COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002			REQUIREMENTS FOR IMAGE CAMERAS
C00009 00003			SOLID STATE IMAGE SENSORS
C00015 ENDMK
C⊗;
		REQUIREMENTS FOR IMAGE CAMERAS



This sets down our requirements for a camera for our industrial automation
research.  The requirements center on the following attributes:
As usual, we find that there are system requirments and device requirements.



			DEVICE REQUIREMENTS


RESOLUTION
       at least 200x200. 
BANDWIDTH
       compatible with Unibus rates, 1 sample per 600 ns. 
SIGNAL/NOISE
       as high as possible.   DI/I > 32/1 over an  intensity range of
       200.    In  usual  tv  terminology,  this  amounts  to  6400/1
       signal/noise, but the requirement is much weaker if the sensor
       noise characteristic is  log.  Our concern is  that the signal
       variation  from element  to element  be small compared  to the
       noise.  We  define the uniformity in  this way: for a  uniform
       white  background and  defocussed  image, measure  the  sum of
       (f(x)-f(x-1))↑2.  This has the  value 2*ε↑2, twice the  square
       of the standard deviation  of such differences.  This  will be
       our definition of noise, sqrt(sum(f(x)-f(x-1))↑2)/1.4.
DYNAMIC RANGE		
       We define dynamic range as the range of light level over which
       the  signal/noise  is  greater  than  our specification.   For
       vidicons, the  dynamic range as  defined here  is very  small,
       because signal/noise is so poor at low light level (typically,
       dynamic range is 10/1 for vidicons). 
UNIFORMITY OF RESPONSE
       Element to element variation should be small enough to satisfy
       the definition  of signal to noise.   Smooth variations of the
       uniformity are not  critical, but should  be under 10%.   This
       could be  generalized as a limit on  the amplitudes of spatial
       frequencies of all scales. 
MODULATION TRANSFER FUNCTION
       Ideally, if  there is an  illumination difference  between one
       cell  and the  next adjacent cell  then the  signal difference
       between  the  two  cells   should  be  proportional  to   that
       difference. 
        ..........				....
	         .				    .
		 .				     .
	         .				      .
		 .				       .
		 ..........				......
       That is, the  response should look  like the left  figure, not
       the  right.  For most  tv cameras, the mtf  is only 10% at  the
       quoted resolution of the camera. 
LAG
       The is  the time  equivalent of  modulation transfer  function
       requirement.  We  want the signal vs time to accurately follow
       the illumination.  If the signal goes from black to white, the
       signal  should go  from black  to white  at the  next reading.
       Typically,  for  vidicons, this  figure  is 10%  at  the third
       field.  We would like better. 
SIZE
       The  cameras  should  be as  small  as  possible.    Only  the
       electronics for low  signal level should be in the camera.  We
       will want a hand-held camera, and for it, size is crucial, but
       it is important for others also. 
SENSITIVITY
       The sensitivity  requirement is determined  by depth  of field
       and  resolution requirements.   For  our purposes,  that means
       that the device should be  as sensitive as a silicon  vidicon,
       which is about the sensitivity of a solid state device. 
RETAINED IMAGES
       The intensity of retained images is to  be less than the noise
       level. 
DARK SIGNAL
       Only the  spatial/time variation of  the dark  signal matters.
       That should be less than the noise level. 
BLOOMING
       There should  be no blooming.   We  expect light overloads  as
       large as  a factor of 100 to 1000.   The signal need no longer
       be  proportional to  light  intensity  (ie  clipped)  but  the
       clipping should be confined to the areas of high intensity. 




			SYSTEM REQUIREMENTS


COLOR 
       is very important.   We call  it desirable now, because  we do
       not see  an acceptable device.  If  we could buy an acceptable
       device at an acceptable cost, we would call it necessary. 
STEREO
       is very important.  This is a system requirement, which we can
       initially separate from camera requirements. 
Note that color and stereo imply 3 solid state sensors for color (or
degraded resolution spatially (color stripe) or time (switching filters))
and 2 cameras for stereo (or a switching system with two optical paths).
		SOLID STATE IMAGE SENSORS


Fairchild
Frank Bauer (sales), Gil Amelio, Rudy Dyck
  spoke to Bauer


AREA SENSORS
  100x100
  244x190	May or June
  488x382	4th quarter; not known whether it will be a commercial product
		if a product, then Jan 1976

LINEAR SENSORS
  512
  1728  CCD121	13u centers x 17u


The 100x100 has 70/1 dynamic range and 3x10↑-3uJ/sqcm for 2854 deg Tungsten.
The commercial version of the 244x190 camera won't have floating gate
multi-stage amplifier, but special version might be available.  Both would
be on the chip, but only the simpler amplifier would be tested and bonded.

For the 1728 linear sensor, it will be possible to have electrical inputs
to do some signal processing (trivial) on chip.

R.H.Dyck  send a copy of the report to him
SIGNAL/NOISE ref Sept Edinburgh, conf on ccd
The DFGA, distributed gate floating amplifier has a noise equivalent level
of less than 20 electrons, probably about 10 electrons.  To a first approx
noise is constant, independent of signal level.  The device is aimed at
low light level, and in fact, if I understood his final analysis of the
circuit, has lower noise for lower signals.
BLOOMING
They are designing in column anti-blooming.  To do element anti-blooming
requires a structure for each element.  They are still working on anti-
blooming.  How bad is it?  Just as bad as you can imagine it to be.  Charge
spills over into a column, then into the output.
MTF is very good.  If we put a spot on one element with the rest of the
tube black, the neighbors read 5% or less, often 1%.  The MTF at the
Nyquist limit for square waves is 80%-90%.  There is some crosstalk in the
infrared, a property of Si being transparent.  OK out to .7u.  The MTF
is a property of high transfer efficiency.  Nε is about .1 from an element
to readout.
NO LAG
UNIFORMITY
Some smooth non-uniformity.  Element to element variation averages 1%.
They find variations up to 5% or so.  Question of mask perfection, cleanliness.
SENSITIVITY
That of bulk silicon.  Quantum efficiency of 50%, with 50% opaque real estate.
The quantum efficiency will go up about a factor of two.

GE
Fred Sachs

CID (Charge Injection Devices)
operating cameras
  100x100 $4950 sequential read, internal clock controllable
    500/1 dynamic range
  132x192 $6500 sequential read, internal clock controllable
  244x188 $6500 tv compatible; odd shape cells 1.4mil x 2.4 mil
    1/30 second, rate not changeable, 45800 samples in 1/30 second
    no blooming up to 10↑4 overload; 400/1 dynamic range (smaller cell,
    smaller charge)
coming August
  244x248, 340mil x 450 mil chip size, 1.4mil x 1.8 mil cell size
    more flexibility, non-destructive readout; use in shuttered
    mode or to integrate.  No blooming

signal/noise
  500/1 for 100x100, slightly less for 244x188 (smaller cell, smaller charge)
  gain by cooling, by integrating, by lower bandwidth
  improvements in electronics??

random access is possible.  Need to change shift registers for decoders.
MTF is very good.
uniformity is very good he thinks.  No detailed measurements
Dark uniformity is not so good.


RCA
Roy Minet
RCA sales 948-8996  Bill Dyall
low blooming, better than Si target vidicon
they aren't able to let us try one out.  They are only on
a sales basis.
Send him our requirements.