 |
 |
 |
 |
 |
 |
 |
|
|
|
||||||
|
||||||||
|
||||||||
 |
||||||||||||||||||
|
||||||||||||||||||
|
Image Sensors 2008 San Diego, CA Image Sensors 2008 is a small conference focused on an important segment of semiconductor technology. The attendance was 100. Consistent with other Intertech Pira events the presentations are of high quality from important players. This not only makes for a good conference but for networking among the attendees. At this event many seemed to know each other in this specialized market. Intertech Pira has brought another high quality event to the needs of this marketplace.
Smaller and Better Quality With the migration to digital, imaging sensing has undergone 10 years of continuous change. In only the last few years cameras in cell phones has driven changes in imaging technology. Key areas highlighted in this conference include the following. Wafer Level camera Cameras fabricated as a complete module on the scale of the CMOS sensor on a wafer EDoF Enhanced Depth of Field. This enables closer objects to be in focus. It can enable arms length Backside Illumination Silicon fabrication technology to allow the CMOS sensor to be used in an “upside configuration” for Small sensors Continued reduction in size of sensors to 1 micro pitch Pricing Driving to CMOS sensors at $1/Mp
Presentations Advasense – Arie Gavriely The mobile market continues to be dominated by the pixel count – this is the main consumer feature by which camera phones are evaluated. Currently 1.75 micron pixels are the mainstream in 3Mp cameras. Small pixel sizes of 1.4 microns are here and moving to 1 micron. The image quality challenges line in: low light sensitivity and improvements in Optical Efficiency and the improvement in CFA patterns. Overall efforts focusing on reducing the sources of noise in the image while making the pixel size smaller. Advasense has Deep Photodiode technology which allows it to improve the Full Well Capacity. This implements a deeper photodiode. It is claimed that in Red QE it will be possible for Advasense to have a .9 micron sensor with the QE of a normal 1.75 micron pixel sensor. Advasense has a technology called FCP – Feedback Controlled Pixel. This has the advantage of imaging without full charge transfer. This results in no image lag. The value of FCP is in image stabilization. The pixel array serves as the memory during the stabilization process. A series of short images are captured, aligned and summed up. AltaSens – Lester Kozlowski AtlaSense is seeking to develop CMOS system-on-chip sensors for HDTV and beyond. Lester began by giving credit to the early work of Peter Noble who developed the concepts behind CMOS imaging. (Peter J. W. Noble, Self-Scanned Silicon Image Detector Arrays. ED-15, IEEE. pp. 202–209, Apr. 1968.) AltaSense projects a target for the noise floor in imaging at 1 e by 2010. Already system-on-chip sensor are surfacing in the latest cameras that have noise values around 5 e. The current state of the art is 2.5 to 5 e noise in 4T pixels. The key factor in achieving low noise is suppressing kTC noise. AltaSens projects it will be possible by 2010 to achieve 1e at ISO of 2000 in single frame cameras. Aptina Imaging – Mansour Behrooz By 2015 it is projected that cameras will be everywhere. Camera linkage to YouTube, Facebook, Myspace and Flickr is becoming increasingly important. The major disruptive imaging sensing technology is the CMOS sensor. By 2010 it is projected that in all market segments it will have superior performance. The market which is driving integration is mobile. This is pushing pixel pitch to 1 micron. Canesta – Cyrus Bamji Canesta believes there is one component missing from today’s imaging technology suite – the distance to what is being imaged. They have implemented an IR time-of-flight sensor on CMOS which has ready integration into products. This allows for the detection of gestures in front of a camera. In fact, the Canesta technology was used in at least one of the gesture demos at CEATEC. There are three problems with time of flight technology. Sun and ambient light which they address with their SunShield technology; Range Ambiguity and the FarSight technology handles and then latency and motion blur and their approach is called ClearShot. When Canesta was asked – at CEATEC the gesture demos suffered with slow response time and large hand movement they responded – we are just learning how to use the technology. It was stated that what was not shown on the floor is better. DxO Labs – Stephane Laveau In order to keep up with the rapid development of mega pixel sensors a new approach is needed to the ISP – image processor. DxO cites that the ISP requirements are 10X to 100X beyond that of CPUs or DSPs. DxO has a programmable IRIS core which is SIMD RTL dedicated to image processing. The design is based on the assumption that the same operations are done on all pixels and a given pixel involves only its neighborhood. The footprint of this processor is 250Gops/s in a 3 sq mm chip including RAM. The processing can go to 400Gops/sec. e2v – Gareth Powell & Sophie Caranhac e2v has designed image sensors for 30 years. It provides high quality sensors for industrial and professional applications. The applications include: consumer electronics products, machine vision, bar code scanners and medical imaging. In the automotive market they have developed sensors for NIR spectroscopy for fuel quality sensing and imaging for occupancy detection for air bag deployment, lane