Eyepoint Teaching Using Geometric Feature Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional wire bonding techniques face challenges in accurately aligning semiconductor devices due to variations in surface properties such as color and texture, leading to low scores and reduced productivity, as they rely on pattern matching methods that are sensitive to differences in reflectivity and appearance among devices from the same batch or different batches.
Innovation Solution
A method of teaching an eyepoint using a group of shapes selected from a semiconductor device region, where the locations of these shapes are defined relative to each other, and either a sample device or predetermined data is used to create a geometric feature-based pattern that is invariant to surface variations, allowing for robust alignment and processing regardless of device-specific differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional pattern matching methods (NGCS or edge-based) are used to teach eyepoints, then the system can identify device features, but the alignment accuracy deteriorates due to surface variations in color, texture, and reflectivity among devices
Solution Approach 1:
The patent extracts only the geometric shape information of features (bond pads, circuit elements) while deliberately excluding surface property information (color, texture, reflectivity). The vision system is taught to recognize features based solely on their geometric characteristics, which are invariant across devices. This extraction of essential geometric information while discarding variable surface properties resolves the contradiction between measurement precision and reliability.
Solution Approach 2:
The patent changes the parameters used for feature identification from surface properties (grayscale values, edge detection based on intensity changes) to geometric parameters (shape, size, position, orientation). By transforming the recognition criteria from appearance-based to geometry-based parameters, the system achieves consistent alignment accuracy across devices with varying surface properties.
2Productivity
If the vision system relies on visual properties for alignment, then it can detect device features, but productivity decreases due to frequent assists and re-teaching operations
Solution Approach 1:
The patent performs preliminary teaching of the vision system using geometric feature extraction from a single sample device. This preliminary action establishes a robust geometric model that can be applied to all subsequent devices without re-teaching. The system is pre-configured to recognize geometric patterns invariant to surface variations, eliminating the need for frequent assists and re-teaching operations, thereby improving productivity while maintaining alignment precision.
3Adaptability or versatility
If conventional teaching methods use surface appearance data, then the system can capture device features, but adaptability deteriorates when processing devices from different batches or suppliers
Solution Approach 1:
The patent creates a geometric abstraction (a simplified geometric representation) of the device features during the teaching phase. This geometric copy captures the essential shape, size, and positional relationships of features while eliminating batch-specific surface variations. The vision system then matches this geometric copy against actual devices, enabling adaptability across different batches and suppliers while filtering out irrelevant variability information.
Data Source
AI summary
A method of teaching an eyepoint for a wire bonding operation is provided. The method includes (1) selecting a group of shapes from a region of a semiconductor device for use as an eyepoint, and (2) teaching the eyepoint to a wire bonding machine using at least one of (a) a sample semiconductor device, or (b) predetermined data related to the semiconductor device. The teaching step includes defining locations of each of the shapes with respect to one another.