Academic Company Events NI Developer Zone Support Solutions Products & Services Contact NI MyNI

Customer Solutions

Using LabVIEW and IMAQ Vision to Create High-Resolution Image Mosaics

Author(s):

Edward Delaplaine, Mink Hollow Systems; Eric Lyness, Mink Hollow Systems; David McAndrew, Mink Hollow Systems

Industry:

Semiconductor

Product:

LabVIEW, Vision

The Challenge:

Adding high-resolution, large area image acquisition to a semiconductor failure analysis tool to directly overlay magnetic data images, visible spectrum images, and near infra-red (NIR) spectrum images.

The Solution:

Using LabVIEW and IMAQ Vision to rapidly developed MosaicVIEW, a tool for creating high-resolution, large area mosaic images mapped to real-world coordinates with virtually any camera and translation stage.


image

Before MosaicVIEW
Neocera’s semiconductor analysis tool uses magnetic imaging to detect chip failures. The magnetic data images show current density maps that easily allow for failure analysis but do not effectively disclose the location of the flaw on the system that proved to be inaccurate. At Mink Hollow Systems, we selected a 1 megapixel Hamamatsu IEEE-1394 camera and Thales-Optem NIR 7x zoom lens for the hardware upgrade. We chose this hardware, so mosaic images could be constructed with both visible and NIR illumination.Using NIR illumination, the system can see into the chip because silicon becomes transparent at these wavelengths. Therefore, the complete system can correlate the magnetic image to surface high-resolution mosaics and NIR illuminated, subsurface, high-resolution mosaics.

The software, called MosaicVIEW, is broken into three main components: Calibration, Build Mosaic, and View Mosaic. It allows the semiconductor failure analysis tool to produce high-resolution, large area images from the visible and NIR spectrums that overlay and directly correlate to a magnetic field map of the device under test. The correlation of images and the ease of analyzing the immense quantity of data provide more accurate failure analysis. The following section describes MosaicVIEW’s three main components in detail:

MosaicVIEW’s Components
Calibration assures that each image within a mosaic accurately aligns with adjacent images and the user can overlay multiple mosaics so each pixel on each mosaic corresponds to the same physical point on the target object.

An automatic calibration algorithm determines the camera resolution, pixels per millimeter, and the rotation of the camera with respect to the stage. MosaicVIEW requires both measurements to calculate the scaled orthogonal coordinate system of the target object relative to the camera. Thus, the user accurately can overlay other scaled images or insert them within the mosaic.

The user provides two heuristic calibration values to improve the accuracy of the final mosaic. Overlap specifies the number of pixels by which each mosaic image will overlap the next. This is a useful feature for post-acquisition corrections. With the crop factor allows, the user can ignore a specific number of pixels on the edges of an image to reduce error due to lens distortion.

To build a mosaic, the user needs only to select upper-right and lower-left real-world coordinates. MosaicVIEW uses this rectangle and the calibration parameters to calculate image acquisition locations, so each subimage in the mosaic has known real-world coordinates.

The following automated process constructs a mosaic:

  • Move the stage into position
  • Snap the image
  • Rotate, crop, and decimate the image according to calibration settings
  • Save the resulting mosaic subimage
  • Repeat until the requested region is scanned

Before the building of a mosaic completes, MosaicVIEW allows users to embed an arbitrary number of higher resolution, zoomed images. With the true scale feature of MosaicVIEW, engineers can return the camera to exact locations of areas of concern and record images at a higher magnification.

Neocera engineers use the image-embedding feature to acquire submicron resolution images at fiducials. This allows for precise visual inspection at known high-interest points and lower resolution at less interesting areas. Without this feature, a 7x magnified full mosaic would take about 49 times the memory, making a 1 GB mosaic a 49 GB mosaic. This would also and take 49 times as long to acquire the mosaic, resulting in four hours instead of five minutes.

The View Mosaic application displays the full mosaic in a “helper window,” so the user can always see the full mosaic. The user can select a rectangle, or region of interest (ROI), on the helper window to expand. The software uses the ROI selected by the user to determine which files and parts of files in the mosaic should be loaded and displayed.

It was important to Neocera to run the software on an ordinary computer without overextending the computer’s memory. Since full mosaics can consist of hundreds of image files, each containing several megabytes of image data, viewing can be unwieldy for an average computer. To counter this problem, we implemented an active decimation algorithm that decimates based on a user-configurable maximum image size. The ratio of maximum image size to amount of data available results in a decimation factor. When this ratio is greater than or equal to one, the algorithm loads 100 percent of each image file. When the ratio is less than one, the software decimates each file as it loads.

Viewers can observe mosaics in IMAQ image windows or LabVIEW intensity plots. Engineers easily can export IMAQ image windows to a new file in various formats. Intensity plots provide real-world coordinates (mm, inches, etc.) on the axes of the graphs as the user zooms and pans.

As the user zooms in on an area of the mosaic containing a zoomed image, the zoomed image will be loaded and displayed. This results in the lower resolution pixels appearing larger in size next to the higher resolution image.

Even after calibration, we sometimes found the individual mosaic images did not line up perfectly. This was due to several factors, including fish-eye distortion, small errors in stage motion, and small errors in calibration. Post-processing algorithms aided in making the images align more precisely by adding repeated offsets, custom offsets, or rotation to all or individual mosaic subimages. These features help in system debugging by allowing the user to tweak a correction mask to visually align the mosaic then back-calculate calibration errors to determine systematic problems. Also, users can mask large, time-consuming data sets taken with an out-of-date calibration file that would otherwise be worthless to assemble visually correct and valuable data.

Improvements
We developed a high-resolution image mosaic software tool to correlate magnetic field map data to visible and NIR mosaic images and successfully incorporate the tool into the existing software. By adding the capability of NIR imaging, the system can correlate magnetic field images to UUT surface imagery, as well as subsurface imagery. These data provide visual feedback to an engineer for faster debugging and improved failure analysis.

For more information, contact:
David McAndrew
6880 Mink Hollow Rd.
Highland MD 20777
Tel: 301-854-1579
Fax: 301-854-9746
E-Mail: dmcandrew@minkhollowsystems.com