Bob Newman looks at the pros and cons of Sigma’s Foveon sensor, particularly the new sensor in the dp2 Quattro

Foveon sensor

Left: The top blue-sensitive layer has four times the resolution of the green and red- sensitive layers, and is used to determine luminance. Middle: The processor combines the luminance and colour data. Right: Sigma says full colour resolution is preserved in the final image

The standout feature of Sigma’s cameras from the start has been the company’s use of the unique three-layer Foveon sensor, first as Foveon’s major customer and more recently as the owners of the sensor company. Whereas almost every other colour camera uses a Bayer colour matrix, with red, blue and two green pixels arranged in groups of four across the face of the sensor, the Foveon design stacks blue, green and red pixels with three colour pixels occupying the same space on the surface of the sensor. This works because light photons with different energies penetrate to different depths in the body of the silicon, which forms the sensor. Blue light penetrates the least, while red penetrates furthest. Thus, with three stacked pixels, the uppermost receives predominantly blue light, the middle mostly green, and the lower red, without further colour filtering. This elegant design produces the only sensor currently available that can sense all three colour primaries at each location.

However, there is rarely such a thing as a free lunch, and the design does suffer from some drawbacks. First, it is very hard to design the electronics of the sensor in such a way that all the accumulated photocharge is read from the pixels, particularly the lower ones. This has led to the Foveon sensors tending to have poor low-light capabilities. Second, the filtering is not as precise as the dye filters used in a conventional sensor, which leads to some very complex colour decoding software and also what are called ‘metamerism’ problems, whereby some colours which should look similar can turn out to be wildly different. At the same time, since the human eye’s colour resolution is not as high as its luminance resolution, capturing all three colour channels at every location might be seen as producing more information than is actually useful.

In the new sensor for the dp2 Quattro, Sigma has undertaken a thorough design review of its sensor to address some of the perceived problems. The first change is that the three layers of pixels no longer have the same resolution. The lower two layers (the green and red layers) have one quarter of the resolution of the top (blue) layer, with each pixel in the lower layers covering the same space as four in the upper layer. In this design the luminance information is captured by the top layer, while chrominance is captured at a lesser resolution by the lower layers, thus addressing the ‘excess information’ issue. Making the lower pixels larger also makes it easier to design for a complete charge collection from them, and maybe thereby increase the low-light performance of this sensor.

Possible downsides to the new design arise from the fact that the top (blue) pixel layer has the wrong spectral response to act as a luminance layer, thus reconstruction of a ‘correct’ luminance channel will require the use of information from the lower resolution layers below. This means that the raw decoding from this sensor will involve the same kind of spatial interpolation as the normal Bayer sensor, which the Foveon design has, until now, been free of.