time 0 to 0.1), the jerk is constant at 50, and the acceleration increases linearly from 0 to 5. The velocity integral increases to 0.25, and the position integral accumulates to 0.0083.

[0044] In the second segment (dimensionless time 0.1 to 0.2), the jerk remains at 50, and the acceleration reaches a peak of 10 at the end of time 0.2. The velocity integral increases to 1.0, and the position integral accumulates to 0.0416.

[0045] After completing ten recursive steps, the dimensionless acceleration and velocity precisely return to 0, and the dimensionless position reaches 1.0. Subsequently, the system performs inverse normalization mapping to obtain physical commands.

[0046] Taking the endpoint of the second segment (dimensionless time 0.2, dimensionless position 0.0416) as an example for mapping: The corresponding actual time is equal to the dimensionless time multiplied by the total time, that is, 0.2 multiplied by 0.1s, which gives 0.02s.

[0047] The corresponding actual acceleration is equal to the dimensionless acceleration 10 multiplied by the distance traveled 0.03m, then divided by the square of the total time, calculated to be 30m / s². 2 This value is exactly equal to the maximum safe acceleration threshold set in step 101, proving that the command is absolutely safe and makes full use of the device's performance.

[0048] The corresponding actual position is equal to the dimensionless position 0.0416 multiplied by 0.03m, which gives 0.00125m (i.e. 1.25mm).

[0049] The actual time (from 0.01s to 0.1s) of these 10 discrete nodes is smoothly connected with the corresponding actual position data to obtain a complete time-position curve, which is then sent to the controller as a command.

[0050] 105. The collaborative control module is used to send position movement commands to the die bonder motion controller. At the same time, it constructs a feedforward torque based on the speed and acceleration corresponding to the position movement command, and iteratively learns and corrects the feedforward torque based on the position error of the previous delivery cycle to form an updated feedforward torque. Together with the feedback control, it drives the robotic arm to complete the automatic chip delivery.

[0051] It should be noted that the motion controller parses commands with a sampling period of 0.01s. Taking the 0.02s mark of the chip acceleration phase as an example, if the target position of the command is 1.25mm, the corresponding command acceleration is 30m / s². 2 The commanded speed is 0.3 m / s. Based on the nominal moment of inertia and friction model, the controller calculates the basic dynamic feedforward torque at this moment to be 0.40 N·m.

[0052] To overcome unmodeled mechanical disturbances, the system retrieves the position error record from the same moment in the previous transport cycle (cycle 5). Cycle 5 exhibits a 10µm hysteresis error at 0.02s. Using a learning gain of 0.005 N·m / µm, the system calculates a feedforward correction of 0.05 N·m. The system then superimposes the base feedforward and the correction to obtain a feedforward torque of 0.45 N·m after updating at this node in cycle 6.

[0053] At 0.02s of the sixth cycle, the 0.45 N·m feedforward torque is directly injected into the servo drive. Simultaneously, the feedback controller outputs a feedback torque of 0.01 N·m based on the real-time residual error. The combined torque of the two is 0.46 N·m, driving the robotic arm.

[0054] Table 1 below shows the data evolution and computational consistency at three typical time points from cycle 5 to cycle 6: Table 1 0.02 (end of acceleration phase) 1.25 30.0 10 0.40 0.45 2 0.05 (Center of uniform velocity) 15.00 0.0 4 0.10 0.12 1 0.08 (Initial deceleration phase) 28.75 -30.0 -8 -0.30 -0.34 -1 The updated feedforward torque accurately predicts dynamic demands and converges the trajectory error to within 2µm without exceeding any safety boundaries.

[0055] Please see Figure 2 , Figure 3 Another embodiment of the die bonder's automatic chip delivery system in this invention includes a chip physical transfer mechanism, having a suction nozzle for vacuum adsorption of bare semiconductor chips, and a servo drive actuator for driving the suction nozzle to reciprocate between the chip pick-up position and the die bonder position; a die bonder main controller, communicatively connected to the chip physical transfer mechanism, and the die bonder main controller is internally configured with: 201. Constraint Parameter Module: This module is used to calculate the speed threshold, acceleration threshold, and jerk threshold based on the servo driver current loop bandwidth, mechanical structure resonant frequency, and nozzle vacuum suction force of the die bonder, combined with the current chip mass and motion stroke, thus forming a constraint parameter set.

[0056] Specifically, the rated speed of the servo drive is obtained as the reference value for the speed threshold; the first acceleration upper limit is obtained based on the ratio of the motor peak torque to the load inertia, and the second acceleration upper limit is obtained based on the ratio of the suction nozzle vacuum suction force to the chip mass. The smaller value between the first and second acceleration upper limits is taken as the acceleration threshold; the first jerk upper limit is obtained by multiplying the current loop bandwidth by the acceleration threshold, and the second jerk upper limit is obtained by multiplying the square of the mechanical structure resonant frequency by a preset coefficient. The smaller value between the first and second jerk upper limits is taken as the jerk threshold; the speed threshold, acceleration threshold, and jerk threshold together form a constraint parameter set, which is used to normalize the motion time and displacement.

