CMOS technology challenges CCD in Europe
In Europe, a number of academic researchers are working on CMOS imaging devices, and a few companies are producing them. The major company providers are the Imaging Division of STMicroelectronics (Corsotphine, Edinburgh, Scotland), which has acquired the former Edinburgh University spinoff VLSI Vision Ltd.
To achieve higher volumes and lower prices, several European companies are producing CMOS imaging sensors using custom techniques.
By Brian Dance,Contributing Editor
In Europe, a number of academic researchers are working on CMOS imaging devices, and a few companies are producing them. The major company providers are the Imaging Division of STMicroelectronics (Corsotphine, Edinburgh, Scotland), which has acquired the former Edinburgh University spinoff VLSI Vision Ltd.; Fillfactory (Mechelen, Belgium), which is using technology developed by the Inter-university Microelectronics Centre (IMEC; Leuven, Belgium); and Philips Semiconductors (Nijmegen, The Netherlands). In addition, the Fraunhofer Institut für Mikroelektronische Schaltungen und Systeme (IMS; Duisberg, Germany) has developed a range of CMOS sensors for a variety of applications.
A few years ago, IMEC developed a CMOS technique to achieve a fill factor of nearly 100% and, therefore, high sensitivity. In this approach, a p-doped layer is formed under the sensitive pixel electronics surface to create a small, but effective, electrostatic potential barrier of up to 100 meV. This barrier can divert the charge carriers formed by the incident light in the semiconductor material toward the collecting junction and away from other junctions. These carriers contribute to the signal current from the collecting junction.
Advances in chip processing enabled IMEC to produce devices with no defective pixels, and the short junction perimeter achieved a low dark current. This n-well pixel technique was used in almost all of the active pixel image sensors recently developed by IMEC for industrial customers, which combined high sensitivity, low noise, and a high fill factor.
The integration of double-sampling circuitry on-chip eliminated the nonuniformities that are usually unavoidable in CMOS technologies. These nonuniformaties increase fixed pattern noise and produce a "snow-like" effect over the whole image. On-chip correction electronics are used for every image sensor column to make sure that no new nonuniformities are introduced. These corrections are executed at the same speed as the image data-transfer rate.
Standard CMOS imagers show considerable degradation after undergoing just a few kilorad of ionising radiation. IMEC has developed radiation-resistant CMOS imagers using a pixel architecture in standard CMOS that can tolerate more than 20 Mrad in silicon from a cobalt-60 source. These imagers are valuable for space applications where small size and low power consumption are mandatory requirements.
Two visual monitoring cameras, each 100 x 60 x 60 mm and weighing 430 g, developed by IMEC and Delft Sensor System/OIP, were deployed on a spacecraft. They provided images during the unfolding of the large solar panels and manoeuvring of the spacecraft thrusters. One of the cameras provided a logarithmic response to achieve a wide dynamic range. Fillfactory has cameras on the Cluster satellite.
IMEC has developed a 35-mm CMOS image sensor with 6.6-Mpixel resolution in collaboration with Scitex (now the Leaf Division of Creoscitex, Herzlia, Israel) for digital photography. This device can be placed in the focal plane of a standard 35-mm camera to capture full-size, high-resolution images. Exceptional image quality, high sensitivity, and low noise are claimed.
IMEC has also designed a fast CMOS 512 x 512 active pixel sensor in a high-speed camera for Vision Research Inc. Applications include automotive crash and safety system testing, production line-failure diagnosis, and medical applications.
The FUGA1000 is a 1024 x 1024-pixel addressable CMOS image sensor from Fillfactory that offers a logarithmic response (see Fig. 1). This response enables a dynamic range of more than 120 dB to be achieved. A camera based on this sensor requires no iris or integration time control, in contrast to conventional linear response sensors. It is claimed that this sensor excels in applications such as machine vision, where information is extracted from an image to control a process. The logarithmic response greatly reduces the constraints imposed by illumination control, whereas random addressing increases the speed by reading only the significant parts of the image and, therefore, processing fewer pixels. The infrared sensitivity permits full illumination by visible light in a factory-floor environment. A four-transistor-pixel architecture is used in this sensor.
