`
`Embargoed News: June 22, 2009
`
`OmniVision Technologies, Inc.
`
`
`At the heart of any digital camera lies the image sensor. The image sensor is an integrated circuit, like any memory chip
`
`or microprocessor, except that it’s designed to be very sensitive to visible light. Image sensor technology has undergone
`
`rapid technological improvements in recent years. OmniVision’s BackSide Illumination (OmniBSI™) pixel technology
`
`truly offers a giant leap forward in the world of imaging technology.
`
`Image Sensor Trends
`
`Understanding the key trends in the image sensor business helps to illustrate the substantial value offered through
`
`OmniBSI technology. Specifically, the image sensor market is being driven towards lower cost devices, smaller pixels,
`
`higher resolution, thinner camera modules and better image quality.
`
`As with all electronics, the primary trend in image sensors is lower cost and higher performance. Lower cost is achieved
`
`primarily by shrinking all features on the die. Aside from transistors, whose size reduction is driven mainly by the
`
`microprocessor and memory industries, the largest devices on an image sensor are the actual pixels themselves. It is the
`
`innovation in the pixel technology that drives higher performance.
`
`The second trend in image sensors is better performance. ‘Make them smaller and better’ has been the industry’s
`
`mantra for many years, from salespeople to the pixel designers. The first image sensors for mobile cameras used 5.6 µ
`
`pixels, shifting to 2.8µ, 2.2µ, 1.75µ and smaller. Current leading‐edge devices use 1.4µ pixels, with 1.1 µ and 0.9µ pixels
`
`in development. For comparison, the wavelength of visible light varies from 0.65 µ (deep red) to 0.40 µ (blue). Evidently,
`
`pixels are getting close to some fundamental physical size limits. With the development of smaller pixels, engineers are
`
`asked to pack in as many pixels as possible, often sacrificing image quality.
`
`The third most important trend in image sensors is increased resolution. While resolution, measured in megapixels, is
`
`easily the most defining specification of a digital camera, it doesn’t tell the whole story. Consumers demand higher
`
`resolution coupled with other performance factors.
`
`The fourth trend in sensors is reduction in camera module size. In the majority of consumer electronics, the camera is
`
`built as a small, self‐contained module integrated into the final product. The camera module contains the image sensor,
`
`a lens, a tiny circuit board and a plastic housing to hold it all together. Like sensors, camera modules are under
`
`Sony Corp. v. Raytheon Co.
`IPR2016-00209
`
`Raytheon2028-0001
`
`
`
`enormous pressure to shrink in size and cost while increasing in resolution. A primary focus in recent years has been to
`
`make modules as thin as possible, a trend emphasized by ultra‐thin mobile phones.
`
`All the while, consumers expect mobile cameras to deliver image quality and features comparable to high‐end digital still
`
`cameras. And why shouldn’t they, with features like auto‐focus, the resolution of digital still cameras (DSC) resolutions
`
`and image processing features like red‐eye reduction, image stabilization and HD video recording?
`
`Traditional CMOS Image Sensor Design
`
`At its core, a solid state image sensor is a two‐dimensional array of light‐sensitive pixels. Each pixel contains a single
`
`photodiode with supporting electronics. Photodiodes are the actual light sensing structure, and convert visible light
`
`photons into electrons. The number of electrons collected by each pixel is measured by the in‐pixel electronics, and the
`
`values are transmitted to the rest of the camera’s electronics for further amplifying, image processing, storage and
`
`display.
`
`In traditional CMOS Image sensors the “front” of most integrated circuits is the top surface, which contains all
`
`photodiodes, transistors, metal wiring layers and other electronics. This front‐side layer uses only the top 1 percent or
`
`so of the silicon wafer. The remaining 99 percent becomes the “backside”, and serves only as mechanical support and
`
`bonding surface, as well as an electrical ground terminal.
`
`In image sensors, the photodiode array is fabricated along with the transistors on the front‐side surface. The lens
`
`projects its image onto the front surface, where the light is collected by the photodiodes and digitized into an image.
`
`Pixel Fundamentals
`
`To understand the advantages of OmniBSI technology, some background information on pixel performance and the
`
`limitation of current technology is useful.
`
`Electrical circuits can’t sense light directly; the light as photons must first be converted to electrons. That’s the job of
`
`the photodiode which absorbs the light and converts each photon into one electron and stores those electrons as you
`
`would store rain drops in a bucket. Like a heavy rain, the brighter the image, the more photons that land on the sensor
`
`over a given period, the fuller the photodiode buckets become.
