Friday, May 03, 2024

Foveon sensor development "still in design stage"

Full-frame Foveon sensor "still at design stage" says Sigma CEO, "but I'm still passionate"

"Unfortunately, we have not made any significant progress since last year," says Sigma owner and CEO Kazuto Yamaki, when asked about the planned full-frame Foveon camera. But he still believes in the project and discussed what such a camera could still offer.

"We made a prototype sensor but found some design errors," he says: "It worked but there are some issues, so we re-wrote the schematics and submitted them to the manufacturer and are waiting for the next generation of prototypes." This isn't quite a return to 'square one,' but it means there's still a long road ahead.

"We are still in the design phase for the image sensor," he acknowledges: "When it comes to the sensor, the manufacturing process is very important: we need to develop a new manufacturing process for the new sensor. But as far as that’s concerned, we’re still doing the research. So it may require additional time to complete the development of the new sensor."

The Foveon design, which Sigma now owns, collects charge at three different depths in the silicon of each pixel, with longer wavelengths of light able to penetrate further into the chip. This means full-color data can be derived at each pixel location rather than having to reconstruct the color information based on neighboring pixels, as happens with conventional 'Bayer' sensors. Yamaki says the company's thinking about the benefits of Foveon have changed.

"When we launched the SD9 and SD10 cameras featuring the first-generation Foveon sensor, we believed the biggest advantage was its resolution, because you can capture contrast data at every location. Thus we believed resolution was the key." he says: "Today there are so many very high pixel-count image sensors: 60MP so, resolution-wise there’s not so much difference."

But, despite the advances made elsewhere, Yamaki says there's still a benefit to the Foveon design "I’ve used a lot of Foveon sensor cameras, I’ve taken a bunch of pictures, and when I look back at those pictures, I find a noticeable difference," he says. And, he says, this appeal may stem from what might otherwise be seen as a disadvantage of the design.

"It could be color because the Foveon sensor has lots of cross-talk between R, B and G," he suggests: "In contrast, Bayer sensors only capture R, B and G, so if you look at the spectral response a Bayer sensor has a very sharp response for each color, but when it comes to Foveon there’s lots of crosstalk and we amplify the images. There’s lots of cross-talk, meaning there’s lots of gradation between the colors R, B and G. When combined with very high resolution and lots of gradation in color, it creates a remarkably realistic, special look of quality that is challenging to describe."

The complexity of separating the color information that the sensor has captured is part of what makes noise such a challenge for the Foveon design, and this is likely to limit the market, Yamaki concedes:
"We are trying to make our cameras with the Foveon X3 sensor more user-friendly, but still, compared to the Bayer sensor cameras, it won’t be easy to use. We’re trying to improve the performance, but low-light performance can’t be as good as Bayer sensor. We will do our best to make a more easy-to-use camera, but still, a camera with Foveon sensor technology may not be the camera for everybody."

But this doesn't dissuade him. "Even if we successfully develop a new X3 sensor, we may not be able to sell tons of cameras. But I believe it will still mean a lot," he says: "despite significant technology advancements there hasn't been much progress in image quality in recent years. There’s a lot of progress in terms of burst rate or video functionality, but whe
n you talk just about image quality, about resolution, tonality or dynamic range, there hasn’t been so much progress."

"If we release the Foveon X3 sensor today and people see the quality, it means a lot for the industry, that’s the reason I’m still passionate about the project."


  1. The major advantage compared to Bayer sensor would be improvement in low light sensitivity, with upper bound of 2x improvement if they get the photodiode process to be perfect in color separation. The improvement in resolution upper bound is up to 1.4x but as mentioned here many high resolution sensors exist and solve the resolution problem.

    1. Agree, but the risk for them is that it could arrive at the same time than color routing systems using metasurface for example that would provide high sensitivity and potentially better color separation at the same time, and fully compatible with current pixel structure.

    2. For photography, imho, none of these matter. Many of the photographers ( users of these cameras) are for recipes that apply film simulations and/ or other lightroom processing to get different effects. Maybe the concept may have application in machine vision or industrial where color separation is important on a pixelated sensor. But does it have the necessary quality factor for image separation?

    3. I think you do not read the text carefully, low light is a challenge for X3 tech: "We’re trying to improve the performance, but low-light performance can’t be as good as Bayer sensor."

    4. The colour separation can not be "perfected" as it relies on process which is inherently probabilistic, or statistical in nature. Well, placing colour filters inside the silicon could do it, but hardly practical. Also, as far as I've know, marjority of the vertical area where the three photodiodes are is essentially "dead" for photon collection, thus reducing QE (of longer wavelengths) a lot. It's hard to see meaningful purpouse for Foveon any more.

  2. If this pixel cannot provide more sensitivity and has a fuzzy boundary between the colors, there is still an advantage - having better resolution for the same pixel pitch Bayer sensor.

    1. In principle you are right, but with today's advanced demosaicing algorithms the abovementioned advantage is really, really minimum. For sure the Bayer pattern still need demosaicing, the Foveon solution does not.


All comments are moderated to avoid spam and personal attacks.