Laser-based Thickness Control in a Double-Side Polishing System for Silicon Wafers

Liang Zhu, Biao Mei, Weidong Zhu, Wei Li2020 · DOI

TL;DR

This paper presents a novel double-side polishing (DSP) system for silicon wafers that uses a laser probe for real-time thickness measurement and a B-spline model for accurate thickness characterization. This approach enables precise feedback control, significantly improving wafer thickness uniformity, which is crucial for advanced semiconductor manufacturing.

What Problem Does This Paper Address?

The semiconductor industry's increasing demand for larger and thinner silicon wafers for high-density integrated circuits necessitates stringent control over wafer thickness and uniformity. Traditional polishing methods often struggle to achieve the required precision, leading to variations that can impact device performance and yield. This paper addresses the challenge of achieving highly accurate and consistent wafer thickness control in automated double-side polishing (DSP) systems.

Methods

The core of this work is a double-side polishing system integrated with a laser-based thickness control (LTC) system. The LTC system utilizes a laser probe to scan the wafer's thickness in real-time. To handle the potentially noisy scan data and accurately represent the wafer's thickness profile, the authors employ a B-spline model. This model offers flexibility in capturing complex surface shapes.

Key methodological steps include:

  • Mechanical Design: A detailed description of the DSP system's mechanical structure, including the polishing unit with its rotating plates and carriers for wafers.
  • Laser-Based Thickness Measurement: An optical coherence tomography (OCT) based laser probe is used to measure the time difference of back-reflected signals from the wafer's upper and lower surfaces, allowing for thickness calculation.
  • Precise Probe Location: A crucial aspect is accurately determining the laser probe's position relative to the wafer. This is achieved by calculating the physical coordinates of carriers, wafers, and the probe using pulse information from the polishing equipment and timestamps.
  • B-spline Modeling: The raw thickness data is fitted using a B-spline model. This approach is chosen over traditional polynomial fitting for its superior geometric modeling capabilities. The paper discusses the selection of the degree of the B-spline basis functions (chosen as 3 for C2 continuity) and, importantly, the optimal number of control points.
  • Feedback Control: The thickness deviation, calculated from the B-spline model, is used to provide feedback to the DSP system for online machining compensation, aiming to achieve the desired thickness.

Datasets

The paper does not explicitly mention pre-existing datasets. Instead, it relies on experimental data generated from their custom-built double-side polishing system. The "dataset" consists of raw thickness points obtained from scanning silicon wafers (specifically, "P-" type wafers) along 0° and 90° measuring paths during the polishing process.

Key Insights and Ideas

  • B-spline for Thickness Characterization: The B-spline model proves to be a robust tool for accurately characterizing wafer thickness variations, outperforming biquadratic polynomial fitting, especially in handling noisy data and complex profiles.
  • Importance of Control Point Selection: The number of control points in the B-spline fitting significantly impacts the accuracy of the thickness model. Too few points lead to underfitting and inability to capture variations, while too many can lead to overfitting and fitting random errors. The study identifies an optimal range (around 25 control points) for achieving good fitting performance.
  • Real-time Feedback is Key: The integration of real-time laser scanning, precise probe localization, and B-spline modeling enables effective feedback control, allowing the DSP system to dynamically adjust its polishing path to achieve target thickness.
  • Precise Probe Localization is Non-Trivial: Accurately mapping the laser probe's position to the wafer's surface is critical for generating valid thickness data. The paper details a method using rotational encoder data and timestamps to achieve this.

Key Results

  • The B-spline fitting with an optimized number of control points (25) effectively characterized the thickness variations of Si wafers, with residual errors approximating a zero-mean normal distribution.
  • The fitted curves, along with upper and lower bounds representing ±3σ, indicated that 99.73% of scanning points fell within the thickness range.
  • Experiments demonstrated that the thickness variation could be effectively controlled using the presented DSP system with B-spline-based thickness evaluation.
  • The paper suggests that after implementing thickness control, the root mean square errors are significantly reduced, fulfilling the quality requirements for 300mm diameter Si wafers.

Limitations and Future Work

  • Limitations:
    • The paper focuses on a specific type of Si wafer ("P-" type). The applicability to other wafer types or materials might require further validation.
  • While the B-spline model is shown to be effective, the paper doesn't extensively explore alternative advanced fitting techniques or their comparative performance.
  • The study primarily focuses on thickness control. Other critical aspects of wafer polishing, such as surface roughness and flatness, are mentioned as goals but not deeply analyzed in the context of this specific control method.
  • Future Work:
    • The authors imply that the presented system and method can be used to further polish Si wafers to achieve even higher quality.
  • The feedback control mechanism, once the thickness deviation is calculated, can be used for online machining path compensation, suggesting further refinement of the control loop.
  • The paper's findings on optimal control point selection for B-spline fitting could be extended to other applications requiring precise surface characterization.

Summary updated

BibTeX
@misc{zhu2020laser,
  author = {Liang Zhu and Biao Mei and Weidong Zhu and Wei Li},
  title = {{Laser-based Thickness Control in a Double-Side Polishing System for Silicon Wafers}},
  year = {2020},
  doi = {10.3390/s20061603},
}
Figure