WO2008152020A1

WO2008152020A1 - Method for semiconductor substrate inspection

Info

Publication number: WO2008152020A1
Application number: PCT/EP2008/057169
Authority: WO
Original assignee: Icos Vision Systems Nv
Priority date: 2007-06-12
Filing date: 2008-06-09
Publication date: 2008-12-18
Also published as: EP2165312A1

Abstract

A method is provided for detecting anomalies in a semiconductor substrate comprising the steps of: -providing a semiconductor substrate -capturing an inspection image of the substrate -generating a reference image of the substrate -subtracting the reference image from the inspection image, thereby generating a resulting image -examining the resulting image characterized in that the reference image is generated from the inspection image.

Description

METHOD FOR SEMICONDUCTOR SUBSTRATE INSPECTION

FIELD OF THE INVENTION

The present invention relates to a method for detecting anomalies in semiconductor substrates.

BACKGROUND OF THE INVENTION

In semiconductor processing and manufacturing of semiconductor components and integrated circuits, quality control is very important at every stage of the manufacturing process.

Quality control in the field of semiconductor processing is to a high extend directed to detection of defects, in particular semiconductor substrate anomalies, such as cracks and micro-cracks, scratches, dirt, voids, etc. Since even micro-cracks, penetrating or non-penetrating in the substrate, can cause breaking of the substrate during further processing, it is very important to be able to detect these cracks in an early stage of processing. For example for solar cell production, polycrystalline silicon substrates are used which are very brittle. If micro-cracks are present, probably the substrate will break during further processing.

Quality control of a semiconductor substrate relies heavily on optical inspection, because for detecting anomalies optical inspection methods are beneficial in terms of throughput compared to other inspection methods.

A common method to optically detect anomalies on semiconductor substrates compares an image of the substrate part to be inspected with an image of such substrate part containing substantially no anomalies of at least any anomalies of the kind to be detected. The first is usually called the inspection image, while the latter is usually called the reference image. To compare both images, the reference image is then subtracted from the inspection image. The pixel values which after subtraction are higher than a fixed threshold value are labeled as surface anomaly. However, this method can only be applied if the reference image has substantially the same gray values, i.e. the same background image, as the inspection image, if there is no geometrical variation, e.g. scaling or distortion, between the inspection and reference image, and if both images can be well aligned in order to subtract images from exactly corresponding substrate parts from each other and not to cause false positives by misalignment.

In some cases a reference image having substantially the same gray values as the inspection image is unavailable simply because the semiconductor substrate to be inspected is never identical to a corresponding substrate which could be used as reference surface.

An example demonstrating the shortcomings of a referential inspection method is the inspection of polycrystalline silicon substrates used in solar cell production. The pattern of crystal boundaries at their surface is never identical. Consequently, a reference image having the same gray values as the inspection image can never be captured.

Methods to potentially alleviate the above problem are so called non-referential methods, i.e. detecting defects without a reference image. C-H. Yeh and D. -M. Tsai (Int J Adv Manuf Technol 17:412-424, 2001 ) suggest a non-referential method for substrate conducting path inspection, and Sheng-Uei Guan, Pin Xie, Hong Li (Machine Vision and Applications 13 (5-6), pp. 314-321 , 2003) suggest a non- referential method for repetitive patterned wafer inspection.

Both methods are, however, not applicable for substrates with low contrast (small grey value variations) or non repetitive structures, e.g. polycrystalline silicon substrates. In other words, such referential methods have the risk to not detect certain anomalies. Moreover, the proposed methods are time consuming.

To overcome the disadvantages of prior art, a preferred method would be a method without a need for alignment of corresponding images and also without the risk of not detecting anomalies which should have been detected. In contrast to the prior art methods, the method according to the present invention does not comprise alignment of corresponding images and the risk of missing anomalies to be detected is diminished.

