ige over many
r gate area, the
be increased.
ices have been
s can easily be
called a double
lly high charge
emperature and
£5
Dynamic range: 79 dB | S
00 100000
cket
'acteristics and
ensor with an
Reference [18]
the slow-scan
of light. Under
lows statistical
(4)
already 1.4%.
els from being
large using the
e substantially
nch capacitors
recently, it can
/ briefly in the
and red part of
1etrate through
e sensor from
ed with a thin
ght is incident
22-24 1995
[24]. In this way, the sensitivity of CCD image sensors can be extended into the blue and near-UV spectral region, with a quantum
efficiency of typically 20%.
The charge-transfer-efficiency, i.e. the average fraction of actually transported charge in one charge-transfer operation, can be as
high as 99.99996% [25]. Typical values for modern CCDs are of the order of 99.999%, even for large-area CCDs [26]. After the
transfer through 512+512=1024 pixels, requiring 1024x3 transfer operations in a three-phase CCD, a charge signal has dropped to
97% of its original value. Such a CTE is sufficient for video applications. For large-area CCDs, however, it should be appreciably
higher. As an example, consider a scientific four-phase CCD with 4kx4k pixels. The signal charge at an extreme corner has to
undergo (4k+4k) x 4 = 32768 charge transfers, leading to a signal that is reduced to 67%. It is clear, therefore, that for large-area
CCD image sensors CTEs of 99.9999% or better are required.
Another limitation of large-area CCDs is the readout time. The highest achieved readout rate in a silicon CCD structure is 325
MHz [52]. While standard video CCDs are read out at frequencies of only 10-14 MHz, it is already very difficult to achieve
readout-rates of 72 MHz as required by the future HDTV standards. A simple solution to this problem is the introduction of more
than one output stage, which can be read out in parallel, so called multi-tap readouts. For HDTV image sensors, for example, it is
customary to use two 36 MHz outputs, as described for example in reference [12]. Large-area CCDs are offered with four output
stages, one at each corner, or more. In this way, it is not difficult to obtain combined pixel readout rates of several 100 MHz. To
obtain an idea of the frame rates achievable with high-speed solid-state imaging, consider a 128x128 pixel CCD with 16 taps, each
with a readout rate of 32 MHz. Assuming that the storage problem can be solved, an impressive frame rate of 31 kHz would result.
Although high-speed imaging systems with frame rates exceeding 10 kHz are commercially available [27], in ultra-high-speed
camera applications electronic imaging lags far behind the capabilities of cameras using photographic film : Commercially
available film cameras exist that are capable of acquiring 25 million frames per second [28].
A last property to be mentioned is the dark current, i.e. the thermal generation of charge carriers, filling the pixels with unwanted
charge, even in the absence of a light source. Modern CCD technology is capable of producing image sensors with a dark current at
room temperature of less than 1 nA per cm? , in best cases even as low as a few 10 pA per cm”. Using a special clocking technique
called multi-phase-pinning (MPP), it becomes possible to reduce the dark current even further, to values of below 10 pA per cm“,
although at the cost of a more complicated CCD fabrication process with additional steps. As mentioned above, reduction of the
temperature by about 8K halves the dark current.
5. ELECTRONIC PHOTOGRAPHY — PROFESSIONAL AND CONSUMER APPLICATIONS
With the development of the first solid-state image sensors used for the acquisition of still images, the expectations were raised
that these new types of image sensing devices might lead to a replacement for traditional photographic film, to a new field called
electronic photography. It was clear from the beginning that such a replacement could only occur if two conditions are met: Firstly,
performance of electronic photography should come close to the film-based one, and, secondly, the price must be right!
Traditional film has two distinct advantages over an electronic solution: Photographic film is a highly sensitive photodetector with
large dynamic range and high resolution, and it is at the same time the long-term, high-density storage medium for the picture.
Film material can be produced reliably in large sizes and at very low cost, using continuous fabrication processes that require no
expensive geometric patterning ("pixel formation") at all. Electronic photography, on the other hand, needs more components: a
solid-state image sensor, a large-capacity storage device, high-speed data links connecting the two and the outside world for picture
viewing, processing, hard-copying and archiving. For a practical system, digital compression and color processing electronics are
required as well, to reduce the tremendous amount of image information to a more manageable level. As discussed in Section 2, the
price of image sensors and RAM chips depends primarily on the area of silicon required. With the shrinking of the pixel size and
the RAM storage cell size, as illustrated in Figs. 2 and 3, it was just a matter of time until the prices of the components dropped to
a level where electronic photography became attractive. After a dead-start by Sony in the early 80's with their Mavica camera
system [11] offering insufficient picture quality at an excessive price, the recent surge of electronic photography products le
experts to declare that "the waiting is over for digital still cameras" [29]. :
The main consideration in electronic imaging for a long time has been resolution at an acceptable cost. It is difficult to compare the
resolution of film and solid state image sensors because the shapes of the modulation transfer functions (MTF) are so different: An
image sensor system with pixel period p shows an oscillatory MTF with a first zero oy in the spatial frequency domain given by
2x
=> (5)
P
A typical film, on the other hand, has a long-tailed MTF, extending to high spatial high frequencies with low contrast, without any
zeroes [30]. A simple but crude comparison of the resolution can be done by using the spatial frequency at which the MTF has
dropped to 50%. Using this measure, it can be estimated that the resolution of conventional 24x36 mm? color negative film with an
ISO speed of 400 corresponds to about 1250x1900 colour pixels [30]. It is concluded that an electronic photography system that
comes close to the performance of photographic film should at least offer this number of pixels. Traditional video sensors with just
768hx576v pixels fall definitely short of this requirement. Nevertheless, first electronic photography products based on color video
CCDs, such as Apple's QuickTake (manufactured by Kodak) [29], have already found practical applications in desktop publishing.
Its performance is illustrated in Fig. 6, a reproduction of a busy laboratory scene in which the test setup for smart image sensors is
shown, see Section 6. The camera is based on Kodak's KAF-0400L CCD, but using only a reduced number of pixels of
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995
