Wire Bond Shear Force Prediction Using Bonder Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Destructive wire bond pull and shear tests are costly and difficult to perform accurately at close wire distances, often missing early detection of production defects, leading to increased internal defect costs.
Innovation Solution
A method using bonder data and machine learning models, particularly XGBoost Regressor, to predict shear force values in real-time, enabling earlier defect detection and reducing the need for destructive testing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If destructive pull and shear tests are used to determine wire bond strength, then reliable measurement of wire strength can be obtained, but the analysis cost increases and neighbouring wires may be damaged altering measurement results
Solution Approach 1:
The patent creates optical copies (images) of the wire bond structure using a camera system. These images are processed to extract geometric parameters that serve as proxies for mechanical strength properties, eliminating the need for physical destructive testing that damages surrounding wires.
Solution Approach 2:
The patent replaces the mechanical destructive testing system (pull and shear tests) with an optical measurement system. By using image processing and machine learning to predict strength from geometric parameters, the mechanical testing apparatus that causes damage to neighbouring wires is substituted with a non-contact optical method.
2Reliability
If destructive tests are performed to ensure wire bond quality, then production defects can be detected, but the production cost and time increase due to later stage corrections
Solution Approach 1:
The patent performs quality assessment immediately after wire bonding by capturing images and predicting strength values before the bonding process is complete. This preliminary detection allows for real-time quality control and immediate corrective action, rather than waiting for later production stages where defects are harder to correct.
Solution Approach 2:
The patent implements a feedback loop where image-based strength predictions are fed back into the production process. This enables real-time monitoring and adjustment of bonding parameters, allowing defects to be detected and corrected during production rather than after completion, reducing both time and cost losses.
3Reliability
If bonder data is collected and analyzed in real-time using machine learning, then earlier defect detection is enabled, but the device complexity and computational requirements increase
Solution Approach 1:
The patent makes the bonder's imaging and data collection system multi-functional. The same camera and data acquisition hardware used for basic process monitoring are also used for quality assessment and strength prediction, eliminating the need for separate dedicated testing equipment and reducing overall system complexity.
Solution Approach 2:
The patent enables the bonder system to perform quality assessment independently using its own built-in camera and data collection capabilities. The machine learning model processes data generated during normal bonding operations, allowing the system to self-evaluate bond quality without requiring external testing equipment or additional complex infrastructure.
Data Source
Figure 1
AI summary
The invention relates to a method (100) for predicting shear force values in a wire bonding process, comprising the following steps: - Providing (101) bonder data, the bonder data comprising values for at least one process parameter that are acquired during the wire bonding process by at least one sensor (2), - Predicting (102) the shear force values in the wire bonding process based on the provided bonder data using a machine learning model (1). Furthermore, the invention relates to a computer program, an apparatus, and a storage medium for this purpose.