If you were asked to design a camera sensor, you’d naturally put the photo receptors on top, closest to the light. Oddly enough, because of the way chips are fabricated, until recently most camera sensors have captured light at the bottom, underneath layers of interconnections. The recent introduction of back-illuminated (BI) sensor technology (also referred to as backside-illuminated or BSI) has changed all that. It is now possible to build sensors “the right way round” with the photo receptive layer facing the light. Back illumination has made some headlines for allowing better low-light performance, but its worth diving into the technology, as it is going to be a lot more important than that.
Why camera sensors are upside-down, and why it matters
Silicon is both the substrate on which chips are built and the material that performs the magic of turning photon energy into electrical energy that can be used to create images. It is therefore the simplest solution to create the photosensitive areas in the substrate silicon and stack the electronics on top — leaving openings in the wiring over each photosite (pixel) to allow light to pass through. As camera resolutions have increased, pixel sizes have decreased, especially in smartphones with their tiny sensors. The result is that more and more of the surface area of the sensor is covered by wiring, resulting in less and less light reaching the photosites. So there is a natural need to find a way to move the photosensitive region to the top of the chip, allowing it to gather more light.
Curiously, the human eye and most animal eyes are also built with the photosensitive pigments on the side furthest away from the light streaming through the eyeball. It isn’t known exactly why eyes are designed that way, but designing them this way definitely makes it easier to provide circulation to the energy-hungry rods and cones, as well as allowing cellular debris to be whisked away without floating around inside the eyeball. Creatures like cephalopods who rely on their eyes in the dark waters of the deep ocean do indeed have their photoreceptors close to the lenses of their eyes, to maximize the sheer amount of light captured.
Problems with the traditional front-illuminated design
If a sensor was only a layer of photosensitive silicon, it wouldn’t matter much which side was up. A pixel is a lot more than just the photodiode, however. It typically includes transistors and wiring for amplifying the charge, transferring it to the signal processing portion of the chip, and resetting itself between frames. Those electronics get placed on top of the silicon layer, partially obscuring it from the light and resulting in a well-like appearance for a typical pixel.
As you’d expect, putting the photodiode at the bottom of a well reduces the amount of light that reaches it, with some light bouncing off the wiring above, and some just not having the right angle to make it to the bottom of the well. Microlenses are used to reduce this problem (the human eye uses waveguides known as Muller cells), but a meaningful amount of light is still lost before it gets to the photodiode to be captured. Typical sensor fill factors — the portion of light successfully captured — range from 30% to 80%. By contrast, a back-illuminated sensor can have a fill factor of nearly 100%.
Light bouncing around inside the electronics can also cause other problems such as vignetting and crosstalk. Thus a design which puts the photodiodes on top is clearly desirable. Having the photosensitive area on the side of the chip facing the light also dramatically improves the angular response of the sensor. No longer does light have to be aimed “down the well” of the pixel, but it can strike it from nearly any angle.
The devilishly hard process of putting photo receptors on top
The magic that makes camera sensors possible is also what makes them difficult to build with the receptors on top. Because silicon is both the substrate for the sensor chip and the light-sensitive material, it wants to be both at the top and the bottom of the sensor at the same time. Traditional sensor designs simply start with the silicon and pile everything else on top. That allows for a single, relatively thick, layer of silicon and a straightforward manufacturing process.
To get a silicon layer on top of the sensor, it is necessary to build the chip the same way as a traditional front-illuminated sensor, but then place another layer of silicon substrate on top and flip the entire silicon sandwich over. After that comes the hard part. The original base layer of silicon, which is now on top, has to be thinned to make it act as a light-sensitive layer.
While a typical silicon layer is about 1 mm thick, in order to let light through to photosensitive areas, the back layer of a BI sensor has to be accurately thinned to around 5 to 10 microns for the optimal capture of visible light — less than 1% of its original thickness. Precise thinning, and bonding in a way that prevents delamination, represent some of the major manufacturing challenges that have slowed down the deployment of back-illuminated sensors. With these barriers mostly conquered, Applied Materials, a leading provider of the fabrication technology needed for creating back-illuminated sensors, estimates that 75% of the smartphones shipped in 2014 will use BI sensor technology.
Omnivision helped pioneer the use of BI sensors in 2007, but until recently they have been too expensive for production use. Sony’s Exmor R in 2009 was one of the first production sensors to use back-illumination — nearly doubling its low-light sensitivity compared to other similar chips. Back-illumination broke into the mainstream when it was used in flagship phones including the Apple iPhone 4 and the HTC EVO 4G.
Better light sensitivity is just the beginning of what backside illumination makes possible. Sony has shown that BI technology allows creating a sensor from a stack of chips instead of a single chip — so that the top chip can be optimized for capturing the light and the ones underneath can do the signal processing. A stacked design allows for additional processing at the pixel level without reducing the sensor’s light sensitivity. Recent advances in through-silicon vias (TSVs; interconnects between a 3D stack of chip dies) have made this type of layering possible. This stacking is the same technology that is used in multi-layer memory chips, so it is getting significant investment.
Olympus has demonstrated how a stacked (it calls it 3D) architecture can create new possibilities with its research prototypes of a BI sensor, with a nearly perfect shutter that it calls a “global” shutter. Like a rolling shutter the global shutter can work nearly instantly and without any moving parts, but avoids artifacts by placing the read out electronics on a lower layer behind a shield. Since it takes time to read all the data from a sensor, a rolling shutter can create smearing or other artifacts because the sensor is still actively receiving photons while the voltages are being read. The Olympus design transfers all of the charge off the sensor at once — to the lower, shielded layer — where they can then be read out accurately.
Sony has already started sampling a stacked version of a BI chip with its Exmor RS, which is expected to ship in volume this year. As BI technology continues to improve, look for Sony to be one of the first companies to use it in their larger sensor cameras. The larger pixel size of APS-C and full-frame cameras allows a high fill factor even with traditional front-illuminated technology, but as BI gets cheaper and stacked systems allow additional features, expect it to begin to appear in mirrorless and DSLR cameras.