The on-chip circuitry includes a 10-bit flash analogue-to-digital (A/D) converter. Its logarithmic response enables use in applications such as automatic welding. Its spectral range is 400 to 1000 nm, with monochrome and colour versions available.
The sensor architecture consists of a pixel array, address decoders, and an output amplifier. The sensor uses a common input data bus for x and y address data, controlled by two data-enable signals. An extra y-decoder has been added for yield purposes. The readout block contains two sample-and-hold functions, one for column use, which makes it possible to freeze a whole line, and one for pixel use. An 8-bit digital-to-analogue (D/A) converter can be used as an offset control for the output amplifier. The data bus is mapped to the lower region of the address bus. The sensor output can externally be routed to the A/D-converter input.
FIGURE 3. In this colour imager, incident light from an object passes through colour filters and onto the CMOS sensor (top). Each pixel receives a certain colour, and its response depends on the intensity of that colour. The output signal is used to reconstruct a coloured image. Colour camera system array feeds the imaging array signals via an input processor to colour and spatial distortion circuitry and then to an encoder that provides the output signals (bottom). (Figures courtesy STMicroelectronics Imaging Division)
IMEC developed a stand-alone network camera for capturing and transmitting live images on the Internet (see Fig. 2). For maximum flexibility and minimum power consumption, field-programmable-gate-array (FPGA) technology was used throughout. The camera used a 1.3-Mpixel CMOS imager from Fillfactory. The FPGA contains the hardware version of the algorithms required to format the raw image data to a standard format and to carry it over the network to a web browser. The system, including the camera, consumes less than 2 W, which is claimed to be a fourfold improvement over current stand-alone embedded Internet camera systems.
Peter Denyer, director of the Advanced Research, Imaging, and Display division at STMicroelectronics, says, "Our CMOS sensors now have lower noise than some CCD imagers, but we are about a factor of two or three away from the best CCDs available today. Our noise limits are twofold. Dark current leakage is one source of noise. We can keep reducing this by process improvements and will continue to do so. The other noise limit is set by pixel reset noise. To reduce this we either use a new reset-noise-free pixel structure or we can allow the noise to be formed and correct for it later."
The architecture of the STMicroelectronics CMOS imaging chip consists of one red, one blue, and two green colour filters per pixel, which may be added in one of the final fabrication steps to provide colour sensitivity in a 640 x 480-pixel image (see Fig. 3).
Alternatively, CMOS devices can be used as near-infrared imagers. The raw data must be converted into a digital signal for processing; this should be done on the chip or as close to the chip as possible. All current STMicroelectronics CMOS imagers contain an A/D converter on-chip. Denyer says that STMicroelectronics imager chips are made by a 0.5-µm process, but this process will soon become 0.35 µm. Pixel size is 7.5 µm, decreasing to 5.6 µm soon. The resolution corresponds to the 20,000 to 300,000 pixels per chip that are integrated into a simple array with readout. The operating voltage is 3.3 V, but devices operating at 2.8 V have been made.
To use a CCD device, a set of integrated circuits must be purchased for use with it at a typical total cost of £6.8 to £8.2, in volume, for the imaging system. A CMOS imager essentially comes complete, including the circuitry on-chip, all for a total typical cost of some £5.5 to £11, depending on the resolution. The cost of CMOS imagers has been falling at around 25% per year, but the rate of decreasing cost is now slowing.
Philips Semiconductors (Nijmegen, The Netherlands) is developing both CMOS and CCD imagers. For CMOS, it initially concentrated on 640 x 490-pixel imagers for the personal-computer/video-camera market. It plans to introduce imagers for mobile phones. It is also investigating larger imagers for digital still camera use.