`
`The process of converting photons to electrons is perfect; every photon that strikes the photodiode is converted into a
`
`electron by the photodiode structure. The problem is that not every photon that strikes also strikes the photosensitive
`
`area, the photosensor, because the photodiode is just a component of the sensor and so is smaller than the physical size
`
`of each pixel. The ratio of the number of photons hitting the sensor to the number actually converted to detected
`
`electrons is known as quantum efficiency (QE). QE is dependent on the wavelength (i.e. the color) of the light. The
`
`closer to 100 percent, the stronger the electrical signal, resulting in a more vivid image.
`
`QE is an important aspect of sensitivity, which is an important metric of camera performance. In the context of sensors,
`
`sensitivity measures size of the output signal for a known number of input photons. Intuitively, more sensitivity is better.
`
`Raytheon2028-0002
`
`
`
`noise floor also has to be considered. The noise floor is the minimum detectable signal of the photodiode. The same
`
`way it’s hard to see the change in water level caused by a single drop of rain, small changes in signal are hard to
`
`accurately measure. Noise is also present in film cameras, and noisy images are usually termed “grainy”. This noise
`
`floor is a critical performance parameter for image sensors, as it determines how small a signal (or how dim an image)
`
`can be before it’s indistinguishable from random measurement error.
`
`Engineers use the ratio of the measured signal to the noise floor (the signal to noise ratio, or SNR) as one of the most
`
`critical sensor performance metrics. The SNR level relates well to the overall image quality. An image with a low SNR
`
`will look grainy, and an image with a high SNR will look good.
`
`This, however, only considers the image sensor pixel array alone. In reality, the pixels are connected to a good deal of
`
`supporting circuitry by layers of metal wiring on top of the silicon surface. Recall that light is shining downwards onto
`
`the silicon surface, and must pass though these metal routing layers first. Since metal is both opaque and reflective, it
`
`creates a huge problem in pixel design. The wires have to be kept out of the light path, or they will cast shadows or
`
`reflect the light away from the photodiode, negatively impacting image quality; i.e. lower the sensors QE and sensitivity
`
`as explained above.
`
`Any light striking the wires is not just directed away from the right photodiode, but often reflected into another
`
`photodiode, in a phenomenon called “crosstalk” (refer to FSI pixel shown in Figure 1).
`
`Crosstalk creates blurry images by causing the light from a small object (destined for only one or two pixels) to be spread
`
`out into neighboring pixels. Since photons aren’t being collected in the right place, the resulting electrical signal at the
`
`target pixels is smaller than it could be, therefore reducing the SNR of the image. This is especially important in low‐light
`
`images, where there aren’t many photons to begin with. Crosstalk also reduces color quality, as the photons must pass
`
`through color filters on their way to the photodiode. Adjacent pixels possess different colored filters, and crosstalk
`
`distorts the sensor’s color perception. Clearly sharp, low‐noise images and good color are the desirable features in
`
`photographs, thus eliminating crosstalk is an important element.
`
`
`
`Raytheon2028-0003
`
`
`
`
`Figure 1: FSI and BSI crosstalk comparison
`
`
`As Figure 1 illustrates, getting light to the pixel is quite difficult, and the trend toward shrinking pixels isn’t helping.
`
`Although pixels are getting smaller, the metal layer stack‐up isn’t thinning as rapidly. The result is that pixels now
`
`resemble a tall glass, rather than a bucket, commonly referred to as the “pixel straw”.
`
`Since light cannot pass through the metal walls, it’s very hard to get the light where you want it unless it’s perfectly
`
`aimed. Pixel designers go to great lengths to achieve this, including adding tiny lenses (micro lenses) to the top of each
`
`pixel to direct light towards the photodiode and away from the metal layers. More exotic techniques include reducing
`
`the number of metal layers in the pixel array, and innovative circuit designs to minimize the number of metal wires
`
`required. But the problem remains; how do you get light past all of the metal layers while continuing to shrink pixel
`
`size?
`
`Do tall pixels and crosstalk affect the camera design? Yes, and profoundly so, with the greatest impact being on the lens
`
`design. The best way to get light through the pixel straw is to send it straight through, which means the lens has to aim
`
`all rays of light towards each pixel at perfect right angles to the sensor surface. Doing so requires very tall lenses, which
`
`simply isn’t possible in the small space allowed for mobile cameras.
