TL;DR
This paper introduces "statistical metrology" as a systematic approach to understand and control variation in semiconductor manufacturing, specifically focusing on interlevel dielectric (ILD) thickness. By using electrical measurements from specialized test structures and coupling them with TCAD simulations, the authors successfully identified key factors influencing ILD thickness variation in a BPSG reflow process.
What Problem Does This Paper Address?
As semiconductor devices shrink and integration levels increase, controlling variation in manufacturing processes becomes critical for performance and yield. Interlevel dielectric (ILD) thickness is a key parameter that directly impacts parasitic capacitance between interconnect layers, affecting signal integrity and overall circuit performance. Traditional methods often focus on end-of-line measurements, which can be time-consuming and don't provide insights into early process stages. This paper addresses the need for a more proactive and systematic methodology to quantify and identify the sources of ILD thickness variation during the manufacturing process itself, enabling earlier intervention and improvement.
Methods
The authors employ a methodology they term "statistical metrology," which combines several key elements:
- Electrical Measurements: Instead of relying solely on optical measurements, they use electrical measurements from short-flow process test structures. This allows for the acquisition of a large volume of data, including intradie variation.
- Statistical Design of Experiments (DOE): They utilize a two-step DOE strategy. The first step is a screening experiment to identify significant layout and process factors affecting ILD thickness. The second step, planned for future work, aims to quantify and model these contributions more accurately.
- Technology Computer-Aided Design (TCAD) Coupling: Electrical measurements are coupled with 2D capacitance simulations (using Raphael) to convert capacitance data into ILD thickness. This allows for the estimation of ILD thickness even when direct physical measurement is difficult or time-consuming.
- Specialized Test Structures: They designed capacitor test structures with varying layout factors (line width, spacing, finger length, number of fingers, rotation, and interaction ring presence). To account for variations in the underlying polysilicon line width, they also incorporated resistive structures (van der Pauw and 4-point probe) to electrically measure line width.
Datasets
The study utilizes data collected from test structures fabricated on wafers at the MIT process facility. The specific materials system involved polysilicon and BPSG dielectric reflow. The experimental design involved 32 unique combinations of layout factors, with each test structure replicated three times and randomized within a 1 cm x 1 cm die. Measurements were taken from 52 die on each of 12 wafers, with two different nominal BPSG deposition thicknesses (5000Å and 6000Å).
Key Insights and Ideas
- "Statistical Metrology" as a Framework: The paper formalizes a methodology that integrates electrical measurements, DOE, and TCAD for variation analysis, moving beyond traditional approaches.
- Electrical Measurements for Process Control: Demonstrates the power of electrical measurements in short-flow processes for early detection and understanding of variation, which is more efficient than end-of-line testing.
- ILD Thickness is Highly Sensitive to Layout: The study highlights that the physical layout of the underlying conductors (specifically finger line width, spacing, and length) has a significant impact on the resulting ILD thickness.
- Line Width Variation is a Critical Confounder: The authors recognized that variations in the underlying polysilicon line width directly affect capacitance measurements, necessitating a method to independently measure and account for this variation.
- Spatial Variation Matters: The analysis revealed that die position within the wafer, particularly the y-coordinate, contributes significantly to the observed ILD thickness variation.
Key Results
- The statistical metrology methodology was successful in highlighting key factors influencing ILD thickness variation.
- Exploratory data analysis indicated that finger line width had the largest impact on ILD thickness, followed by line spacing, finger length, and number of fingers.
- Histograms and scatter plots showed a bimodal distribution of ILD thickness for certain wafers, suggesting the influence of process splits or systematic variations.
- Analysis of variance (ANOVA) confirmed that line width and capacitor y-coordinate were significant contributors to the variance in ILD thickness.
Limitations and Future Work
- Limited Scope of Planarization: This study focused on BPSG reflow planarization. The authors acknowledge that other planarization techniques, such as chemical mechanical polishing (CMP), may have different dominant variation sources (e.g., area/density concerns).
- Line Width Metrology Under Development: While they accounted for line width variation, the statistical metrology for lithography and etch processes themselves had not yet been fully developed for the specific equipment used.
- Future Work:
- Conducting second-pass experiments to more accurately quantify and model the relationships between factors and ILD thickness variation.
- Investigating interactions between ILD thickness metrology and line width variation metrology.
- Applying additional statistical approaches, such as filtering techniques.
- Adapting the ILD statistical metrology approach for alternative planarization processes, particularly CMP, with modified test structures.
Summary updated