SUMMARY OF THE INVENTION

The present invention is directed to a method for detecting anomalies in a semiconductor substrate comprising the steps of:

- providing a semiconductor substrate - capturing an inspection image of the substrate

- generating a reference image of the substrate

- subtracting the reference image from the inspection image, thereby generating a resulting image

- examining the resulting image characterized in that the reference image is generated from the inspection image.

It is further directed to an apparatus for applying a method in accordance with the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 shows a general algorithm in accordance with the invention.

Figure 2 shows an algorithm to generate a reference image from an inspection image.

DESCRIPTION OF THE INVENTION

A person skilled in the art will understood that the embodiments described below are merely illustrative in accordance with the present invention and not limiting the intended scope of the invention. Other embodiments may also be considered.

In a first object, the present invention provides a method for detecting anomalies in a semiconductor substrate comprising the steps of: - providing a semiconductor substrate

- capturing an inspection image of the substrate

- generating a reference image of the substrate

In this method a reference image is generated from the inspection image itself. In contradiction to prior art methods using image subtraction, there is no need to record a separate reference image from the same substrate part as the part to be inspected, and containing substantially no anomalies of at least any anomalies of the kind to be detected. Consequently, there is no risk of misalignment. Thus, the method has a non-referential approach, although image subtraction is applied.

The general inspection algorithm in accordance with the present invention is shown in figure 1. A semiconductor substrate of which the surface has to be inspected is positioned on a substrate holder. An image of the surface part under inspection is captured by a camera. From this inspection image a reference image is generated containing no anomalies of at least any anomalies of the kind to be detected. This reference image is then subtracted from the inspection image and the resulting image is examined by grouping and locating the anomalies.

The step of generating the reference image from the inspection image may be done by low pass filtering. Low pass filtering means the low frequencies are kept in the image, while the high frequencies are removed from the inspection image, since high frequencies may be generated by anomalies.

Low pass filtering methods may be, but are not limited to mean and median filtering, or wavelet transformation .

In a preferred method in accordance with the present invention, wavelet transformation is used. Wavelets are mathematical functions that cut up data in different frequency components, and then study each component with a resolution matched to its scale. They process data at different resolutions, which makes it possible to distinguish between small and large features in an image.

In figure 2, an algorithm for generating a reference image from an inspection image is shown. The inspection algorithm is transformed by a wavelet transformation resulting in a low pass image containing low frequencies and a high pass image containing high frequencies. The transforming process may be repeated by a number of iterations N, usually 1 , 2 or 3, optionally with a different wavelet compared to the first wavelet, resulting in a final low pass image. This low pass image is then inversely transformed by an inverse wavelet transform. This inversely transforming process may also be repeated by a number of iterations N, usually 1 , 2 or 3, optionally with a different wavelet compared to the first inverse wavelet, resulting in a final reference image. The inverse transform wavelets should be the inverse of the transform wavelets.

In a more preferred method in accordance with the present invention, the wavelet transformation uses n^th order polynomials, in particular Daubechies wavelets or Haar wavelets.

The resulting image may optionally be pixel wise raised to the power of two and finally pixels are labeled by a labeling algorithm and grouped to anomalies.

In accordance with the present invention, the semiconductor substrate to be inspected may be of any material used as substrate in semiconductor processing, such as but not limited to silicon, germanium, silicon germanium, gallium arsenide, silicon oxide or silicon nitride. In particular polycrystalline silicon used as solar cell substrate may be inspected.

In accordance with the present invention, the step of capturing an inspection image may comprise illuminating the semiconductor substrate with a backlight having wavelengths where the substrate is transparent. Anomalies should be less transparent than the substrate or not transparent within the same wavelength range. In case of silicon substrates, the wavelength range should be in the infrared (IR) band, preferably in the near infrared band (NIR), and more preferably above 1 micrometer, because silicon is transparent above 1 micrometer and opaque for shorter wavelengths. For germanium, the wavelength where it is transparent is around 1.88 microns. For gallium arsenide, the wavelength where it is transparent is around 0.87 microns.