`
`Practical mobile phone lenses produce images where the rays of light form relatively flat cones. The industry term
`
`“chief ray angle” (CRA) defines the cone angles that the light passing through the center of the lens will follow. Mobile
`
`phone lenses typically have CRAs ranging from 20 to 30 degrees. Making the lens thinner (shorter) requires squishing
`
`the cone, which further increases the CRA. Twenty or thirty degrees might not seem so large, but compared to most
`
`DSC lenses where the cone angles are 5 degrees or less, it is significant. Sensor designers can compensate to some
`
`extent by adjusting the positions of the micro lenses to better align with the actual paths the light follows to the
`
`photodiode. This matching isn’t perfect, since light rays destined for a particular pixel can pass though different parts of
`
`the lens and at different angles. The sensor also ends up tuned to a very specific lens design, limiting the selection of
`
`lenses a manufacturer can use, or causing problems if the lens properties change, as in the case of autofocus or zoom
`
`lenses. Mismatches between the sensor and lens can cause color shifts across the image, darkened corners and other
`
`image artifacts. The ideal sensor would not be so selective, accepting light from a range of angles and lens designs
`
`without suffering crosstalk or reductions in sensitivity.
`
`Backside Illumination (OmniBSI) Technology
`As previously described, FSI pixels have been shown to suffer many disadvantages – low QE, low sensitivity, high
`
`crosstalk and high sensitivity to CRA mismatch. So what if there was a way to get the light to the photodiode through
`
`the backside of the silicon? By avoiding the metal layers, one could avoid all the drawbacks associated with FSI pixels.
`
`Raytheon2028-0004
`
`
`
`This is the essence of OmniBSI technology. As discussed, an image sensor’s backside is the bottom surface of previously
`
`unused silicon. OmniVision has developed a technology for reducing the thickness of the silicon wafer without
`
`compromising the image sensor’s strength or flatness.
`
`Manufacturing CMOS Image Sensors
`Image sensors are made in the same silicon fabrication process as other CMOS semiconductors as reflected in Figure 2.
`
`Devices such as transistors and photodiodes are defined and created in a pure silicon crystal wafer. Three to six layers of
`
`metal wires (aluminum or copper) are deposited on top of the silicon to connect the silicon devices together. The image
`
`sensor process involves a few more steps to create the color filters and micro lenses that enable the sensor to see color
`
`and to improve image quality.
`
`Figure 2: Manufacturing FSI pixels
`
`
`
`Although the same initial process steps (fabricating transistors and depositing metal layers) are used for both FSI and BSI
`technologies, during manufacturing of OmniBSI sensors, the process takes a radical turn ─ literally. Once the metal
`
`layers are deposited, the silicon wafer is flipped over and ground thinner than a strand of human hair. This is a crucial
`step and is key to the performance of OmniBSI technology. The process of grinding a 200 mm to 300 mm diameter
`
`silicon wafer without causing it to shatter is extremely challenging. OmniVision and its manufacturing partner, TSMC,
`
`worked closely together to develop the necessary technology that allows grinding the wafer to the levels required for
`
`high‐volume production.
`
`Raytheon2028-0005
`
`
`
`
`
`Figure 2: Manufacturing BSI pixels, showing the flip process step
`
`Once the wafer is thinned, the color filters and micro lenses are deposited on the backside. The wafer is then tested,
`
`diced up, packaged, and is now ready for integration into camera modules.
`
`Conceptually, the process is very simple; however the practical limitations have prevented previous BSI technologies
`
`from being adopted in all but the most costly and high performance applications, such as astronomy and scientific
`
`cameras.
`
`By making the silicon “backside” many times thinner than a strand of human hair, light can now pass easily though to
`
`the photodiodes. So easily in fact, that the quantum efficiency is significantly higher than any FSI design could ever be
`
`(and remember ‐ higher QE is a good thing). With no metal layers between the lens and photodiode, many more
`
`photons are collected. Between the increased efficiency and lack of obstructing metal layers, OmniBSI technology
`
`provides a wide range of performance advantages.
`
`Performance Advantages of OmniBSI Sensors
`OmniBSI technology offers a multitude of advantages over conventional FSI designs, including higher sensitivity, smaller
`
`pixel size, reduced crosstalk and greater CRA tolerance. OmniBSI technology offers almost double the sensitivity over a
`
`similarly sized FSI pixel. Since there is less material through which photons must pass, fewer are absorbed or reflected.
`
`The lack of obstructing metals means more photons reach their target, or photodiodes. This combines to produce
`
`greater sensitivity over FSI pixels designs. Consumers enjoy brighter, less noisy images, clearer photographs.
