Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Non-Destructive Detection of Sub-Micron Imperceptible Scratches On Laser Chips Based On Consistent Texture Entropy Recursive Optimization Semi-Supervised Network

Pan Liu

TL;DR

This work tackles non-destructive detection of sub-micron laser-chip scratches under severe low-contrast and data-scarce conditions. It introduces TexRecNet, a diffusion-inspired, recursive optimization network that refines segmentation by leveraging prior high-confidence predictions and a time-aware encoder–decoder, combined with a texture-entropy–based semi-supervised training strategy. The approach yields a supervised loss L_l and an unsupervised loss L_u, with supervisory signals for unlabeled data derived from consistent texture entropy across recursive prediction sequences, enabling end-to-end training from limited labels. Experiments on a laser chip scratch dataset show TexRecNet achieving foreground IoU up to about 75% and shallow scratch recall improvements up to roughly 33% over Unet, with robust performance across data partitions, demonstrating practical impact for improving yield and reducing costs in chip production. The method offers a principled way to exploit unlabeled data and edge-texture cues to detect imperceptible defects, enhancing non-destructive quality control in semiconductor manufacturing.

Abstract

Laser chips, the core components of semiconductor lasers, are extensively utilized in various industries, showing great potential for future application. Smoothness emitting surfaces are crucial in chip production, as even imperceptible scratches can significantly degrade performance and lifespan, thus impeding production efficiency and yield. Therefore, non-destructively detecting these imperceptible scratches on the emitting surfaces is essential for enhancing yield and reducing costs. These sub-micron level scratches, barely visible against the background, are extremely difficult to detect with conventional methods, compounded by a lack of labeled datasets. To address this challenge, this paper introduces TexRecNet, a consistent texture entropy recursive optimization semi-supervised network. The network, based on a recursive optimization architecture, iteratively improves the detection accuracy of imperceptible scratch edges, using outputs from previous cycles to inform subsequent inputs and guide the network's positional encoding. It also introduces image texture entropy, utilizing a substantial amount of unlabeled data to expand the training set while maintaining training signal reliability. Ultimately, by analyzing the inconsistency of the network output sequences obtained during the recursive process, a semi-supervised training strategy with recursive consistency constraints is proposed, using outputs from the recursive process for non-destructive signal augmentation and consistently optimizes the loss function for efficient end-to-end training. Experimental results show that this method, utilizing a substantial amount of unsupervised data, achieves 75.6% accuracy and 74.8% recall in detecting imperceptible scratches, an 8.5% and 33.6% improvement over conventional Unet, enhancing quality control in laser chips.

Non-Destructive Detection of Sub-Micron Imperceptible Scratches On Laser Chips Based On Consistent Texture Entropy Recursive Optimization Semi-Supervised Network

TL;DR

This work tackles non-destructive detection of sub-micron laser-chip scratches under severe low-contrast and data-scarce conditions. It introduces TexRecNet, a diffusion-inspired, recursive optimization network that refines segmentation by leveraging prior high-confidence predictions and a time-aware encoder–decoder, combined with a texture-entropy–based semi-supervised training strategy. The approach yields a supervised loss L_l and an unsupervised loss L_u, with supervisory signals for unlabeled data derived from consistent texture entropy across recursive prediction sequences, enabling end-to-end training from limited labels. Experiments on a laser chip scratch dataset show TexRecNet achieving foreground IoU up to about 75% and shallow scratch recall improvements up to roughly 33% over Unet, with robust performance across data partitions, demonstrating practical impact for improving yield and reducing costs in chip production. The method offers a principled way to exploit unlabeled data and edge-texture cues to detect imperceptible defects, enhancing non-destructive quality control in semiconductor manufacturing.

Abstract

Laser chips, the core components of semiconductor lasers, are extensively utilized in various industries, showing great potential for future application. Smoothness emitting surfaces are crucial in chip production, as even imperceptible scratches can significantly degrade performance and lifespan, thus impeding production efficiency and yield. Therefore, non-destructively detecting these imperceptible scratches on the emitting surfaces is essential for enhancing yield and reducing costs. These sub-micron level scratches, barely visible against the background, are extremely difficult to detect with conventional methods, compounded by a lack of labeled datasets. To address this challenge, this paper introduces TexRecNet, a consistent texture entropy recursive optimization semi-supervised network. The network, based on a recursive optimization architecture, iteratively improves the detection accuracy of imperceptible scratch edges, using outputs from previous cycles to inform subsequent inputs and guide the network's positional encoding. It also introduces image texture entropy, utilizing a substantial amount of unlabeled data to expand the training set while maintaining training signal reliability. Ultimately, by analyzing the inconsistency of the network output sequences obtained during the recursive process, a semi-supervised training strategy with recursive consistency constraints is proposed, using outputs from the recursive process for non-destructive signal augmentation and consistently optimizes the loss function for efficient end-to-end training. Experimental results show that this method, utilizing a substantial amount of unsupervised data, achieves 75.6% accuracy and 74.8% recall in detecting imperceptible scratches, an 8.5% and 33.6% improvement over conventional Unet, enhancing quality control in laser chips.

Paper Structure

This paper contains 8 sections, 15 equations, 10 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Methodology
  3. TexRecNet
  4. Network Training Process
  5. Experiment
  6. Comparison with existing alternative methods
  7. Ablation
  8. Results

Figures (10)

  • Figure 1: Laser Chip Manufacturing Site (a) Mechanical Calibration (b) Production scenario (c) Eutectic Bonding (d) Chip Retrieval
  • Figure 2: Schematic diagram of laser chip image and location of scratches (a) 3D structure of chip, (b) Chip general view, (c) Chip enlargement view, (d) deep scratches schematic diagram, (e) shallow scratches schematic diagram
  • Figure 3: Overview of TexRecNet pipeline for unlabeled images.The highly reliable prediction is obtained by the recursive optimized prediction process, and then the reliable supervision signal is extracted by texture entropy consistency.
  • Figure 4: TexRecNet Network Architecture. Images and historical predictions are input simultaneously for encoding and fused through attention structures.
  • Figure 5: Performance demonstration of different methods (a) Chip collection image (b) Unet prediction result (c) CutMix prediction result (d) MT prediction result (e) Prediction result of our method (f) Ground truth
  • ...and 5 more figures