Whom is This Article For?
The recently announced Phase One IQ260 represents a major achievement: it uses a Dalsa sensor which is capable of long exposures. In fact, it can do a one-hour exposure at ISO 140. For most photographers this specification will be enough – the details of how it was accomplished, unimportant. But some will be interested in the details, the story behind the massive R+D project and the history behind the development of the Phase One IQ260’s long exposure. Just how did they do it?
Wait, What? Dalsa Long Exposure?
When I first heard that long exposure was returning to the Phase One IQ series I was very surprised. As a technical expert at Phase One’s largest dealer I was very curious how this was accomplished. I had learned, for so long, that the underlying design of a Dalsa sensor did not allow for long exposures – that it seemed like the natural order of things had been upturned. In fact, it had: “Get ready to see us pull a rabbit out of a hat: long exposure on a Dalsa sensor” said Lau Nørgaard, Firmware engineer at Phase One.
In interviewing Phase One’s development team for this article I found a story more interesting than I imagined; a story which Phase One itself has never told before. This article represents my personal thoughts and opinions, shaped by interviews with Phase One’s development team and the history of Digital Transitions as one of the world’s leading medium format dealer. For a proper appreciation of this accomplishment a trip down memory lane is in order.
Sensor manufacturer musical chairs: The sensor division at Kodak was spun off as TrueSense Imaging in 2011.
Dalsa was acquired by Teledyne the same year. Dalsa acquired the sensor division of Phillips,
supplier of sensors to some early Phase One digital backs, in 2002.
The History of Long Exposure: Kodak and the P 45+
Today at the heart of every medium format digital back is a sensor manufactured by one of two companies:
– Dalsa (now Teledyne Dalsa after a recent acquisition, previously Phillips) and…
– Kodak (now Truesense Imaging after being spun off during the Kodak implosion)
From 2001 to 2008, Phase One used only Kodak sensors. During this period Phase One worked closely with Kodak and, through significant R+D, produced a series of backs* capable of very long exposures. This included the much vaunted Phase One P 45+.
The Phase One P 45+ was capable of a one hour exposure at ISO 50 at room temperature (a bit longer if cold, a bit shorter if warm). It reigned as the king of medium format long exposure from 2007 until the announcement of the Phase One IQ260.
The Phase One P65+ was the first Phase One back with a Dalsa made sensor since the Phase One H10 in 2002.
It featured Sensor+, fast shooting speed, and a full-frame 60mp sensor, but it maxed out at 60 seconds.
The Loss of Long Exposure: Switching to Dalsa
In the late 2000s the pace of development at Kodak’s sensor division slowed considerably, partly attributable to the turmoil of its parent company; Kodak’s most recent sensor (the 50mp KAF-50100) was announced in July 2008, and was not full frame.
Around this time the R+D team at Phase One had a novel idea that would trade resolution for higher ISOs and increased shooting speeds. The idea required binning (aka grouping) pixels together at the sensor level, prior to recording the file, and would therefore require a close collaboration with the sensor manufacturer.
Phase One approached both Kodak and Dalsa with this idea. Dalsa was eager to invest in this long-term development project; Kodak was not. Dalsa was also ready and eager to produce a 60mp full-frame sensor which was larger, higher in resolution, and faster than anything on Kodak’s roadmap. Accordingly, Phase One changed teams.
As a result, Phase One gained a full-frame sensor, faster shooting speeds** and a series of backs with their now-patented Sensor+ technology. As production units started to ship they also found near-universal praise for the skintones, tonal smoothness, and color differentiation of the new sensor. But they lost long exposure. To get it back would require the talent, investment, and several years of failed prototypes, and dead-ends.
Phase One IQ260. 2 min @ ISO140. Photo by Anthony Spencer.
The Long Road Back to Long Exposure: the Phase One IQ180
Almost immediately after the switch to Dalsa, Phase One began R+D projects to develop a Dalsa-based digital back which could do long exposures. “It was a tough decision to put development of additional long exposure sensors on hold in order to bring out a brand new unique range of sensors together with Dalsa,” recalled Phase One CEO Henrik Håkonsson. “As soon as we were certain that we could succeed with the new sensors I told our R+D team to put long exposures back on top of our priorities.”
The effort was split into two prongs: a short term project with modest ambitions to improve long exposures by means of a series of minor tweaks and optimization (squeezing the last drop from the sponge so to speak), and a long-term project to regain the hour-long exposure possible with the Phase One P45+ via a complete redesign of the sensor.