`
`OmniBSI pixels can also be made smaller than FSI pixels. In FSI pixels, the photodiode and metal layers compete for the
`
`same space. With OmniBSI technology, there is no competition: metal wiring can be placed right on top of the
`
`photodiode. This allows the pixel components to be packed together more tightly, making for smaller and lower cost
`
`image sensors.
`
`Crosstalk is also greatly reduced in OmniBSI sensors. Again, without the shiny metal layers redirecting photons, more
`
`pixels end up where they’re targeted. Images become sharper and the color reproduction is truer. With truer color
`
`comes lower noise, since the process of correcting color introduces noise into the images.
`
`Raytheon2028-0006
`
`
`
`Finally, without metal layers forming a “pixel straw”, OmniBSI pixels impose fewer restrictions on lens CRA matching.
`
`The following section describes in more detail how wider CRA flexibility provides lens and module designers with greater
`
`freedom and flexibility.
`
`Advantages of OmniBSI Sensors for Lens and Module Design
`OmniBSI technology provides camera module and lens designers three new options over FSI pixel designs:
`1. Better support for larger CRAs, enabling thinner modules
`2. Better support for larger CRAs, enabling lenses with larger apertures (faster lenses)
`3. Support for variations in CRAs, enabling high‐quality zoom lens designs
`
`These advantages are illustrated in
`
`, and discussed below.
`
`
`
`Figure 3: The wider optical path of OmniBSI enables new camera designs
`
`OmniBSI Sensors for Thinner Modules
`
`
`
`Without metal layers shadowing the photodiode, OmniBSI pixels can accept light at a wider range of angles. The most
`
`immediate advantage in camera module design is the allowance of the lens to be placed closer to the image sensor,
`
`therefore providing a thinner camera module. By allowing a wider range of CRAs, OmniBSI technology allows lens
`
`designers to make their designs more extreme, enabling thinner lens and modules, without sacrificing image quality.
`
`Raytheon2028-0007
`
`
`
`OmniBSI Sensors for Large Aperture Lenses
`
`Good low‐light sensitivity is a crucial requirement for mobile cameras, since they rarely have a powerful flash and are
`
`used in a variety of challenging lighting environments. One promising solution is to use lenses with larger apertures
`
`(also referred to as faster lenses or smaller f‐numbers), such as f/2.0 instead of the conventional f/2.8. The problem is
`
`that large‐aperture lenses are also larger in diameter, and force the camera module to be taller to maintain the same
`
`CRA profile. This directly contradicts the trend for smaller camera modules.
`
`By allowing lenses with larger CRAs, OmniBSI technology enables lens designers to place large‐aperture lenses closer to
`
`the sensor, making the module thinner than FSI pixels allow. Besides its inherently larger QE and sensitivity, OmniBSI
`
`technology can further improve low‐light performance by enabling the effective use of larger aperture lenses.
`
`OmniBSI Sensors for Zoom Lenses
`
`Zoom lenses are extremely popular in nearly all consumer products, with the exception of mobile devices. One
`
`significant optical design problem is that the lens CRA changes with the zoom position. As previously discussed, FSI
`
`sensors have to be carefully matched to the lens CRA curve. A close match is impossible if the curve changes. With
`
`OmniBSI technology, lens CRAs don’t have to be as closely matched, giving lens designers the optical design freedom
`
`they need to make high‐quality zoom lenses.
`
`OmniBSI Sensor Compatibility with Module Design and Manufacturing
`
`With its radical new manufacturing process, one might expect OmniBSI technology to create challenges in module
`
`design or manufacturing. In fact, OmniBSI sensors are completely compatible with all existing module design rules,
`
`handling procedures, and manufacturing processes. OmniBSI sensors can be made as thick or thin as any FSI device, and
`
`can be supplied for either chip‐on‐board (COB) wire bonding or in a chip‐scale package (CSP). OmniBSI technology
`
`provides module designers with many new degrees of freedom, without incurring any module design or manufacturing
`
`disadvantages.
`
`Conclusion
`
`With its revolutionary OmniBSI technology, OmniVision delivers some of the highest quality image sensors in the world.
`
`OmniBSI technology enables higher sensitivity, smaller pixels, better image quality, thinner camera modules, faster
`
`lenses and greater camera module design freedom than ever before, making it truly a giant step forward for consumer
`
`imaging. Consumers will appreciate higher resolution images that are brighter, more colorful and less noisy than FSI
`
`pixels could ever produce.
`
`
`
`Author: Michael Okincha, Senior Staff Technical Product Manager
`
`
`
`Raytheon2028-0008