Philips uses a three-transistor pixel with a 22% fill factor, but it is investigating the use of four-transistor pixels with correlated double sampling for low noise. This sampling reduces sensitivity for a given pixel size, but the use of microlenses above the pixel improves sensitivity. Philips uses a 0.35-µm technology and feels signal-to-noise ratios may be a problem in 0.25-µm imagers.
Tower Semiconductor Ltd. (Migdal Haemek, Israel) claims its colour filter deposition process makes it the first silicon foundry to offer a complete turnkey manufacturing service for CMOS image sensors. The colour filter deposition process uses the same clean-room production equipment as that used for the CMOS image sensor itself. Tower has also developed a lithographic stitching technique that enables the foundry to produce large devices. Tower estimates the current worldwide image sensor market as some £478 million, with a potential to grow annual revenues up to £2.05 billion within the next five years.
Siemens (Regensburg, Germany) researchers, working in collaboration with IMS Duisberg, claim to have developed the first three-dimensional (3-D) object-recognition sensor that is based entirely on semiconductors. Its initial application is the sensing of people within cars to improve safety systems. In particular, it aims at improving the control and safety of airbag-triggering systems. Simple mat or weight sensors in a vehicle seat can determine only the occupant classification and approximate size. However, 3-D imaging enhances these data with positional measurements.
Sensors might also be used to detect when a person other than the car owner, such as a thief, is driving a vehicle. Siemens says that as position measurements gain increased mapping capability and resolution, occupant safety systems will become more intelligent. This Siemens technology will make its product debut in an airbag safety car system in 2003.
The Siemens sensing system uses a technique known as multiple double-short-time integration in which near-infrared, low-power laser pulses are directed through the passenger compartment of the vehicle. A CMOS image sensor detects the reflected radiation and delivers a signal that is processed by circuitry integrated onto the CMOS chip. The system requires only 1 ms to measure the position of 1000 object points to an accuracy of 10 mm and then forms a 3-D image of the objects within the vehicle.
Occupant sensing had previously been achieved using a pulsed laser system that was not robust or fast enough to meet the requirements of the automobile industry. The time it takes for the system to continuously re-acquire data about the exact positions of the occupants must be as close to real time as possible. This is necessary because knowledge of the position of the occupants at all times is required to enable the system to deploy airbags intelligently when a collision occurs. Consequently, the circuit-processing speed and memory are extremely critical.
Another application, developed by Given Imaging Ltd. (Yokneam, Israel) in collaboration with Photobit Corp. (Pasadena, CA, USA), is a camera that can be swallowed to produce high-quality video images of the small intestine (see Vision Systems Design, July 2000, p. 8). It uses a Photobit 256 x 256-pixel, custom-built CMOS imaging device to provide 2 frames/s with an image quality adequate to show the small intestinal villi.
The swallowable capsule contains a light-emitting-diode (LED) white-light source, colour video camera, battery, and radio transmitter with antenna. The LED provides the light intensity needed to diagnose disorders from slight differences in tissue colour. The capsule consumes just 2 mW, which is much lower than that of other CMOS imaging devices.
The images from the capsule camera are transmitted to a recorder on a patient's belt. This transmission technique results in the painless imaging of regions of the small intestine that endoscopes cannot reach. The recorded details are processed by software that produces the images and identifies the location of the capsule when each image was taken.
Lou Hermans, vice president of marketing and sales at Fillfactory, says, "The application of standard CMOS technologies for image sensor development has opened the doors for customised devices. We have several customers that need only a small number of devices, but we can offer a customised sensor at low cost."
Creoscitex Corp. Ltd.
Herzlia B, Israel 46103
B-2800 Mechelen, Belgium
Fraunhofer Institut für Mikroelektronische Schaltungen und Systeme (IMS)
Given Imaging Ltd.
Inter-university Microelectronics Centre (IMEC)
Nijmegen, The Netherlands
Pasadena, CA, 91101-1758 USA
Corsotphine, Edinburgh, Scotland
Migdal Haemek, Israel