departure warning and pedestrian detection. To improve sensitivity e2v is using micro lenses to improve the QE and electron multiplying CCD design. Image examples were show at .001 and .0001 lux. Another area of focus is in high dynamic range sensors. One technique uses in-pixel double exposure to increase the dynamic range. e2v’s technology is claimed to provide 100dB linear dynamic range, will support true color and has a low noise based on kTC noise suppression. e2v is also supporting backside illumination. Eastman Kodak – Fas Mosleh Kodak has developed TrueSense color filter technology which enables higher low light sensitivity, increased capture speed and faster sampling speed. The High-ISO technology uses the color pixels as chrominance signals and thus improves the signal level. This is compared to the Bayer CFA pattern which uses the color pixels at all light levels for color imaging. The claimed improvement is 1 – 2 f stops. Imaging to .5 Lux was shown. Kodak forecasts that camera phones will be able to move up scale in application – as they call – capture important events – by 2010. This is a direct result of the improvement in image quality. Kodak is the supplier of the 37 Mp image sensor in the Leica S-system camera. Heptagon – Jyrki Saarihnen Heptagon has developed wafer-level lenses. A master wafer level lens is created using step and repeat. This master enables lenses which align with the sensors when the lenses and sensors are bonded. Wafers are applied between the sensor and the lenses. This allows for the proper optical spacing. It is possible to use multiple lenses to create doublets and triplet lenses. CIF modules have been implemented which are 2.7mm thick and 1.7mm on a side. There are 2000 lenses per wafer for a 3.6 micron pixel. The MTF yield is >98%. Heptagon has no-focus VGA lens modules. The singlet module is 1.7mm thick and the doublet is 2.1mm. It is claimed that the image quality is on a par with glass or injection molded lenses. A 2Mp lens module has been developed which is 3.3mm thick and 4.2mm on a side. There is no focus and no adjustment of the lens assembly required. Nokia – Saku Hieta The market opportunity for cell phones is in emerging markets. Cell phone cameras play an important role but the technologies must be at the lowest cost. Nokia cited the price point of $1/Mp for the camera cost. Nokia market research indicates that by 2010 that approximately 2/3 of the emerging market will seek a cell phone with a camera. Another factor is that replacement phones will be key in camera growth. Nokia sees the following as requirements for cameras which meet these market needs: Wafer level cameras and EDoF camera modules. They feel that these two technologies will rapidly change the playing field of cell phone cameras. OmniVision – Michael Okincha The industry is being driven to smaller pixels for three reasons: cost, higher resolution and camera miniaturization. The impact of this is that the cameras need more image processing. To support these trends a new approach is needed to make the sensors better suited to the applications. This is back side illumination. The technique is to fabricate the CMOS sensor with the usual metal layers and processes. But when the sensor semiconductor fab has been completed the wafer is turned over and ground to the level of the back side of the pixel. On top of this is put the color filter and lens. The result is a much more efficient detector. The advantages include wider angle of light ray acceptance, large aperture ratio, lower cross talk, no limits on the number of metal routing layers and higher yield and lower cost. PerkinElmer – Rudi Blondia Running counter to many of the earlier presentations PerkinElmer it was stated that to get best performance in low light is to have a Xenon flash lamp in each camera. The basis for this is: 100X faster shutter speed with a flash lamp, less blur, high quality images, and natural daylight. Sarnoff Corp. – John Tower One of the major advantages of Backside Illumination is quantum efficiency (QE). Sarnoff has been working for many years to use this process in CMOS sensors for space applications. With a 21 micron backside, a 8 micron pixel QE above 80% in the visible spectrum. Sarnoff has a method of fabrication of the backside called Ultra-Sense SOI. It is based on a SOI wafer which is claimed to scale to 12” wafers. The advantages include 1 – 2 electron RMS noise, a dynamic range of 14 bits The current roadmap calls for a 2Mp chip in January 2009 with scaling to 50 – 100 Mp chips. A 4 Mp chips in design which will operate at 480 f/s. Scalado – Sami Niemi Scalado has a camera engine which works in JPEG that allows for the processing of images at the pixel level at very high rates. This is done with tagging of the image as it is collected. With this processing engine it is possible to accomplish: zero shutter lag, instant zoom and pan; instant shot to shot and burst mode. Technical details were not provided but the demos were impressive. These included: CAPS AutoRama – fully automatic panorama capturing to images with 12Mp and done with very low memory requirements. CAPS ClearShot and PhotoArt – image quality enhancements and image effects in real time. CAPS SpeedView and PhotoFlow – very fast viewing, scaling and zooming of images in photo galleries Second Sight – Kelly McClure It is seldom that a technical talk hits ones emotions but this one did. Kelly described the efforts to create a retinal prosthesis to allow