[0057] It should be noted that before a single chip delivery task begins, the control system reads basic data from the underlying servo driver, mechanical sensors, and the host computer's engineering configuration file. The linear motion distance from the chip pick-up point to the placement point for the current delivery batch is determined to be 0.15m. The rated speed of the servo driver, after conversion based on the mechanical transmission ratio, corresponds to a linear velocity baseline value of 1.5m / s. After conversion based on the motor peak torque and the system's equivalent load inertia through the lead screw or linear motor thrust, the theoretical maximum physical acceleration that the hardware can output is 50m / s². 2 The servo driver's underlying current loop bandwidth parameter is set to 800Hz. The first-order resonant frequency of the mechanical transmission structure, obtained through frequency sweep testing, is 60Hz. The current system's preset anti-resonance safety factor for this machine is 8. The sensor measures a stable vacuum suction force of 0.03N after the nozzle is opened, and the mass of the bare chip to be transported is 0.001kg.

[0058] Based on the multi-dimensional basic parameters collected above, the system sequentially performs boundary value calculations to obtain various thresholds. For the speed threshold, the system directly obtains the linear velocity corresponding to the rated speed of the servo drive as a reference, strictly limiting the speed threshold value to 1.5 m / s to ensure that the motor does not operate in the field weakening overspeed zone.

[0059] Next, the acceleration threshold is determined. The upper limit of acceleration is limited by the motor's own physical and electromagnetic driving capability and the vacuum nozzle's ability to hold the chip's end face. These two physical extremes need to be calculated separately to determine a safe boundary. The first upper limit of acceleration is determined by the ratio of the motor's peak torque to the load inertia, with a converted value of 50 m / s². 2 The upper limit of the second acceleration is determined by the ratio of the vacuum suction force of the nozzle to the chip mass. The system divides the vacuum suction force of 0.03N by the chip mass of 0.001kg. According to Newton's second law, the maximum permissible centrifugal acceleration under this suction force without the chip falling off and sliding is 30m / s². 2 The system compares the first and second acceleration limits, and takes the smaller value to determine the acceleration threshold as 30 m / s². 2 .

[0060] Finally, the jerk threshold is obtained. The upper limit of jerk (i.e., the rate of change of acceleration) is limited by the electrical system's response to sudden current changes and the mechanical structure's resistance to resonance. The first jerk upper limit is based on the driver's electrical response characteristics and is obtained by multiplying the current loop bandwidth by the determined acceleration threshold. The system multiplies 800Hz by 30m / s². 2 The upper limit of the first jerk is thus determined to be 24000 m / s². 3 The second jerk limit is based on the mechanical anti-shake characteristics and is obtained by multiplying the square of the mechanical structure's resonant frequency by a preset coefficient. The system squares 60Hz to get 3600, then multiplies it by the preset coefficient of 8, resulting in a second jerk limit of 28800 m / s². 3 .

[0061] The system compares these two upper limits and takes the smaller value, determining the jerk threshold to be 24000 m / s². 3 After complete logic processing, the system sets the velocity threshold to 1.5 m / s and the acceleration threshold to 30 m / s². 2 And the jerk threshold is 24000 m / s. 3 These parameters are combined to form the core constraint parameter set for this chip delivery mission.

[0062] 202. Motion Model Module: This module is used to establish a dimensionless motion model. The model uses the total transport time as the time normalization coefficient and the motion stroke as the displacement normalization coefficient. It defines the mapping relationship between normalized time variables, normalized position variables, normalized velocity variables, normalized acceleration variables, and normalized jerk variables, thus forming a definition scheme for the dimensionless motion state.

[0063] Specifically, the total transport time is used as the time normalization coefficient, and the motion distance is used as the displacement normalization coefficient. The normalized time variable is defined as the ratio of the actual time to the total transport time. The normalized position variable is defined as the difference between the actual displacement and the starting point of the motion divided by the motion distance. The normalized velocity variable is defined as the first derivative of the normalized position variable with respect to the normalized time variable. The normalized acceleration variable is defined as the first derivative of the normalized velocity variable with respect to the normalized time variable. The normalized jerk variable is defined as the first derivative of the normalized acceleration variable with respect to the normalized time variable. The above definitions are used as the definition scheme for dimensionless motion states. This definition scheme is used to construct a quadratic programming problem with the jerk constant within each segment as the decision variable.

[0064] Furthermore, when using the total transport time as the time normalization coefficient, the motion process is divided into acceleration, constant speed, and deceleration segments based on the motion stroke, speed threshold, and acceleration threshold. The acceleration and deceleration segments are made to have equal times, and both are set to the shortest time allowed by the acceleration threshold. Meanwhile, the time of the constant speed segment is made non-negative. The theoretical shortest total transport time is then used as the initial reference value for the time normalization coefficient. In subsequent processing, the theoretical shortest total transport time is used as the minimum feasible value for the candidate total transport time.

