CN-121981068-A - Design scheme of bonding wire, electronic equipment, server and storage medium

CN121981068ACN 121981068 ACN121981068 ACN 121981068ACN-121981068-A

Abstract

The invention provides a bonding wire design scheme, electronic equipment, a server and a storage medium, wherein the scheme comprises the steps of collecting characteristic data of the bonding wire and a working environment thereof, carrying out numerical coding on the characteristic data, transmitting the characteristic data subjected to numerical coding into a design model to obtain a predicted value of the height and the wire diameter of the bonding wire, adjusting the design model by using an activation function with physical constraint by using a reliability weighting loss function, optimizing the design model by using multi-task learning until the design model meets process requirements, and outputting the bonding wire design scheme by using the adjusted and optimized design model. The invention has the advantages of accuracy, reliability, realization, high efficiency and sustainable optimization, provides a systematic, scientific and intelligent solving path for the bonding wire design, and is better than the traditional artificial experience method or the simple mathematical optimization method.

Inventors

LI CONGBIN
WANG DIE
ZHOU YU
LIANG JIE
Remy Alan Jiyeman

Assignees

赛晶亚太半导体科技(浙江)有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. The design scheme of bonding line, its characterized in that includes: collecting characteristic data of a bonding wire and a working environment thereof; Performing numerical coding on the characteristic data; transmitting the characteristic data after numerical coding into a design model to obtain the predicted value of the height and the wire diameter of the bonding wire, wherein the design model uses an activation function with physical constraint; Adjusting the design model by using a reliability weighted loss function, and optimizing the design model by using multi-task learning until the design model meets the process requirement; And outputting the bonding wire design scheme by using the adjusted and optimized design model.
2. The bonding wire design of claim 1, wherein: The characteristic data of the bonding wire comprise bonding wire materials, current bearing capacity, working frequency and packaging type, and the characteristic data of the working environment comprise thermal circulation and mechanical vibration.
3. The bonding wire design of claim 2, wherein the process of numerically encoding the feature data comprises: The method comprises the steps of mapping bonding wire materials and mechanical vibration into low-dimensional numerical vectors, mapping the bonding wire materials and the mechanical vibration into integer labels according to different current requirements for current bearing capacity, carrying out standardization processing on working frequency and thermal cycle, and carrying out condition judgment and binarization processing on packaging types.
4. The bonding wire design of claim 1, wherein the design model is built based on a regression task model; the number of nerve units of the model output layer is the same as the number of characteristic data, the model hiding layer consists of two full-connection layers, the inside of each full-connection layer is provided with an activation function of physical constraint, and the model output layer is provided with two output nodes respectively corresponding to the height and the wire diameter of the bonding wire.
5. The bonding wire design of claim 4, wherein the activation function with physical constraints is specifically: creating a monotonically increasing and continuously guidable function as an initial function; Adding a normalization constant to the initial function to limit the output of the initial function to a limited interval, thereby obtaining an activated function; The output intervals of the activation functions of the two full-connection layers correspond to physical constraints of the bonding wire height and the wire diameter respectively.
6. The bonding wire design scheme according to claim 1, wherein the reliability weighted loss function is constructed as follows: calculating a weight according to the reliability information of each sample, and combining the weight with a sample prediction error term in a loss function; dynamically adjusting error penalty strengths of different samples and tasks by learning the logarithmic variance of each output; a physical constraint penalty term is introduced in the loss calculation, and a smoothness penalty is applied to predicted values that exceed the allowable range of bond wire height or wire diameter.
7. The bonding wire design of claim 1, wherein the optimizing the design model using multitasking learning comprises: The method comprises the steps of constructing a shared feature extraction network, designing an independent output layer for each task, constructing a weighted total loss function, synthesizing the losses of a plurality of tasks, adding a physical constraint penalty term, and dynamically adjusting task weights through an optimizer training model to balance the importance of the tasks.
8. An electronic device comprising a processor and a memory communicatively connected to the processor for storing instructions executable by the processor, characterized in that the processor is adapted to execute the bonding wire design of any of the above claims 1-7.
9. A server comprising at least one processor and a memory communicatively coupled to the processor, the memory storing instructions executable by the at least one processor to cause the at least one processor to perform the bond wire design of any one of claims 1-7.
10. A computer-readable storage medium, in which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the bonding wire design of any one of claims 1-7.