The result of the short-term project was the Phase One IQ180, based on an 80mp sensor co-developed by Phase One and Dalsa. It raised the 1-min@ISO50 bar set by the Phase One P 65+ to 2-min@ISO35. The improvements were modest, but greatly appreciated by many photographers who found that 1 minute was just at the cusp of what they needed and 2 minutes was enough to placate most of their long exposure needs. But many landscape and architectural photographers needed muchlonger exposures still, and this would require much more than small tweaks. It would require a complete redesign, an entirely new technology, and close collaboration between the hardware, firmware, and software teams.
Phase One IQ260. 1 hour @ ISO140. Photo by Thomas Andersen, Phase One.
The Man Behind the Sensor
The sensor in the Phase One IQ260 is the culmination of years of collaboration between Phase One and Dalsa. On Phase One’s side these efforts were led by Thomas Andersen, CTO of Phase One. In the highly specialized world of sensor design Thomas is one of the world’s leading experts. He has his name on patents forpixel binning(sensor+),long exposuresubsystems (expose+), andhigh-quality raw file compression(IIQ-L and IIQ-S). Every digital camera system Phase One has produced since the days of scan backs has benefited from his intimate knowledge of sensor design.
One of the most valuable skills an engineer can have is an accurate gut feeling regarding what will be possible to achieve with a certain technique, even before fully developing that technique. Thomas is known among his colleagues for just this. Jacob Sørensen, head of the IQ Hardware team says “Thomas can see where the road is going even before the pavement is laid.” When Phase One was considering switching to Dalsa, a major influence was Thomas’ confidence that it would be possible to produce a long exposure back around a Dalsa sensor. Even if he knew it would require a lot of work to get there.
Phase One IQ260. 10 min @ ISO140. Schneider 55mm LS. Photo by Niels V. Knudsen
Sand, Heat, and Laser Beams
A sensor starts its life as high-purity silicon sand. Through the precise application of heat and very careful timing that sand is layered into a silicon wafer. That wafer is then etched by a fabricator. The more precise the fabricator and the better the silicon wafer, the smaller the physical features that can be reliably placed on the silicon.
Dalsa, and the industry as a whole, have improved these processes over time, shrinking the size of what can be etched into the silicon, and increasing the sophistication of the electronics inside and around each pixel. There is practically no limit to the sensor designs that can be put on paper, but much like an architect who must make blueprints for a building which will stand in the real-world, a sensor designer must adhere to the circuits, shapes, and sizes which can be physically etched on silicon with a given technology.
The general industry trend is that fabrication processes are improved to create new sensors with higherresolution than ever before (which requires smaller pixels). Such was the case with the development of the sensor in the Phase One IQ180 and Aptus-II 12, which required massive fabrication improvements to allow a design which shrank the pixel size from 6 microns to 5.2 microns while increasingdynamic range and color fidelity.
For the IQ260 Phase One reversed this tradition, backporting these fabrication improvements to design and manufacture a new 60mp sensor for the IQ260. Using these improved fabrication processes allowed Phase One and Dalsa greater flexibility in the design of the sensor, including modifications to the pixel level electronics that would have been previously impossible.
100% crop from previous image.
Thomas the Shepherd, and His Flock of Electrons
The key to great dynamic range and clean shadows is having lots of signal, and not much noise. In a digital sensor the “signal” can be measured by the number of electrons that are collected as a result of light from the scene striking the sensor. During an exposure electrons are captured and stored temporarily in the storage well of the pixel. This stored pool of electrons is the “signal”; there is only one signal.
Noise on the other hand comes from several places, but for long exposures the most important noise source is called “dark current.” Whenever a sensor is on it generates heat. This heat gathers and accumulates on the surface of the sensor. In the world of sensors heat and noise are two words for the same thing. Dark current heat is especially problematic for long exposures because it accumulates over time; the longer the sensor is on the more the dark current affects the stored pool of electrons, causing noise in the final image.
The core innovation at the heart of the Phase One IQ260 sensor is a strategy to keep the stored pool of electrons away from the dark current heat at the surface of the sensor. The well in which the electrons are temporarily stored has a physical width, length, and depth, even if they are measured in microns. This means that part of the well is closer to the surface dark current while other parts of the well are further away. In the IQ260 the deepest section of the storage pool is located far from the surface of the sensor, providing a place of refuge for the electrons.
The process is similar to a shepherd herding his sheep away from the wolves. Thomas explains, “Because electrons have a negative charge we are able to guide them by using a sophisticated set of positively and negatively charged steering gates, herding them to the deepest recesses of the well where they are insulated from the effects of the dark current at the surface.”
When asked if there was a Eureka moment in the development of the Phase One IQ260 sensor, Thomas replied, “No, there was no single moment, there was no eureka. Instead there was a long, difficult and challenging road to produce the result we wanted. In a way this is what I am most proud of.” The development of this technology was neither direct nor easy.