[0065] It should be noted that the system calculates the theoretically shortest total transport time to determine the physical limits under current hardware conditions. The control system extracts a motion travel of 0.15m, a speed threshold of 1.5m / s, and an acceleration threshold of 30m / s². 2 To minimize the time consumption, the system adopts the classic trapezoidal velocity programming model, which strictly divides the motion process into acceleration, constant velocity, and deceleration segments.

[0066] In the acceleration phase calculation, the system instructs the driver to continuously accelerate at full load corresponding to the acceleration threshold until the speed threshold is reached. The time required for the acceleration phase is defined as the speed threshold divided by the acceleration threshold, i.e., 1.5 m / s divided by 30 m / s. 2The acceleration phase's time limit is determined to be 0.05 s. During this period, the displacement of the acceleration phase is calculated as half the product of acceleration and the square of time, which is 0.0375 m. Since high-speed die bonders typically employ symmetrical trajectory planning to maintain mechanical force balance, the system ensures that the deceleration phase is completely symmetrical to the acceleration phase. Therefore, the deceleration phase also has a time of 0.05 s and a displacement of 0.0375 m. Summing the displacements of the acceleration and deceleration phases yields 0.075 m.

[0067] Subtracting this sum from the total travel distance of 0.15m yields a displacement of 0.075m for the constant-speed segment. This displacement is checked to be greater than zero, ensuring the existence of the constant-speed segment and that its time is non-negative. The running time of the constant-speed segment is obtained by dividing 0.075m by the speed threshold of 1.5m / s, resulting in 0.05s. The system adds the 0.05s for the acceleration segment, the constant-speed segment, and the deceleration segment to arrive at the theoretically shortest total transport time of 0.15s. This value represents the limit time achievable by the hardware under the current safety constraints and is set by the system as the minimum feasible value for the candidate total transport time.

[0068] Based on the above extreme value calculations, the system establishes normalization coefficients for the dimensionless motion model. The displacement normalization coefficient is directly taken as the absolute value of the current chip's transport distance, i.e., 0.15m. The time normalization coefficient is taken as the initial reference value of the theoretical shortest total transport time, i.e., 0.15s.

[0069] The system defines the dimensionless mapping relationships for each motion variable in the underlying algorithm module of the controller. The normalized time variable is defined as the ratio of the absolute time actually experienced by the current motion to the time normalization coefficient. This operation perfectly maps the motion cycle, originally in seconds, to a dimensionless pure numerical range of 0 to 1. The normalized position variable is defined as the difference between the current actual coordinate system displacement and the starting point of the motion, divided by the displacement normalization coefficient, similarly mapping the position to a dimensionless range of 0 to 1.

[0070] Based on the principles of calculus, we define the normalized velocity variable as the pure numerical first derivative of the normalized position with respect to the normalized time, the normalized acceleration as the first derivative of the normalized velocity with respect to the normalized time, and the normalized jerk as the first derivative of the normalized acceleration with respect to the normalized time.

[0071] 203. The optimization solution module is used to discretize the definition scheme of the dimensionless motion state at the cost of the weighted sum of the total transport time and the square integral of the jerk. It constructs a quadratic programming problem with the jerk constant in each segment as the decision variable. The quadratic programming problem is solved by scanning the candidate values ​​of the total transport time to obtain the optimal total transport time and the corresponding jerk constant sequence.

[0072] Specifically, the normalized time interval is divided into multiple sub-intervals, with the jerk remaining constant within each sub-interval. Using the jerk constant sequence of all sub-intervals as decision variables, the objective is to minimize the weighted sum of the total transport time and the square integral of the jerk, constrained by the principle that the velocity, acceleration, and jerk of the dimensionless motion state at discrete points do not exceed the dimensionless upper limit corresponding to the constraint parameter set. A quadratic programming problem is constructed. A minimum and maximum feasible value for the total transport time are pre-defined. Multiple candidate total transport times are selected between the minimum and maximum feasible values. The quadratic programming problem is solved for each candidate total transport time to obtain the corresponding jerk constant sequence and objective function value. The candidate total transport time with the smallest objective function value is taken as the optimal total transport time, and the corresponding jerk constant sequence is taken as the optimal jerk constant sequence. The optimal total transport time and the optimal jerk constant sequence are used to obtain the discrete dimensionless position sequence.

[0073] Furthermore, the selection method for multiple candidate total delivery times is as follows: the interval between the minimum feasible value and the maximum feasible value is divided equally with a fixed time step, and the value corresponding to each division point is used as a candidate total delivery time to form a candidate list; the fixed time step is set according to the servo cycle of the die bonder motion controller, so that each candidate total delivery time in the candidate list is an integer multiple of the servo cycle; the candidate list is used to solve a quadratic programming problem for each candidate total delivery time, and the candidate total delivery time with the minimum objective function value is determined as the optimal total delivery time.