Description

Design scheme of bonding wire, electronic equipment, server and storage medium Technical Field The invention belongs to the field of bonding wire design, and particularly relates to a bonding wire design scheme, electronic equipment, a server and a storage medium. Background At present, aluminum wire bonding is widely applied to manufacturing of low-power devices such as T0 series DFN series and IGBT power modules, and plays an indispensable key procedure and role, in the bonding process, the quality required by bonding is achieved through parameter control, besides, the loop height formed by the aluminum wires among objects is also a factor which has higher influence on bonding quality, in the bonding process, the wire arc (Looping) height is moderate, the wire is easy to collapse when the wire is subjected to a subsequent plastic packaging process due to the excessively high wire arc height, and if the wire arc height is excessively low, a large stress is easily formed inside a bonding point, so that the bonding point reliability is reduced; the prior art has the advantages that the neutral line arc height is defined as GAP of an aluminum line and a contact surface is larger than 2 times of the line diameter, but the positive and negative values of loop are +/-500 mu m, the highest loop is defined in the IGBT industry and cannot exceed the glue filling height, but in different objects, connecting surfaces and connecting processes, the requirements on the loop are required to be considered, and each object has different requirements on the thermal cycle and the loop path of the loop, so that the 2 times of the line diameter is not an absolute value. Disclosure of Invention The invention aims to provide a bonding wire design scheme, electronic equipment, a server and a storage medium, so as to solve at least one technical problem in the prior art. In order to solve the above technical problems, a first aspect of the present invention provides a design solution of a bonding wire, including: collecting characteristic data of a bonding wire and a working environment thereof; Performing numerical coding on the characteristic data; transmitting the characteristic data after numerical coding into a design model to obtain the predicted value of the height and the wire diameter of the bonding wire, wherein the design model uses an activation function with physical constraint; Adjusting the design model by using a reliability weighted loss function, and optimizing the design model by using multi-task learning until the design model meets the process requirement; And outputting the bonding wire design scheme by using the adjusted and optimized design model. Further, the characteristic data of the bonding wire comprise bonding wire materials, current bearing capacity, working frequency and packaging type, and the characteristic data of the working environment comprise thermal circulation and mechanical vibration. Further, the process of performing numerical coding on the characteristic data includes: The method comprises the steps of mapping bonding wire materials and mechanical vibration into low-dimensional numerical vectors, mapping the bonding wire materials and the mechanical vibration into integer labels according to different current requirements for current bearing capacity, carrying out standardization processing on working frequency and thermal cycle, and carrying out condition judgment and binarization processing on packaging types. Further, the design model is constructed based on a regression task model; the number of nerve units of the model output layer is the same as the number of characteristic data, the model hiding layer consists of two full-connection layers, the inside of each full-connection layer is provided with an activation function of physical constraint, and the model output layer is provided with two output nodes respectively corresponding to the height and the wire diameter of the bonding wire. Further, the activation function with physical constraint specifically includes: creating a monotonically increasing and continuously guidable function as an initial function; Adding a normalization constant to the initial function to limit the output of the initial function to a limited interval, thereby obtaining an activated function; The output intervals of the activation functions of the two full-connection layers correspond to physical constraints of the bonding wire height and the wire diameter respectively. Further, the construction process of the reliability weighted loss function is as follows: calculating a weight according to the reliability information of each sample, and combining the weight with a sample prediction error term in a loss function; dynamically adjusting error penalty strengths of different samples and tasks by learning the logarithmic variance of each output; a physical constraint penalty term is introduced in the loss calculation, and a smoothness penalty is applied to