The Human eye versus Silicon

Bart Dierickx, Caeleste, update 30sept2010

 

In the technical debate (the good one against the bad ones, or CDD against CMOS, or Silicon against Film) too often the single most important competitor is forgotten: the human eye.  For fundamentalists the Human Eye [2] is the nec-plus-ultra, for contra-revolutionaries, it is one of our weaker parts that deserves an urgent bug fix. For all performance specs related to image rendering, it is the de facto standard.

In het next tables we collected the major spcifications of four competing imaging systems: the eye, the two major flavors of semiconductor imager technology (CCD and CMOS), and chemical film.  We silently generalized all flavors of  CCD (FT, IL, CS, CMD) and CMOS (diode array, passive pixels, pinned diodes…), and neglected some important outsiders, as camera vacuum tubes, photon counters, and the klingon eye.

 

Table 1: optical performance [1]

criterion

eye

CCD

CMOS APS

film

spectral response

400-700 nm
peaked at 555

350-1000 nm

350-1000 nm

300-700 nm

peak quantum efficiency

<20%

>>50%

>50%

<10%

dynamic range

1000000:1 logarithmic
100:1 linear

10000:1 linear

6000:1 linear
100000000:1 logarithmic

10:1 ...100:1 non-linear sigmoid-like response
10000:1 by exposure latitude extension

dark limit

0.001 lux
1E-6 W/m2

typ: 0.1 lux
<0.0001 possible

typ: 0.1 lux
<0.0001 possible

virtually zero

noise photons (*)[3]

10

typ:100 best:10

100

100

integration time or response timeconstant
(room temperature)

0.02-0.5 s

33 ms typical
5 minutes possible

33 ms typical
2 minutes possible

virtually unlimited

max. frame rate

ca. 15 Hz

10 kHz

>> 100 kHz

1 shot only

speed of automatic illumination control

1 minute

none

typ. 1 frame

none

Table 2: optical resolution

criterion

Eye

CCD

CMOS APS

film

#pixels

120M cones

typical: 800K
record: 90M (Fairchild)

typical: 800K…10Meg

typical: >1e6 grains, 10 to 0.01 grains/um2

pixel pitch

2-3 µm

2-10 µm

3-10 µm

effectively 10..20 um (grain size is not equivalent to pixel size!)

focal plane size

3 cm

<1 mm  to wafer scale

<1 mm to wafers size

only limited by film size

Table 3: operating conditions

criterion

Eye

CCD

CMOS APS

film

radiation hard

1 mrad

10 krad to 100 krad

10 krad to 20 Mrad

operating Temp

36 C

-200 C...+200 C

0 K ... +200 C

0K ... 100 C

power dissipation

< 1mW

500 mW typ

50 mW typ

nihil

Table 4: on-board image processing

criterion

eye

CCD

CMOS APS

film

cosmetic quality

perfect by definition

very good

very good

as good as perfect

color

ideal hue gamut by definition

3-color acquisition

3-color acquisition

3-color print quality

absolute photometry

impossible

easy

easy

possible

focal plane processing

extensive

none

often

none (apart from DIR: development inhibitor releasing couplers)

access method

data driven (focus of attention)

serial only

serial, random access, ...

optical only

datapath

5M nerves

8-14 bits

8-14 bits

none

Table 5: logistics

criterion

eye

CCD

CMOS APS

film

price

invaluable

typ 20 EUR

typ 1 EUR

typ 0.1 EUR

2nd source

none

few

many

few

technology development cycle

500 Myears

3 years

1 years

20 years

fabs

3E9

5

10

5

Table 6: Eye, CCD and CMOS principles

Eye

CCD

MOS diode array

MOS active pixel

film

Description: V:\Caeleste Vault\caeleste_website\lexicon\eye\Image7.gif

Description: V:\Caeleste Vault\caeleste_website\lexicon\eye\Image4.gif

Description: V:\Caeleste Vault\caeleste_website\lexicon\eye\Image5.gif

Description: V:\Caeleste Vault\caeleste_website\lexicon\eye\Image6.gif

photo-chemical detection, transfer by neurons

multplexing of charge packets by lossless transfer

multiplexing by switches on photodiode node

multiplexing of the local amplifier output

photo-chemical detection and development

rod and cone signal levels are "processed" by ganglion cells

one charge sense amplifier at the end of the CCD register

one charge sense amplifier on the output bus

(charge sense) amplifier in every pixel

pure local change in optical absorption

 



 [2] the real name is the vertebrate’s eye.  It’s not really a man made development.

[1] Bart Dierickx, “Silicon versus the Eye”, presented at the IEEE Workshop on CCD&AIS, 4 june 97

[3] noise photons is defined as: noise photons = noise electrons / (quantum efficiency * fill factor)