[0074] It should be noted that the system uses the theoretically shortest total delivery time of 0.15s determined in step 202 as the minimum feasible value for the search interval. The control system reads the machine's process cycle requirement configuration file and sets the maximum feasible value for this operating condition to 0.20s. The control system obtains the fixed servo cycle of the die bonder's underlying motion controller as 0.001s. Using this servo cycle as the time step, the system equally samples the closed interval from 0.15s to 0.20s, obtaining a list of 51 discrete values ​​for the candidate total delivery time, including 0.150s, 0.151s, 0.152s up to 0.200s.

[0075] When discretizing the spatial state, the system divides the normalized time closed interval [0,1] into 100 uniform time subintervals, thus the total number of time subintervals N is fixed at 100, and the normalized time step of each subinterval is 0.01. The system stipulates that within these 100 independent subintervals, the dimensionless accelerometer constant cannot change abruptly and remains constant. These constants constitute the 100-dimensional decision variable sequence that the subsequent quadratic programming solver needs to optimize. During the algorithm's traversal and solution process, for a candidate total time in the list (currently iterated to T=0.156s), the physical threshold obtained in step 201 is strictly mapped to the dimensionless upper bound constraint under the current candidate time.

[0076] Using a physical velocity threshold of 1.5 m / s, a current candidate total time of 0.156 s, and a motion distance of 0.15 m, the dimensionless velocity upper limit at each discrete point is calculated to be 1.56 m / s. Using the square and cube of the candidate time, the dimensionless acceleration upper limit and dimensionless jerk upper limit boundary matrix are calculated according to the similarity law for the substituted time.

[0077] To balance production cycle efficiency with suppressing the flexible vibration energy of the mechanical structure, the system constructs a cost function with the dual objectives of minimizing the weighted sum of the total delivery time and the square integral of the jerk. The system employs a calculation formula that includes a dimensional calibration factor. ; In this formula, C represents the dimensionless total numerical cost of the output. T represents the candidate time substituted, in seconds. S represents the actual motion distance, 0.15m. N represents the total number of dimensionless subintervals, 100. Let be the dimensionless accelerometer constant decision variable for the k-th subinterval. The term within parentheses in the formula represents the time discrete integral of the square of the physical accelerometer, with dimensions m. 2 ·s -5 W T This is the time weighting coefficient, set to 100 in this example, with the unit defined as seconds. -1 To eliminate the dimension of time.

[0078] W J The accelerometer energy weighting coefficient is set to 1 × 10⁻⁶. -6 Its unit is defined as m -2 ·s 5 The system uses its built-in solver to solve a quadratic programming model with 100 decision variables for each of the 51 candidate time periods, obtaining 51 local minimum costs C. After scanning the data array, it is confirmed that the overall cost is minimal when T = 0.156s. The system then locks 0.156s as the optimal total delivery time and extracts the corresponding 100-dimensional dimensionless sequence as the optimal jerk constant sequence, writing it into the underlying register.

[0079] 204. The instruction acquisition module is used to perform integral recursion based on the jerk constant sequence to obtain a discrete dimensionless position sequence, and then obtain the actual time-position curve as the position motion instruction through inverse normalization mapping.

[0080] Specifically, based on the optimal jerk constant sequence, the normalized acceleration, normalized velocity, and normalized position at the end of each sub-interval are calculated sequentially using numerical integration to obtain a discrete dimensionless position sequence. Each value in the discrete dimensionless position sequence is multiplied by the motion stroke and the starting point is added to obtain the actual position sequence. The actual position sequence is combined with the corresponding time point calculated based on the optimal total transport time to form a time-position curve as a position motion command. The position motion command is used to drive the robotic arm and construct the feedforward torque.

[0081] Furthermore, the numerical integration employs the second-order Runge-Kutta method. Within each subinterval, the acceleration, velocity, and position of the starting point of the subinterval are first calculated. Then, based on the current subinterval jerk constant in the optimal jerk constant sequence, the acceleration and velocity of the intermediate point of the subinterval are calculated. Finally, the normalized acceleration, normalized velocity, and normalized position of the endpoint of the subinterval are updated based on the values ​​of the intermediate points. The above calculations are performed sequentially on all subintervals to obtain the normalized acceleration, normalized velocity, and normalized position at the endpoint of each subinterval, which serve as a discrete motion sequence. The discrete motion sequence is used to convert the motion sequence into an actual position sequence by multiplying it by the motion distance and adding the motion starting point.