Several prototypes were developed which did not meet Phase One’s expectations. “Having developed the previous king of long exposure we knew we could not stop until we had a back with long exposure even better than the P 45+” said Espen Beck, Phase One Senior Product Manager. Learning from each failure, Phase One and Dalsa arrived at a prototype which performed in the real-world as they had calculated on paper: clean one hour exposures.
Phase One IQ260. 8 min @ ISO140. Cambo Wide RS w/ Schneider 43mm XL. Photo by Doug Peterson
The Phase One IQ260 vs. the Phase One P 45+
All engineering is compromise; there is no free lunch. But in this case the compromise is minimal: to temporarily reduce the volume where electrons are stored the native ISO must be increased to ISO140 which modestly reduces dynamic range when compared to ISO50 on thesame sensor.
Fortunately for users of the Phase One P 45+ the advances in sensor technology in the eight years since the P 45+’s Kodak sensor was announced mean that the performance of the IQ260@ISO140@10min is as as good as the Phase One P 45+@ISO50@10min. In addition the IQ260 has a larger, higher resolution sensor and a nearly 2-stop advantage in light gathering from the higher native ISO.
In addition to long exposure the case for an IQ260 vs. P 45+ is greatly aided by the addition of a retina LCD, touch interface, focus mask, adjustable exposure warning, two-axis electronic spirit level which allow auto-correction of horizon, in-back rating, faster frame rate, wifi for wireless review (including at 100%) on an iOS device, USB3 for connectivity to new/small laptops, better color, sensor+ for higher ISO and faster shooting, and availability of a five year warranty.
Phase One IQ260. 8 min @ ISO140. Cambo Wide RS w/ Schneider 43mm XL. Photo by Doug Peterson
Beyond the Sensor: the A/D convertor and FPGA controller
The sensor is the heart of any digital back. But a better sensor alone does not guarantee better quality; there are many other links in the chain of image quality. This is evident when comparing several backs with identical sensors. For instance the same KAF-39000 sensor was used in several Phase One and Hasselblad digital backs, but launched with varying results concerning long exposures:
– Phase One P 45: approx. 5 minutes***
– Phase One P 45+: 60 minutes
– Hasselblad H3D-39/ii: 32 seconds (later increased to 64 seconds)
Two such links in that chain are the A/D convertor and the FPGA processor. The A/D convertor is the piece of electronics which convertors the purely analog signal (consisting of voltages) into a digital signals (comprised of 1s and 0s). Many users get caught up on the bit-depth of the A/D convertor. In truth this is only a small part of the complex engineering game that is the selection of the best possible A/D convertor. In addition to bit depth, the speed, heat generation, reliability, flexibility, and non-bit-depth related conversion quality attributes vary from one to the next. Phase One selects its A/D convertor based on making Image Quality the first, second, and third priority.
The FPGA is a programmable processor which oversees the process of reading data off the sensor and piping that data through the A/D converter. An FPGA is similar to the ASIC processors used in most cameras (e.g. a Canon/Nikon dSLR) with two important differences: the FPGA can be updated with new code after it’s installed in the camera (whereas an ASIC chip is fixed the moment it is manufactured), and an FPGA is cost prohibitive for lower-cost cameras.
The FPGA + A/D convertor serve as a critical link between the sensor and the rest of the system. It is in this area that Phase One has often made the most progress relative to others. Much like an auto mechanic learns how to best tune the engine of a car by both study and experience, the engineering team at Phase One continues to improve how they tune the activation, readout, and a/d conversion long after the first system ships. That learning is translated into updated FPGA code and released as firmware updates to existing users. This is especially important since Phase One digital backs are designed with an anticipated active service life of more than a decade (based on the experience that many Phase One H 20 users are still using their backs purchased in 2001.
100% crop from previous image.
The Dark (Frame) Side
Alongside the FPGA and A/D convertor is the dark-frame subtraction hardware/firmware which automatically takes an additional exposure after the main exposure. This additional exposure is used to map and measure the location, type, and amount of heat (aka noise) on the sensor. Using this as a guide the noise can be accurately mapped out with minimal impact to detail, color, or dynamic range. Phase One’s patents and engineering capabilities in the area of dark-frame technology are the best in the world, allowing them to push sensors far beyond what other manufacturers have achieved.
Computers are, in general, very poor at reducing noise when little is known about the characteristics of the noise, but if the noise can be profiled and well understood its removal becomes arbitrary. Recent cell phones, for instance, have started to incorporate extra microphones, away from the mouth of the speaker, to listen to the noise around the speaker. By listening to both the [speaker+noise] and the [noise alone] the phone is able to greatly negate the undesired noise and leave a clear voice even in a noisy room. The same principle applies in dark-frame subtraction. By taking separate pictures of both [scene + noise] and [just noise] the removal of the noise becomes much easier.
