Exclusive: Sigma and Foveon discuss the forthcoming SD1
One of the few real surprises at Photokina 2010 was Sigma's announcement of its forthcoming SD1 DSLR, and that at its heart would be a new Foveon sensor that would offer 15.4x3MP resolution rather than the existing 4.7x3MP. We spoke to Sigma Chief Operating Officer Kazuto Yamaki, Foveon Vice President for Technology and Operations, Shri Ramaswami and its Vice President for Strategic Marketing, Rudy Guttosch about the work that's gone on.
The latest chip, with 46 million photodiodes providing full color information at 15.4 million positions will be at the heart of the SD1, which is due in early 2011. But work started two years ago, the companies say: 'Foveon was working on a different project when we bought to company, after the acquisition we started work', explains Yamaki. It was immediately clear that higher resolution was needed, he says: 'We want to be able to offer the best image quality, the highest resolution. Our customers appreciate the color detail and crisp image quality. If we enlarge the images from our current sensors they still look good but we knew we needed to increase resolution.'
'We knew we could make pixels smaller due to the previous project we'd been conducting for mobile phones,' says Guttosch: 'this gave us the confidence to be able to propose a small pixel, high pixel count sensor. It was discussed during acquisition: the goal was by Photokina 2010 to have a sensor with the right characteristics to make a big impact.'
'It gave us a good chance to reflect on what the strengths of the X3 technology are', he says. 'The spectral response of human visual system has its peak in green. The Bayer design takes advantage of that by having two green pixels in every four. With our X3 technology every darned pixel captures green. The only way to take advantage of this difference is to get to the same pixel size. What we do better than anyone is provide that green luminance information, which is perceived as image sharpness.' 'We knew this advantage but hadn't really made the most of it', admits Yamaki.
'We responded by driving pixel size down to equivalent sizes,' says Guttosch: 'a sensor with an equivalent number of pixels has a √2 resolution advantage over Bayer for the green channel and about 2 times for other colors. Clever demosaicing reduces the gap but you still have an anti-aliasing filter with Bayer. And, if you look at the maximum resolution of most Bayer cameras, they give a figure that's around 70% (1/√2) of the potential maximum value (Nyquist), whereas we can get right up towards the Nyquist frequency.
'The [15.4x3mp] goal was set at the beginning', says Ramaswami. The team was confident that the 15.4x3mp figure would be high enough to make the required impact, he says: ''It's hard to know where someone else is going to be but looking at historic trends, the rest of the industry has kept pace with what you'd expect, maybe a little late but not hugely so.'
'We'd been working on cellular phone sensors which was a greater challenge in terms of sensor design. But, while working on this much larger scale is technically a little easier the image quality challenges are so much harder. In terms of image quality and dynamic range, for example.'
This also explains the move from the smaller 1.7x crop size of previous Foveon sensors up to full 1.5x crop APS-C size, says Ramaswami: 'We don't want the pixels much smaller than they are: not too big, not too small.'
The reason is simple, explains Yamaki: 'Image quality is the number one priority - not losing the sharpness and detail our customers love from our products. Dynamic range and noise are also both important.'
But designing the sensor is only part of the story, Yamaki stresses: 'There are two challenges during the development process - improvements in the sensor and improvements in the image processing. We can guarantee that there will be improvements in both [noise and dynamic range] over the previous generation but we can't yet say by how much. Our existing customers expect a different type of image quality - very crisp, very clear. You could make a low noise camera by blurring away all the detail but we don't do that. That doesn't mean we don't care about it - we will improve.'
'We need to be in the same class as other DSLRs on the market' says Guttosch.
The higher pixel-count itself presented a challenge for the designers in terms of getting the data off the sensor. High readout speeds are useful not only in terms of continuous shooting rate but also in terms of providing smooth live view or fast contrast detection AF in mirrorless cameras (such as Sigma's DP series).
'Readout speed was another challenge for the team that produce the circuitry for the chip', says Guttosch: 'they had to redesign the power supply, the row drivers, the ADC, everything associated with CMOS chips to enable fast readout rates. We're confident that it is up to contemporary standards.'
'This sensor is very different for the one we have used until now. The new generation sensor has seen a complete change in pixel architecture', confirms Yamaki.
Sadly, these high readout speeds won't initially be used to provide video. 'There is no video on the SD1,' confirms Yamaki: 'we will consider it for a future model. We had many things to do and we have to prioritize. The existing customers are very keen on still images and the first priority is to product the highest quality stills. Maybe the second or third step will be to perfect video.'
And the results should be worth the wait, they suggest: 'Video is a special case', says Ramaswami: 'the number of pixels is set. When you look at Foveon 1080, Bayer doesn't look anywhere close.'
There were simply too many other things to be done first, explains Yamaki: 'With a CFA [Bayer] sensor they can implement incremental improvements. In the case of Foveon this is just the third generation of sensor and we have many things to do. But after the Foveon acquisition we were able to set the goals so the speed at which we're making such improvements is faster. For now we're focused on photo enthusiasts and high-end cameras so those improvements have been accelerated.'
And Guttosch thinks this focus is correct: 'We're totally focused on the DSLR market because it's that market that really appreciates what we do', he says.
Unlike many manufacturers who out-source their manufacturing, Sigma makes almost the entire camera in-house. There are still a handful of components it buys in, however. 'We worked with a semiconductor manufacturer for the AF sensor and again purchased the AE sensor from a supplier', says Yamaki. This hasn't meant buying from other camera makers though, he said: 'this is a specified industry - they supply us with parts to our specifications. We didn't work with any other camera maker.'
The in-house method will put pressure on the company's ability to fulfil demand, it says: 'We are aiming to have everything ready for next spring but we haven't announced release dates yet. We may have to change the release date in different markets to match manufacturing capacity.'
There's plenty of work for Foveon to do still, too: 'Image processing is something you're always working on, always improving. But on a certain date you have to have a product ready. It's very different to when you're just supplying a sensor to another manufacturer', says Guttosch. But it has its benefits, he says: 'it's very satisfying for the team at Foveon to work towards a final product, rather than having to hand it over to someone else. They have a great enthusiasm for photography, just like the guys at Sigma.'
It's a point Yamaki echoes: 'We are fortunate we only work in photography. Engineers come to work for us because they want to work in photography. They know we only develop lenses and cameras so we get the people who are passionate about it.'