[0082] It should be noted that the motion trajectory acquisition module of the control system completely extracts the 100-dimensional optimal accelerometer constant sequence output from the previous step from the read-only memory area, and reads the locked optimal total delivery time of 0.156s and the actual physical motion distance of 0.15m. The system sets the current stationary pickup point of the robotic arm as the starting point of motion in the absolute coordinate system, with its physical coordinates set to 0.000m. In the dimensionless model analysis at this stage, the normalized total time interval is uniformly divided into 100 segments, so the normalized time step size for each derivation is fixed at 0.01. The boundary initial conditions of the entire motion sequence are strictly zeroed out: that is, at the motion starting point corresponding to the zeroth time, the values ​​of normalized position, normalized velocity, and normalized acceleration are all initialized to 0.

[0083] The precise state of each node is calculated piecewise using a high-precision discrete kinematics analytical integral formula. Since the jerk is defined as a constant within the set time sub-interval, the state change of the kinematic system is a precise forward analytical recursive process, which can avoid the cumulative truncation error caused by numerical approximation algorithms. Taking the derivation process of the first time sub-interval corresponding to the trajectory starting point as an example: the control algorithm reads the optimal dimensionless jerk constant value of the first sub-interval as 120. The system multiplies this jerk constant (120) by the time step (0.01) and adds it to the initial acceleration (0) of the previous moment to obtain the normalized acceleration at the endpoint as 1.2.

[0084] Multiply the jerk constant (120) by the square of the time step and divide by 2 (resulting in 0.006), add the product of the initial acceleration and the time step, and add the initial velocity (0) to obtain the normalized velocity at the end of the first interval as 0.006.

[0085] Multiplying the jerk constant (120) by the cube of the time step and dividing by 6 (resulting in 0.00002), and then adding the initial velocity and time integral terms, the system obtains the substantive displacement of the endpoint normalized position of this sub-interval, with a value of 0.00002. The system then seamlessly replaces this set of endpoint values ​​with the starting state of the next sub-interval, extracts the constant values ​​of the second interval, and continues the deduction. This process is repeated 100 times to obtain a dimensionless position sequence consisting of 101 nodes, with the data smoothly increasing from 0 to 1.

[0086] After obtaining the absolutely precise dimensionless position sequence, the trajectory generator executes the denormalization physical mapping logic. The system extracts each of the 101 dimensionless position values ​​in the sequence, multiplies them by the motion distance of 0.15m, and superimposes the calculation result onto the starting point's 0.000m coordinates. The dimensionless value of 0.5 located at the midpoint of the sequence is converted into a physical coordinate of 0.075m. This completely unfolds the dimensionless sequence into actual physical position commands with a starting point of 0.000m and an ending point of 0.150m.

[0087] The 101 distribution points on the normalized time axis (uniformly incremented from 0.00 to 1.00) are each multiplied by the optimal total time of 0.156s, mapping them to a physical time axis starting from 0.0000s, generating a timestamp every 0.00156s, and ending at 0.1560s. The system binds the physical time series and physical position series with a bidirectional index in the data structure, synthesizing a high-resolution actual time-position curve table. This table serves as the basis for issuing position motion commands, directly directing the movement of the robotic arm while also providing extremely high-sampling-rate foundational data for the underlying feedforward torque construction.

[0088] 205. The collaborative control module is used to send position movement commands to the die bonder motion controller. At the same time, it constructs a feedforward torque based on the speed and acceleration corresponding to the position movement command, and iteratively learns and corrects the feedforward torque based on the position error of the previous delivery cycle to form an updated feedforward torque. Together with the feedback control, it drives the robotic arm to complete the automatic chip delivery.

[0089] Specifically, the position motion command is taken as input, and the velocity command value and acceleration command value at each moment are obtained by differentiating the position motion command command. Based on the velocity command value and acceleration command value, combined with the load inertia and viscous damping coefficient, the basic feedforward torque is calculated. The position error sequence recorded in the previous conveying cycle is obtained, and the position error sequence is multiplied by a preset learning gain to obtain the correction compensation amount. The correction compensation amount is added to the basic feedforward torque to form the updated feedforward torque. The updated feedforward torque is added to the feedback torque output by the feedback controller to obtain the total driving torque, which drives the robotic arm to automatically convey the chip according to the position motion command. The error sequence between the actual position and the position motion command in the current conveying cycle is recorded as a new position error sequence for iterative learning in the next conveying cycle.

[0090] It should be noted that the underlying drive controller receives the time-position curve, which includes 101 high-density spatiotemporal data points, output in step 204 in real time. Within the digital signal processor of the actuator, the algorithm performs high-precision differential calculations on the position commands of adjacent timestamps to obtain a real-time velocity command sequence, and then performs a second differential calculation on the velocity command sequence to obtain an acceleration command sequence.

[0091] The system's internal dynamic identification module pre-sets two core physical parameters of the mechanical transmission system referred to the motor shaft end: the equivalent load inertia coefficient is set to 0.05 N·m / (m / s²). 2 The equivalent system viscous damping coefficient is set to 0.1 N·m / (m / s). Within each microsecond-level high-frequency control cycle at the bottom layer, the system extracts the current instantaneous acceleration command value and multiplies it by the load inertia coefficient to obtain the torque component required to overcome system inertia; it also extracts the velocity command value and multiplies it by the viscous damping coefficient to obtain the torque component required to overcome frictional damping. The system algebraically sums these two components to calculate the basic feedforward torque, independent of any error feedback.

