eal PSF are
"hey exhibit
understood
y chips” [3].
seeing chip:
ere are also
‚considered
je expected,
s in silicon,
led that also
Xf biological
t sensitivity
ive functio-
| dimension
on of optical
e-to-contact
segregation,
impressive
> advantage:
Xf biological
a pixel site
area). As a
t planes: an
n plane and
uarantees a
ent parts of
anes can be
for stacking
f a low-cost
re certainly
hips. It can
> part of an
rts. Special
-uniformity
pe, on-chip
nomer, and
-processing
ble scenes,
ns where a
1 electronic
e in digital
re machine
icture. The
recognition
st case, the
ic accuracy
ie output of
t or absent,
mber could
strates that
ub-fields of
24 1995
11
While it is unrealistic for the near future to expect large-area CCD image sensors with even larger numbers of pixels, exceeding
about 8kx8k, qualitative improvements in the fabrication of these sensors can be expected. Less expensive image sensors with
fewer bad pixels will be available in the future. It will still be very difficult to produce flawless image sensors, so that cost-
effective digital imaging systems will have to rely heavily on pixel correction or interpolation mechanisms provided by software.
Research and development work dealing with this problem will have a significant impact.
In the wake of image sensor developments for the vast multimedia mass market of the future, applications will prosper that can
take advantage of the new CCIR or HDTV imagers: Inexpensive image sensors of high quality can be expected. At the beginning of
the next decade, pixel size will have reached a lower limit of about 2-3 um; further miniaturization will not make economical sense
for optical reasons (diffraction and aberrations of the lens systems). CCIR image sensors with 768hx576v pixels will be of the 1/6"
type, i.e. covering an area of 2.5hx1.9v mm?. HDTV image sensors following the European standards will have 2048hx1152v
pixels on an area of about 7hx4v mm?. Once the fabrication problems are mastered and the yield reaches the high values obtained
for today's video image sensors, it is expected that HDTV color image sensors of high quality will be offered for less than $100.
One application area that will profit directly from this development is promising a large practical potential and big business:
Electronic photography. Professional photographers are already making good use of high-quality digital cameras, for which 2kx2k
or 2kx3k color pixels (as in the Kodak M6 CCD) are a minimum requirement. While today's high-resolution cameras with about
5kxSk color pixels are still using scanning line CCDs, future products will employ area-CCDs for single-shot operation. The
situation is different in the consumer market, where digital cameras do not yet quite offer the performance at an acceptable price,
required to replace the traditional film-based photographic cameras. As soon as HDTV color image sensors are available for a few
hundred dollars, electronic camera products will start to invade film-based photography profoundly, the same way as it currently
happens in the professional domain. Such a significant technology shift in the consumer photography market is expected within the
next five to ten years.
While it can be safely predicted that the performance of conventional image sensors will be improved, for example with lower
noise, higher D/R, color reproduction quality, etc., much more momentous progress is expected in the increased functionality of the
image sensors: Novel capabilities with a significant practical impact include pixel-programmable sensitivity, on-chip non-
uniformity and shading correction, variable exposure and timing control, region-of-interest capability, dynamic pixel size and
shape, on-chip image pre-processing which can be carried out for all pixels concurrently, and many others. These advanced types of
image sensors, combined with other developments in optics, applied solid state physics and computer science, see for example
[50], will be the basis of exciting developments for adaptive vision systems, giving practical meaning and real life to the concepts
of "active vision" [51]. Applications in many relevant fields will abound, such as in automatic manufacturing, surveillance and
security, ALV guidance and transportation, medicine, environmental control, optical metrology and digital photogrammetry, etc.
Previously it was expected that adding more and more functionality to an image sensor's pixels would eventually lead to a single-
chip machine vision system, a seeing chip. While specialized applications might indeed see the advent of single-chip smart image
sensors, doubts have recently been cast on the concept of a seeing chip as a complete system. The reason is that the large amount of
processing and functionality required of a single pixel would require 3D (volume) integration, for which standard silicon
technology is not well suited; after all, "vision is difficult", as one of the early proponents of seeing chips put it [3]. Practical
integrated vision systems may therefore be realized as optimized multi-chip solutions, using suitable communications paths and
protocols, which are under active research now.
Electronic imaging has come a long way in the past 25 years since the publication of the first image that was produced with a CCD
image sensor. Thanks to the amazing developments of semiconductor technology, from which image sensing could directly profit, a
wide variety of image sensors is commercially available today, ranging from large-area multi-Mega-pixel CCDs to the first smart
image sensors in industrial applications. It might well be claimed, therefore, that electronic imaging has started to realize a dream:
From pixels to answers.
9. REFERENCES
[1] M.F. Tompsett, G.F. Amelio, W.J. Bertram, RR. Buckley, W.J. McNamara, J.C. Mikkelsen jr. and D.A. Sealer,”Charge-
Coupled Imaging Devices: Experimental Results”, IEEE Trans. Electr. Dev., Vol. 18, 992-996 (1971).
[2] P. Suni, "CCD Technology at Orbit Semiconductor Inc.", International Conference on.Scientific Optical Imaging, Georgetown,
Grand Cayman Islands, Dec. 2-6, 1992.
[3] C. Koch, "Seeing Chips: Analog VLSI Circuits for Computer Vision”, Neural Computation, Vol. 1, 184-200 (1989).
[4] Special Issue on Electronic Imaging, Optonics and Photonics News, Vol. 5, (1994).
[5] E.C. Tam, “Smart electro-optical zoom lens”, Optics Letters, Vol. 17, 369-371 (1992).
[6] E. Hecht, "Optics", 2nd edition, Addison-Wesley (1987).
[7] S.M. Sze (Ed.), "VLSI Technology", McGraw Hill (1983).
[8] H.F. Tseng, J.R. Ambrose and M. Fattahi, “Evolution of the Solid-State Image Sensor”, J. Imaging Sci., 199-205 (1985).
[9] S. Shibahara et.al., "1 GDRAM Cell with Diagonal Bit-Line (DBL) Configuration and Edge Operation MOS (EOS) FET",
Proc. International Electron Devices Meeting IEDM '94, Dec. 11-15, 1994.
[10] B. Bosiers et.al., “A True Progressive Scan 640x480 FT-CCD For Multimedia Applications”, Proc. IEDM'94, Dec.11-15, 1994
[11] Sony Europe GmbH, Hugo-Eckener-Strasse 20, 50829 Koln, Tel. +49-221-5966-6623, FAX +49-221-5966-157.
[12] E. Oda, K. Nagano, T. Tanaka, N. Mutoh and K. Orihara, "A 1920(H)x1035(V) Pixel High-Definition CCD Image Sensor",
IEEE Trans. Solid State Circ., Vol. 24, 711-717 (1989).
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences”, Zurich, March 22-24 1995
