Critical Dimension Metrology: Accuracy Limits and Solutions

Introduction#

Critical dimension (CD) metrology is the cornerstone of semiconductor process control, yet achieving sub-nanometer accuracy remains a formidable challenge. The measured CD is never the true dimension; it is a convolution of the physical structure, the measurement instrument, and the model used to interpret signals. Understanding and mitigating these errors is essential for yield improvement and device performance.

This guide examines the fundamental limits of CD metrology by comparing optical scatterometry and electron beam (e-beam) techniques. We then dissect uncertainty sources and present practical calibration and algorithm optimization strategies to push accuracy closer to the true dimension.

Fundamentals of CD Metrology#

CD refers to the width of a feature, such as a transistor gate or interconnect line. Two dominant approaches are used: optical scatterometry (also called optical critical dimension, OCD) and e-beam metrology, including critical dimension scanning electron microscopy (CD-SEM). Each relies on different physical interactions, leading to distinct error profiles.

Optical scatterometry measures light diffracted from a periodic grating. The intensity and polarization of reflected orders are compared to a theoretical model, typically rigorous coupled-wave analysis (RCWA). The CD is extracted by solving an inverse problem. E-beam metrology, by contrast, focuses a beam of electrons onto the sample and detects secondary or backscattered electrons. The edge position is determined from the signal profile, often using threshold or model-based algorithms.

Both methods are indirect: they measure a proxy signal and infer CD through a model. Therefore, accuracy depends on the fidelity of that model and the stability of the instrument. EJER, a supplier of precision environment control and dry storage equipment, emphasizes that stable environmental conditions are a prerequisite for reliable metrology, as humidity and oxidation can alter sample dimensions and signal response.

Uncertainty Sources in Optical Scatterometry#

Optical scatterometry uncertainty arises from instrument, model, and sample factors. Instrument-related errors include wavelength instability, polarization misalignment, and detector nonlinearity. These can be minimized through regular calibration with certified reference materials, such as silicon gratings with known dimensions.

Model errors are often dominant. The RCWA model assumes ideal geometry, homogeneous materials, and perfect gratings. Real structures exhibit line edge roughness, sidewall angle variation, and material intermixing. Discrepancies between the model and reality lead to systematic biases. To address this, one can incorporate additional parameters (e.g., sidewall angle, oxide thickness) into the model, but this increases parameter correlation and uncertainty.

Sample-related errors include film thickness variation, substrate effects, and environmental changes. For instance, moisture absorption can swell polymer films, shifting the effective CD. Using dry storage for samples and reference standards, such as EJER's dry cabinets that maintain humidity as low as 1%RH, helps stabilize the sample and reduce drift.

Uncertainty Sources in E-Beam Metrology#

E-beam metrology faces its own set of challenges. The electron beam interacts with the sample, causing charging, contamination, and beam-induced damage. Charging can deflect the beam and distort the signal, leading to edge detection errors. Contamination builds up under the beam, changing the effective beam diameter and reducing resolution.

Edge detection algorithms are another major source. The true edge is defined by the material boundary, but the secondary electron signal is a convolution of the beam profile and the edge. Threshold-based methods are sensitive to signal-to-noise ratio and can vary with beam current and landing energy. Model-based libraries can improve consistency but require accurate calibration.

Instrument calibration is critical. Magnification errors, stage drift, and beam current fluctuations must be corrected using standards traceable to national laboratories. Additionally, sample cleanliness is paramount; oxidation or moisture layers can alter the work function and secondary electron yield. Storing samples in a dry, nitrogen-purged environment, such as an EJER N2 cabinet with automatic nitrogen saving, prevents oxidation and preserves surface integrity.

Calibration Strategies for Improved Accuracy#

Calibration is the bridge between the measured signal and the true dimension. For optical scatterometry, a multi-step approach is recommended. First, calibrate the instrument using a certified reference grating with known pitch and CD. Second, refine the optical model using multiple samples with varying parameters to decouple correlated variables. Third, validate with an independent technique, such as CD-SEM or atomic force microscopy (AFM).

For e-beam metrology, daily calibration with a gold-on-carbon standard ensures magnification accuracy. Periodic checks with a known pitch grating correct for stage nonlinearity. To reduce charging, use low beam energy and a conductive coating, or operate in a low-vacuum environment. For charging-sensitive samples, a model-based edge detection algorithm that accounts for the signal shape can outperform simple thresholding.

Cross-technique calibration is powerful. By measuring the same sample with both scatterometry and CD-SEM, one can identify and correct systematic offsets. This hybrid approach leverages the high precision of scatterometry and the high spatial resolution of e-beam. It is particularly effective when combined with a reference metrology system, such as a critical dimension atomic force microscope (CD-AFM).

Algorithm Optimization for Model-Based Measurements#

Algorithm optimization focuses on reducing both random and systematic errors. For scatterometry, the inverse problem is often ill-posed. Regularization techniques, such as Tikhonov regularization, can stabilize the solution. Machine learning approaches, including neural networks, can speed up the inverse solution and improve robustness to noise, provided they are trained on a representative dataset.

For e-beam, advanced edge detection algorithms use a library of pre-simulated signal profiles. The measured signal is matched to the library to find the best-fit edge position. This method reduces sensitivity to noise and charging. Another approach is to use a physics-based model that includes the beam profile and sample interaction, solved iteratively.

Uncertainty quantification should be integrated into the algorithm. Monte Carlo simulations can propagate input uncertainties (e.g., beam current, model parameters) to the final CD. This provides a confidence interval, which is essential for process control. Additionally, real-time drift correction using a reference mark on the sample can compensate for stage drift and beam instability.

Environmental Control and Best Practices#

Environmental stability is often overlooked but is a key contributor to measurement uncertainty. Temperature fluctuations cause thermal expansion, altering CD by several nanometers per degree Celsius. Humidity affects both the sample and the instrument optics. For e-beam systems, humidity can lead to oxidation and charging.

Best practices include maintaining a clean, temperature-controlled room (e.g., 23±0.5°C) and using dry storage for samples and standards. EJER's dry cabinets and N2 cabinets offer humidity control from 1%RH to 60%RH, with features like multi-point gas supply and automatic nitrogen saving. These solutions help preserve sample integrity and reduce measurement drift, especially for moisture-sensitive devices.

Regular maintenance of the instrument, including cleaning of optical components and replacement of electron source, is also vital. Documenting all calibration and environmental conditions ensures traceability and reproducibility.

Conclusion#

Improving CD metrology accuracy requires a holistic approach that addresses instrument, model, and environmental errors. Optical scatterometry and e-beam metrology each have unique uncertainty sources, but both benefit from rigorous calibration, advanced algorithms, and stable sample storage. By implementing cross-technique validation and uncertainty quantification, metrologists can achieve sub-nanometer accuracy.

As device dimensions continue to shrink, the demand for precision will only grow. Investing in calibration infrastructure and environmental control, such as the dry storage solutions from EJER, is not just a best practice but a necessity for competitive semiconductor manufacturing.