[0092] The system extracts the absolute position error data based on the same timestamp sequence from the previous chip delivery cycle recorded in the high-speed storage area. The system sets the preset learning gain coefficient for this machine condition to 500 N·m / m. The system directly multiplies the retrieved position error value from the previous cycle with this learning gain to obtain the correction compensation amount required to eliminate the residual, and immediately adds it to the basic feedforward torque to output the dynamically updated feedforward torque.

[0093] To demonstrate this dynamic compensation process, three typical high-frequency sampling points were extracted from the acceleration, constant speed, and deceleration phases of the transport cycle. The relevant underlying dynamic calculation data are detailed in Table 2 below: Table 2 acceleration phase 0.0312 1.2 25.0 1.37 0.0002 0.10 1.47 constant speed segment 0.0780 1.5 0.0 0.15 -0.0001 -0.05 0.10 Deceleration section 0.1248 1.2 -25.0 -1.13 0.00015 0.075 -1.055 At the sampling point in the acceleration phase, the inertial component ( ) and damping component ( The synthesized base torque is 1.37 N·m; combined with the 0.0002 m hysteresis error measured in the previous cycle, a compensation of 0.10 N·m is generated through gain amplification, and the updated feedforward torque is issued as 1.47 N·m. After obtaining the updated feedforward torque, the system injects it into the servo drive loop, and combines it with the feedback torque generated by the position closed-loop PID module to form the total motor drive torque.

[0094] During the forward reciprocating motion, the system continuously acquires the actual physical position using a high-resolution grating ruler, subtracting the theoretical command position from it in real time to obtain a new position error sequence. Once the current delivery cycle ends, this new error array will overwrite the old data, serving as an iterative learning benchmark for the next chip delivery cycle.

[0095] 206. Error initialization module, used to perform no-load test run using a weight of the same mass as the chip to be transported, record the deviation between the actual position and the theoretical position during the no-load test run, convert the deviation into a feedforward torque correction sequence, and store it in memory as an initial feedforward correction table; in subsequent transport, the initial feedforward correction table is used as the position error sequence of the first transport cycle for iterative learning and correction of the feedforward torque.

[0096] Furthermore, during the no-load trial run, the system operates at multiple different constant speed segments, recording the deviation between the actual position and the theoretical position within each constant speed segment. Based on the speed value and deviation of each constant speed segment, the corresponding feedforward torque correction value is calculated for each speed segment. The speed value and the corresponding feedforward torque correction value are associated to form a correction lookup table. The correction lookup table is stored in memory as the initial feedforward correction table. In subsequent transport, the corresponding correction value is retrieved from the correction lookup table based on the current speed command value. This correction value is then superimposed with the correction compensation amount to form the updated feedforward torque.

[0097] It should be noted that before the actual production line begins batch chip delivery, operators install a precisely weighed miniature test block at the die bonder's suction nozzle. This standard weight, set to 0.001 kg, is used to completely and equivalently simulate the load mass characteristics of the actual chips to be delivered, ensuring complete consistency of the system's dynamic model. The system's main control then enters a specially calibrated no-load test run mode.

[0098] During the trial operation command sequence, the underlying control system drove the robotic arm to perform a long-stroke reciprocating sliding test under load. To comprehensively map the system's steady-state tracking performance under different excitation states, the system incorporates a multi-segment constant speed test spectrum covering the entire range of actual working conditions. Based on the maximum permissible speed threshold of 1.5 m / s for the die bonder, the test spectrum rigorously divides it into five independent steady-state speed steps, namely 0.3 m / s, 0.6 m / s, 0.9 m / s, 1.2 m / s, and 1.5 m / s.

[0099] When the robotic arm accelerates and maintains a stable position within each specific constant speed range, the high-frequency data acquisition module is activated. The system captures the actual physical position fed back by the photoelectric encoder at the motor tail in real time, compares it with the theoretical spatiotemporal trajectory given by the servo generator, and after removing high-frequency vibration interference through a low-pass digital filter, extracts the inherent steady-state position hysteresis deviation value of the system under that speed step.

[0100] Based on the iterative learning gain constant (500 N·m / m) pre-calibrated in step 205, the system directly multiplies the captured position deviations of each segment by this fixed gain, and hard-converts them into compensation torque values ​​readable by the feedforward control network. The system pairs the speed command tags with the calculated feedforward correction values ​​and burns an initial feedforward torque correction lookup table into the non-volatile chip. The specific calibration mapping data for each step is detailed in Table 3 below: Table 3 0.3 0.00004 0.020 0.6 0.00009 0.045 0.9 0.00015 0.075 1.2 0.00022 0.110 1.5 0.00030 0.150 After this baseline data table is established, when the die bonder finishes calibration and enters formal operation, and begins to pick up the first real chip for the first cycle of delivery, the system automatically uses this initial table as the missing first cycle historical error library.