In accordance with the present invention, anomalies to be detected may comprise cracks, scratches, or dirt. In particular non-penetrating micro-cracks may be detected. Non-penetrating micro-cracks are very difficult to detect by prior art methods. Because they are non-penetrating, light entering the crack can not leave it at the other side of the substrate. It gets refracted on and in the crack and attenuated in the direction of the sensor. However, by making the semiconductor substrate transparent when using a dedicated wavelength range, the fact they are non-penetrating does not pose a substantial problem anymore, because the non-penetrating crack will appear as dark area.

In a preferred embodiment of a method in accordance with the present invention, the step of capturing an inspection image may comprise illuminating a polycrystalline silicon substrate with NIR backlight and the reference image may be generated from the inspection image by using 2^nd order Daubechies wavelet transformation in order to detect non-penetrating micro-cracks at the polycrystalline silicon substrate.

In a second object, an apparatus is provided for applying a method in accordance with the invention. This apparatus may comprise a sensor, a light source directed towards the sensor, a semiconductor substrate holder positioned between the sensor and the light source, and a calculating unit connected to the sensor.

The sensor may be a camera using CCD, CMOS, or other technology. It has to be sensitive to a wavelength range where the semiconductor substrate is transparent. Preferably it has to be sensitive in the IR and in particular the NIR band.

The camera may be equipped with a lens. In order to improve image quality, the camera may comprise a lens optimized for the used wavelength range and an optical filter only transparent for the used wavelength range. The calculating unit where the sensor is connected to may be machine vision hardware comprising a frame grabber and dedicated software.

The light source may be a continuous or a flash light source. It emits wavelengths in the range, or at least a part of it, where the sensor is sensitive for. In case of silicon substrates to be inspected, a xenon flash backlight may be used with wavelengths in the NIR band.

EXAMPLE

The object of the experiment is to detect penetrating and non-penetrating micro- cracks in as-cut and etched polycrystalline silicon wafers used in solar cell applications.

The wafers are illuminated form the back using a strong Xenon flash light with a spectrum well reaching into the NIR band. A 4MPixel camera is used which has a sufficient sensitivity in this band.

The major challenge is to distinguish between the crystal boundaries of the polycrystalline silicon wafer surface and the micro-cracks. Crystal boundaries have a random orientation as the wafers are cut at an arbitrary angle with respect to the crystal boundaries.

A 2^nd order Daubechies wavelet transformation is used to select grey-scale gradients in a narrow frequency band ignoring low-frequency and high-frequency gradients of the inspection image, thereby generating a reference image which is subtracted from the inspection image. As a result, only gradients with the selected frequency band are visible in the resulting image.

This method is applied to as-cut wafers with artificially created cracks, etched wafers with artificially created cracks and wafers with cracks accidentally created in the production line. In most cases, cracks are detected. In some of these cases, besides cracks, also crystal boundaries are detected as crack.

Claims

CLAIMS :

1. A method for detecting anomalies in a semiconductor substrate comprising the steps of:

- providing a semiconductor substrate

- capturing an inspection image of the substrate

- generating a reference image of the substrate

2. A method according to claim 1 , wherein the step of generating the reference image is done by low pass filtering.

3. A method according to claim 2, wherein the low pass filtering is done by wavelet transformation.

4. A method according to claim 3, wherein the wavelet transformation is done using Daubechies wavelets.

5. A method according to claim 1 to 4, wherein the step of making an inspecting image comprises illuminating the semiconductor substrate with a backlight having wavelengths where the substrate is transparent.

6. A method according to claim 1 to 5, wherein the anomalies comprise nonpenetrating micro-cracks.

7. A method according to claim 1 to 6, wherein the semiconductor substrate is of polycrystalline silicon.

8. A method according to claim 7, wherein the semiconductor substrate is a solar cell substrate.

9. A method according to claim 8, wherein the backlight wavelengths are in the range of NIR.

10. An apparatus using the method according to any of the above claims.