Multi-step image alignment method for large offset die-die inspection

1. A method for obtaining an aligned die-die inspection image, comprising:

receiving a reference image at a processor, the reference image comprising rows and columns of pixels;

selecting a first local section from the reference image using the processor;

receiving a test image at the processor, the test image comprising rows and columns of pixels;

selecting a second local section from the test image using the processor;

determining, using the processor, an estimated rotation offset and an estimated translation offset from the first local section and the second local section;

performing a rough alignment comprising a test image de-skew using the processor, thereby making a partially-aligned test image; and

performing a fine alignment comprising partitioned translation on the partially-aligned test image to obtain an aligned die-die inspection image.

2. The method of claim 1 , wherein the test image de-skew comprises:

determining a skew angle of the test image using the processor; and

de-skewing the test image using the processor.

3. The method of claim 2 , wherein the skew angle of the test image is determined by performing a skew comparison of the first local section from the reference image and the second local section from the test image.

4. The method of claim 3 , wherein the skew comparison comprises:

performing a fast Fourier transform on the first local section from the reference image using the processor to obtain a reference scene function;

performing a fast Fourier transform on the second local section from the test image using the processor to obtain a test scene function; and

comparing the test scene function to the reference scene function using the processor to determine the skew angle.

5. The method of claim 3 , wherein the skew comparison comprises performing a pattern recognition of one or more prominent features in the test image to determine the skew angle.

6. The method of claim 3 , wherein the skew comparison is performed using a machine learning module to determine the skew angle.

7. The method of claim 2 , wherein de-skewing the test image comprises:

determining, based on the skew angle, for each of the pixels in the test image, a column shift vector and a row shift vector using the processor, wherein

the column shift vector comprises a quantity of pixels to shift collinear to the column containing the pixel and a direction; and

the row shift vector comprises a quantity of pixels to shift collinear to the row containing the pixel and a direction; and

shifting each of the pixels according to its column shift vector and row shift vector using the processor.

8. The method of claim 1 , wherein the partitioned translation comprises:

partitioning, using the processor, the reference image into at least one reference image sub-section;

partitioning, using the processor, the test image into at least one test image sub-section; and

translating, using the processor, the test image sub-section to align with the reference image sub-section corresponding to the test image sub-section.

9. A non-transitory computer-readable storage medium, comprising one or more programs for executing the following steps on one or more computing devices:

receive a reference image, the reference image comprising rows and columns of pixels;

select a first local section from the reference image;

receive a test image, the test image comprising rows and columns of pixels;

select a second local section from the test image;

determine an estimated rotation offset and an estimated translation offset from the first local section and the second local section;

perform a rough alignment on the test image comprising a test image de-skew, thereby making a partially-aligned test image; and

perform a fine alignment comprising partitioned translation on the partially-aligned test image to obtain an aligned die-die inspection image.

10. The non-transitory computer-readable storage medium of claim 9 , wherein the test image de-skew comprises:

determining a skew angle of the test image; and

de-skewing the test image, comprising:

determining, based on the skew angle, for each of the pixels in the test image, a column shift vector and a row shift vector, wherein

the column shift vector comprises a quantity of pixels to shift collinear to the column containing the pixel and a direction; and

the row shift vector comprises a quantity of pixels to shift collinear to the row containing the pixel and a direction; and

shifting each of the pixels according to its column shift vector and row shift vector.

11. The non-transitory computer-readable storage medium of claim 10 , wherein the skew angle of the test image is determined by performing a skew comparison of the first local section from the reference image and the second local section from the test image.

12. The non-transitory computer-readable storage medium of claim 11 , wherein the skew comparison comprises:

performing a fast Fourier transform on the first local section from the reference image to obtain a reference scene function;

performing a fast Fourier transform on the second local section from the test image to obtain a test scene function; and

comparing the test scene function to the reference scene function to determine the skew angle.

13. The non-transitory computer-readable storage medium of claim 9 , wherein the partitioned translation comprises:

partitioning the reference image into at least one reference image sub-section;

partitioning the test image into at least one test image sub-section; and

translating the test image sub-section to align with the reference image sub-section corresponding to the test image sub-section.

14. A semiconductor die-die inspection system comprising a sensor to capture images of features of a die and a computing system comprising:

a beam source, wherein the beam source is a light source or an electron beam source;

a stage configured to hold a wafer in a path of a beam produced by the beam source, wherein the beam is a light beam from the light source or an electron beam from the electron beam source;

a detector configured to receive a portion of the beam reflected from the wafer; and

a processor in electronic communication with the detector configured to perform:

a rough alignment of a test image comprising a test image de-skew, thereby making a partially-aligned test image, and

a fine alignment comprising partitioned translation of the partially-aligned test image.

15. The semiconductor die-die inspection system of claim 14 , wherein the processor is further configured to:

receive a reference image, the reference image comprising rows and columns of pixels;

select a first local section from the reference image;

receive a test image, the test image comprising rows and columns of pixels;

select a second local section from the test image; and

determine the estimated rotation offset and the estimated translation offset from the first local section and the second local section.

16. The semiconductor die-die inspection system of claim 14 , wherein the test image de-skew comprises:

determining a skew angle of the test image; and

de-skewing the test image.

17. The semiconductor die-die inspection system of claim 16 , wherein the skew angle of the test image is determined by performing a skew comparison of the first local section from the reference image and the second local section from the test image.

18. The semiconductor die-die inspection system of claim 17 , wherein the skew comparison comprises:

performing a fast Fourier transform on the first local section from the reference image to obtain a reference scene function;

performing a fast Fourier transform on the second local section from the test image to obtain a test scene function; and

comparing the test scene function to the reference scene function to determine the skew angle.

19. The semiconductor die-die inspection system of claim 16 , wherein de-skewing the test image comprises:

determining, based on the skew angle, for each of the pixels in the test image, a column shift vector and a row shift vector, wherein

the column shift vector comprises a quantity of pixels to shift collinear to the column containing the pixel and a direction; and

the row shift vector comprises a quantity of pixels to shift collinear to the row containing the pixel and a direction; and

shifting each of the pixels according to its column shift vector and row shift vector.

20. The semiconductor die-die inspection system of claim 15 , wherein the partitioned translation comprises:

partitioning the reference image into at least one reference image sub-section;

partitioning the test image into at least one test image sub-section; and

translating the test image sub-section to align with the reference image sub-section corresponding to the test image sub-section.