[0101] When the servo executes the initial cycle command, the controller no longer blindly relies on the proportional-integral module for bias correction. Instead, based on the speed command value issued at the current instant, it quickly extracts the torque correction reference value for the corresponding state from the table above using a one-dimensional linear interpolation algorithm. After the first piece is produced and continuous batch operations begin, the dual compensation logic of the control system is activated and runs continuously.

[0102] On one hand, the system continues to retrieve the inherent steady-state torque correction value from the static lookup table based on the current instantaneous velocity. On the other hand, it combines the output of the dynamic compensation amount calculated based on the residual of the previous cycle by the dynamic iterative learning network. These two correction data are merged and superimposed at the bottom layer to form a more accurate updated feedforward driving torque that is sent to the power amplifier.

[0103] Figure 4This is a schematic diagram of an automatic chip delivery device for a die bonder provided in an embodiment of the present invention. The automatic chip delivery device 300 may include a processor 301 and a memory 302. The memory 302 is used to store program instructions and / or data, and the processor 301 is used to execute the program instructions stored in the memory 302, thereby implementing the method described in the above-described method embodiment.

[0104] Optionally, the memory 302 and the processor 301 are coupled. The coupling is an indirect coupling or communication connection between devices, units, or modules, and can be electrical, mechanical, or other forms, for information interaction between devices, units, or modules.

[0105] Optionally, the die bonder chip automatic delivery device 300 may further include a communication interface 303, which is used to communicate with other devices via a transmission medium, such as transmitting received signals from other communication devices to the processor 301, or transmitting signals from the processor 301 to other communication devices. The communication interface 303 may be a transceiver or an interface circuit, such as a transceiver circuit or a transceiver chip.

[0106] This application embodiment does not limit the specific connection medium between the processor 301, memory 302, and communication interface 303. This application embodiment... Figure 4 The processor 301, memory 302, and communication interface 303 are connected via a bus 304. Figure 4 The connections between other components are shown in bold and are for illustrative purposes only, not as limiting information. The bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0107] The present invention also provides an automatic die bonder chip delivery device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the automatic die bonder chip delivery system in the above embodiments.

[0108] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the die bonder chip automatic delivery system.

[0109] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0110] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automatic chip delivery system for a die bonder, used for high-speed, non-slip transfer of bare semiconductor chips from the self-fetching position to the die bonder, characterized in that, include: The chip physical transfer mechanism has a nozzle for holding a bare semiconductor chip by vacuum adsorption at the end face, and a servo drive actuator for driving the nozzle to carry the chip and perform reciprocating transfer motion between the chip pick-up position and the chip bonding position. The die bonder main controller is communicatively connected to the chip physical transfer mechanism to perform real-time control of the chip handling path and the power of the transfer equipment. The die bonder main controller is internally configured with: The constraint parameter module is used to calculate the speed threshold, acceleration threshold, and jerk threshold to prevent the bare chip from slipping or falling off during high-speed displacement, based on the physical constraints and conveying task requirements of the system in material handling, forming a constraint parameter set. The conveying task requirements include the motion stroke. The motion model module is used to establish a dimensionless motion model based on the conveying time to be optimized and the motion stroke to define the discretized dimensionless motion state. The optimization solution module is used to solve the optimization problem under the constraints of the parameter set, taking into account the cost function of the transport time and the dimensionless accelerometer as the optimization objective, in order to achieve the fastest transfer cycle and the minimum mechanical vibration impact. The optimal transport time and the corresponding optimal dimensionless accelerometer sequence are obtained. The instruction acquisition module is used to calculate the discrete dimensionless position sequence based on the optimal dimensionless accelerometer sequence and the optimal transport time, and obtain the chip handling position movement instruction through inverse normalization mapping. The cooperative control module is used to obtain the basic feedforward control quantity to overcome mechanical resistance based on the position movement instruction, and correct the basic feedforward control quantity based on the historical position error to form an updated feedforward control quantity. The updated feedforward control quantity and the feedback control quantity are used to drive the actuator of the die bonder so that the nozzle accurately reaches the die bond position and completes the stable chip transport.

2. The die bonder chip automatic delivery system according to claim 1, characterized in that, The physical constraints of the system include the rated speed of the servo driver, the peak torque of the motor, the load inertia, the vacuum suction force of the nozzle, the current loop bandwidth, and the resonant frequency of the mechanical structure. The delivery task requirements also include the current chip quality. The constraint parameter module is specifically used for: The rated rotational speed of the servo driver is obtained as the speed threshold; A first acceleration upper limit is determined based on the peak torque of the motor and the load inertia, and a second acceleration upper limit is determined based on the vacuum suction force of the suction nozzle and the current chip mass. The smaller value between the first acceleration upper limit and the second acceleration upper limit is taken as the acceleration threshold. A first jerk upper limit is determined based on the current loop bandwidth and the acceleration threshold, and a second jerk upper limit is determined based on the mechanical structure resonant frequency. The smaller value between the first jerk upper limit and the second jerk upper limit is taken as the jerk threshold.