But it’s not just the amount of noise that matters; the kind of noise matters a great deal as well. Consider the analogy of trying to have a conversation above the clamor of a loud cafe vs. having that same conversation over the rumble of the engines in an airplane cabin. The absolute level of noise could be the same in both cases, but the stochastic and voice-filled chaos of the cafe makes it hard to pick out a single voice, whereas the continuous, almost soothing rumble of a plane engine is much easier to tune out. With a digital image the structure, regularity, and distribution of the noise matters enormously. The same amount of noise in a system can be manifested as ugly, blotchy, clumps of noise, or fine, gaussian, well-structured grain. During the development of the IQ260 much effort was made to consider not only how much noise a certain decision/condition might result in, but also how that noise would be manifested. Part of that process involved the close collaboration of the hardware team with the software team.
Phase One’s fine tuning of Capture One software, especially regarding noise and detail controls,
greatly aided the process of creating the IQ260. Hardware and software, developed in tandem, do what neither could do alone.
The Other Half: Software
Hardware and firmware do not create an image. They create a raw file full of raw numbers which, in and of themselves, are nothing. The second half of the process is the conversion of the raw data into a pretty image. Much like the same sheets of music can be played by several orchestras with a different result from each, the raw data from a digital camera can be handled in raw-processing to many different results. Not just in the “style” of the image (contrast, curves, color adjustments) but also in the underlying quality, accuracy, and type of initial conversion.
Image Quality is the highest priority in Phase One’s software department, responsible for Capture One. Phase One started working on raw conversion around 1997; a decade before Adobe launched LightRoom. It would be easy to assume that after 16 years of working on the math to convert a raw file into a processed file that there would be no more room for meaningful improvement. But as the below ISO 3200 example from a Phase One P 65+, compared in Capture One v6 and v7 shows, they are still making improvement.
Phase One P65+ @ ISO3200 as processed in Capture One v6 (left) and v7 (right)
At Phase One this refinement of raw conversion takes place hand-in-hand with the development of new hardware. The software engineers work literally down the hall from the hardware team; they drink at the same bars after work, share holidays, and are in constant collaboration. The result is that Capture One is extensively fine tuned to extract the absolute best quality from a Phase One raw file. This is amongst the tightest and most productive hardware+software coordination in the industry; it cannot be understated how much this contributes to the process of creating a great end result.
This has never been more true than for the development of long exposure on a Dalsa sensor. The software team worked hand-in-hand with hardware R+D, advising the R+D team on the raw-conversion consequences of each hardware/firmware decision. “With every prototype the quality of the raw data improved, and with every prototype we were able to push the processing math a bit further and as a result give the hardware team immediate feedback on what hardware improvements would help us the most.” said Lionel Kuhlmann, Head of the Image Core team, the group responsible for raw processing in Capture One.
The effect went the other way as well. By having early access to raw files from every prototype, years before the final product would ship, the software team was able to rewrite, refine, and fine tune their math, some of which was found to be broadly applicable to processing of any raw file. In fact, it could be said that many of theimprovements incorporated in Capture One v7, especially the single-pixel noise reduction improvements, are only now, with the public introduction of the IQ260, fully actualized.
Time to Celebrate
Everyone I spoke with at Phase One while gathering background information for this article shared the same disposition. Each of them was very excited that their years of work are finally public in the form of a nearly ready to ship product. Customers usually only see and think about the end-result of such R+D projects. I hope this article provides some appreciation for the enormous and length work that goes into a project before it reaches the end-user.
As Thomas said “We’ve been working on this for years. Now, it’s time to celebrate.” Knowing him, he was already thinking about the next big thing while celebrating.
About Doug Peterson, Digital Transitions
Doug Peterson is on the technical team atDigital Transitions, has worked in medium format for five years, and is also a wedding photographer on the weekends – www.doug-peterson.com. Digital Transitions is a dealer specializing in support, sales, rental, and training of medium format. They are Phase One’s largest dealer and have received the Partner of the Year award multiple times – www.digitaltransitions.com.
Note: Additional 100% crops and full resolution downloads of some images from this article can be viewed and downloaded here:
*The original P+ series with long exposure: P 20+, P 21+, P 25+, P 30+, P 45+
**Phase One’s first Dalsa-based back, the P 65+ achieved 1 fps @ 60mp. Contrast this with the Phase One P 30+ which hit 0.8fps @ 31mp.
***The official specification for the P 45 was “several minutes”. In our experience that means approximately 5 minutes at room temperature.