Browse Prior Art Database

Numerical method allowing multi-layer optical model evaluation in production environment

IP.com Disclosure Number: IPCOM000235532D
Publication Date: 2014-Mar-06
Document File: 8 page(s) / 503K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a numerical method to identify instabilities in a multi-layer optical model, especially in the context of obtaining accurate thickness measurements during the microfabrication of Integrated Circuit (IC) chips. The method allows the inline evaluation of complex multilayer ellipsometric model reliability, enabling an improved production control strategy and an assessment of the confidence level in the results for yield learning splits.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 47% of the total text.

Page 01 of 8

Title

Numerical method allowing multi-layer optical model evaluation in production environment

Abstract

Disclosed is a numerical method to identify instabilities in a multi-layer optical model, especially in the context of obtaining accurate thickness measurements during the microfabrication of Integrated Circuit (IC) chips. The method allows the inline evaluation of complex multilayer ellipsometric model reliability, enabling an improved production control strategy and an assessment of the confidence level in the results for yield learning splits.

Problem

Microelectronic fabrication requires strict process control. Within this fabrication process, one of the challenges is obtaining accurate thickness measurements during the microfabrication of Integrated Circuit (IC) chips. Such measurements utilizing optical methods (e.g., ellipsometry) are especially challenging for a multi-layer stack that includes layers with very similar optical properties. The presence of similar optical properties of the layers results in errors in thickness measurements generated by the optical models.

Solution/Novel Contribution

The novel solution is a numerical method to identify instabilities in the optical model. Furthermore, the presented method enables an improved production control strategy and an assessment of the confidence level in the results for yield learning splits (as shown in the Silicon Germanium (SiGe) example).

Method/Process

The method allows the inline (or eventually using offline data processing) evaluation of complex multilayer ellipsometric model reliability. The method consists of:

1. Installing an additional model in the production environment (e.g., reference, REF, model)

2. Comparing the data for the correlated parameters for both models' Process of Record (POR) and an additionally installed REF model. (Figure 1) The thickness data provided by both models is compared for:

A. Individual thickness values of layers with low optical contrast (very similar composition)


B. The summary thickness of correlated layers (Figures 2 and 3)

3. Achieved data can be also used to reduce measured variation due to the model itself (i.e. average data from both models per layer can be reported on process control charts)


Page 02 of 8

Figure 1: Novel Solution: POR Model & REF Model


Page 03 of 8

Figure 2: Buffer/Main variation across model is significant. Buffer/main variation comes from correlation between buffer and main layer due to similar optical properties (e.g., low optical contrast). Four models are compared for varied EPI process conditions.

x%

y%

Figure 3: Analyzing sum of T1+T2 to exclude T1/T2 model correlations

Method utilization using SiGe Stack that Consists of Four Layers



Page 04 of 8

Interlayer correlation depends on optical contrast (~material properties)
 The smaller the optical contrast between the layers, the higher the correlation  Correlations mainly affect Thk ratio (i.e. ratio between the thickness of the

second and thi...