3. The automatic chip delivery system for die bonders according to claim 1, characterized in that, The discretized dimensionless motion state includes normalized time variables, normalized position variables, normalized velocity variables, normalized acceleration variables, and normalized jerk variables; the motion model module is specifically used for: The conveying time is used as the time normalization coefficient, and the motion stroke is used as the displacement normalization coefficient. The normalized time variable is defined as the ratio of the actual time to the time normalization coefficient, and the normalized position variable is defined as the difference between the actual displacement of the chip and the preset motion starting position divided by the displacement normalization coefficient. The normalized position variable is defined sequentially with respect to the normalized time variable to obtain the normalized velocity variable, the normalized acceleration variable, and the normalized jerk variable, respectively.

4. The die bonder chip automatic delivery system according to claim 1, characterized in that, The optimization solution module is specifically used for: The time interval corresponding to the delivery time is divided into multiple sub-intervals, and a quadratic programming problem is constructed with the dimensionless accelerometer constant in each sub-interval as the decision variable, which is the optimization problem. Within a preset time range, multiple candidate delivery times are selected. Under the constraints of the set of constraint parameters, the quadratic programming problem is solved for each candidate delivery time to obtain the corresponding cost function value. The candidate transport time with the minimum cost function value is taken as the optimal transport time, and the corresponding combination of dimensionless accelerometer constants is taken as the optimal dimensionless accelerometer sequence.

5. The die bonder chip automatic delivery system according to claim 4, characterized in that, When the optimization solution module selects multiple candidate delivery times within a preset time range, it is specifically used for: The theoretical shortest delivery time is calculated based on the motion stroke, the speed threshold, and the acceleration threshold, and is taken as the minimum feasible value. The interval between the minimum feasible value and the preset maximum feasible value is divided into equal parts with a fixed time step to obtain multiple candidate delivery times, wherein the fixed time step is an integer multiple of the servo cycle of the actuator.

6. The die bonder chip automatic delivery system according to claim 4, characterized in that, The cost function is calculated using a formula that includes a dimensional calibration factor: ； in, The cost function value; The candidate delivery time to be substituted; For exercise journey; This represents the total number of sub-intervals; Let be the dimensionless accelerometer constant for the k-th subinterval. Square it; This is a time-weighted coefficient, which has the dimension of the reciprocal of time to eliminate the dimension of time. This is the jerk energy weighting coefficient.

7. The die bonder chip automatic delivery system according to claim 4, characterized in that, The instruction-obtaining module is specifically used for: Based on the optimal dimensionless acceleration sequence, the normalized acceleration, normalized velocity, and normalized position at the endpoint of each sub-interval are calculated one by one using the numerical integration method to obtain the discrete dimensionless position sequence. Multiply the values ​​in the discrete dimensionless position sequence by the motion distance and combine them with the preset motion starting position to convert them into an actual position sequence; The actual position sequence is combined with the actual time point deduced from the optimal delivery time to form a time-position curve as the position movement command.

8. The die bonder chip automatic delivery system according to claim 1, characterized in that, When the cooperative control module corrects the basic feedforward control quantity based on historical position errors to form an updated feedforward control quantity, it is specifically used for: The position motion command is differentiated to obtain the corresponding velocity command value and acceleration command value; Based on the speed command value, the acceleration command value, and the system load parameters, the basic feedforward torque is calculated as the basic feedforward control quantity; The historical position error is multiplied by a preset learning gain to obtain the correction compensation amount; The corrected compensation amount is superimposed on the basic feedforward torque to form an updated feedforward torque, which is then used as the updated feedforward control amount.

9. The die bonder chip automatic delivery system according to claim 1, characterized in that, It also includes an error initialization module, used for: Before the first chip delivery, an unloaded test run was performed, and the initial deviation sequence between the actual position and the theoretical position was recorded during the unloaded test run. The initial deviation sequence is used as the historical position error during the first transport; and at the end of each subsequent transport cycle, the deviation between the actual position of the current cycle and the theoretical position corresponding to the position movement command is updated as the historical position error for the next transport cycle.

10. The die bonder chip automatic delivery system according to claim 9, characterized in that, When recording the initial deviation sequence between the actual position and the theoretical position during the no-load trial run, the error initialization module is specifically used for: During the no-load test run, multiple different constant speed segment operation operations were performed, and the initial deviation within each constant speed segment was recorded; The feedforward torque correction value is calculated based on the speed value of each constant speed segment and the corresponding initial deviation. A correction lookup table is constructed that includes the mapping relationship between the speed value and the feedforward torque correction value. The initial deviation sequence is then output based on the correction lookup table.

Images