the blind to see. The early trials were limited, compared to those with full sight, but to those that cannot even tell light from dark the results are enormously impactful on the patients. Second Sight is in clinical trials with 17 patients who cannot tell light from dark. The patients wear dark glasses with a video camera over the bridge of the glasses. The video signal is processed by a small processor about the size of one’s hand. The signals then are sent to an implant on eye which has 60 electrodes. A video was seen of a patient walking a white line. Kelly described the statements made by patients before the regulatory bodies in the US and Europe. One stated that they could shoot baskets with grandchildren. To those with sight 60 electrodes, as the resolution of the imaging, seems crude to those who have had no sight it is incredible. It is expected that regulatory approval could come as early at late 2010. Second Sight is working on higher resolution sensors and implants. Impressive. Spiral Gateway - Graham Townsend Spiral Gateway is focused on the camera image processing. Given the success of camera phones and the continued escalation of Mp they claim that a new architecture is needed to process images at the pixel level. Spiral Gateway has a RICA (reconfigurable instruction cell array). This consists of an array of cells which can be connected together to complete with ASIC based implementations of an image processor. The resulting ISP is claimed to process ASIC netlists on-demand in real time. An array of cells can be connected together to execute C code directly. The processing rate is 150Gops/sec. RICA is capable of processing high quality DSC images, HD video camera and viewfinder modes. It has a “glue less” interface with a wide range of COMS sensors. A “Flat single kernel software pipeline” can operate at 92.6Mp/s and a multipass software pipeline can operate at 44Mp/s. Applications are claimed to be: non-standard optics, face identification, gesture recognition and 3D processing. Tassera – Yehundi Dagan It is claimed that by 2011 that 50% of the cameras will be done with Wafer Scale Integration. The requirements of the cell phone industry are: Camera as thin as an IC; Tassera has developed the technology to produce lens which match the scale of the imaging sensors on a wafer. It is claimed that this technology will allow significant new markets where cameras will be embedded in much the same way microprocessors are embedded in product today. This includes toys, cars and elevators. They believe a new era visual processors is possible. Toshiba – Shri Sundaram The proxy for how the area imaging sensor market will evolve is the linear image sensor market. This latter market has been in place since the 1970s. Over these 30 years there has been consolidation, technical improvement and price reduction. In the 1980s a linear sensor costs $5,000 and now in the consumer market a sensor is $79. The areas of evolution of linear sensors include: size of market, applications, products and technical improvements. One can use these trends to forecast what will happen in area sensors. Area sensors in 2007 reached 1B units. The major markets which has drive the technology are mobile phones and computers. Pricing has taken a nose dive with the VGA module ASP now at $2. It is important to note that we are only 10 years into the evolution of area image sensors while linear sensors are over 20 years old. In that time these has been consolidation to two major suppliers. This has not yet happened in the area sensor market – we expect more consolidation. Varioptic – John Barber Varioptic has developed a liquid lens for use in small cameras. This lens is accomplished by the boundary between two liquids where the surface is spherical and forms a lens. There are lenses for 5Mp (6.2mm thick) and 8Mp (8mm thick). The lens will support continuous autofocus (CAF) for video. This enables applications such as bar code readers, biometrics, and OCR. A demonstration was shown how this lens could be used in video compensation for vibration.
WAVE Comments This was an excellent conference to highlight the rapid pace of events to continuously improve the quality and performance of CMOS image sensors. Wafer Scale integration will drive new levels of economics to cameras. The integration of optics at the level of the sensor die size allows for economies of scale in the camera which is already present in the sensor. At present the lenses are simple single elements but it is likely that this technology will mature with the needs for higher resolution. Cameras on the scale of 5+Mp with low noise and all fabricated at the wafer level will provide a major quantum jump in quality, cost reduction and ubiquity. Noise reduction technologies, as reflected by AltaSense system-on-chip implementation will drive new applications for imaging. Presentations showed imaging in light levels as low as ..0001 lux bring a new level to imaging to the casual photographer. There is a limit to the achievable noise levels and the current designs are reaching those limits at 1 e. Cameras everywhere seems like a play on words from the display sound bite – Displays Everywhere. But this all makes sense. Visual modality is about input output and now the imaging technology is achieving a degree of parity with displays. The key to effective use of sensing technology is vision software which interprets and acts on what is seen. |
|||||||||||||||||
| Comments? |
||||||||||||||||||
