The Art of Making Large Image Sensors (or Quilting with Silicon)

While most of the semiconductor world is busy shrinking chips to ever smaller sizes, high performance image sensor manufacturers are often pushing the bounds of chip size in order to achieve ever higher image resolution.

In the 1980s and early 1990s most image sensors were manufactured with micron lithography where an entire wafer was exposed in a single shot. This was possible because the feature sizes were large enough and the wafer size small enough that a photomask as large as the wafer itself could be projected onto the wafer with enough precision to reproduce the required features. That meant that an image sensor as large as the wafer itself could be “easily” manufactured. In fact, Teledyne DALSA was one of the first companies to produce such devices – in the early 1990’s we developed a sensor that we called the “Canada Chip” on a 150 mm CCD wafer – it included as many pixels in a single imager as was then the population of Canada – 25 million.

As silicon processes moved down to submicron feature sizes (a good thing, else CMOS would never have re-emerged as a viable image sensor technology), and as wafer sizes grew (first to 200 mm and then to 300 mm) lithography moved to much smaller masks, and wafer exposure to a “step and repeat” approach. All of a sudden the largest device that could be manufactured in a single exposure was of the order of 25 mm x 25 mm. Fortunately, clever engineers at several different companies (including DALSA) devised a methodology that is now generally referred to as “stitching”. With stitching, a single image sensor can be built from a sequence of exposures to produce a device that can be many times larger than the size of a single mask.

Typically the pixel array is constructed from blocks of a few thousand pixels. The mask contains only a single instance of this block, but by stepping the mask by the equivalent of the block size, multiple instances of the pixel block can be generated side by side on the wafer. The blocks are analogous to patches in a quilt. Once the circuitry that surrounds the pixel array is added (different blocks on the mask) then the device is complete. One of the attractive features of this method is that a single mask set can be used to manufacture devices of many different sizes.

If you’ve seen a satellite image, a Google Earth image, or a digital Xray image, there is a good chance that you were looking at an image that was produced by a stitched image sensor. At a less exotic level, the screen of your flat panel TV, and many of the printed circuit boards inside your consumer electronics, were probably inspected during manufacturing by a camera with a stitched image sensor (maybe by a Teledyne DALSA image sensor!).

This approach is simple in principle, but as always, the devil is in the details. To produce these types of devices with no evidence of stitch lines requires some pretty specialized IP, as well as close cooperation between the photolithography engineers, the silicon process engineers, and the circuit designers. At the end of the day it turns out to be as much an art as a science.

[UPDATE: February 8, 2012]

I received some excellent feedback from Jan Bosiers who leads R&D in our Eindhoven location (sensor design in Teledyne DALSA happens in Waterloo, Eindhoven, and Santa Clara). The first stitched imager from what is now Teledyne DALSA was produced by Jan and his team in the early 1990s. It was an HDTV CCD image sensors that was used to broadcast the 1992 Winter Olympics. Regarding applications, Jan pointed out that stitched imagers are also used in full-size 35 mm DSC cameras, and in larger medium and large format professional cameras. Jan also caught a typo in my original post — the first sentence in the fourth paragraph should have stated that the pixel array block in stitched designs is typically a few thousand to a few million pixels in size. Thanks for your helpful comments, Jan — much appreciated.

Eric F

About Eric F

Eric has been involved in the world of photonics for more than 25 years. Most of that time has been spent developing image sensors, first with CCD technology, and for the past 14 years with CMOS. Favourite activities outside of work are family and almost anything related to music.
Posted on by Eric F. This entry was posted in Machine Vision and tagged , , , , , . Bookmark the permalink.

Comments are closed.