Lecture Notes 9

Spatial Resolution

• Sensor Modulation Transfer Function (MTF)

• MTF Calculation

• Aliasing

EE 392B: Spatial Resolution 9-1

 Preliminaries

• The image sensor is a spatial (as well as temporal) sampling device (of
the incident photon flux image) — the sampling theorem sets the limits
for the reproducibility in space (and time) of the input spatial (and
temporal) frequencies
• So spatial (or temporal) frequency components higher than the respective
Nyquist rate cannot be reproduced and cause aliasing
• The image sensor, however, is not a point sampling device in space (or
time), and cannot be approximated as such
◦ Photocurrent is integrated over the photodetector area (and in time)
before sampling
◦ Photogenerated carriers in quasi-neutral regions of a pixel may
diffuse and be collected by its neighboring pixels
These effects (in addition to the optics) result in low pass filtering and
crosstalk before spatial (and temporal) sampling

EE 392B: Spatial Resolution 9-2

 • We focus here on spatial sampling
p

p-sub

◦ Assuming a square pixel with width (pitch) p, the spatial Nyquist

1
rate in each dimension is fNyquist = 2p and is typically reported in line
pairs per millimeter (lp/mm)
◦ Signals (photon flux images) with spatial frequencies higher than
fNyquist cannot be faithfully reproduced, and cause aliasing
◦ The low pass filtering caused by integration and diffusion degrades
the reproduction of frequencies below fNyquist — degradation
measured by the Modulation Transfer Function (MTF)

EE 392B: Spatial Resolution 9-3

 Modulation Transfer Function (MTF)

• The contrast in an image can be characterized by the modulation

Smax − Smin
M= ,
Smax + Smin
where Smax and Smin are the maximum and minimum pixel values over
the image
Note that 0 ≤ M ≤ 1
• Let the input to an image sensor be a 1-D sinusoidal monochromatic
photon flux
F (x, f ) = Fo(1 + cos(2πf x)), for 0 ≤ f ≤ fNyquist
The sensor modulation transfer function is defined as
Mout(f )
MTF(f ) =
Min(f )
From the definition of the input signal, Min = 1
• MTF is in general difficult to model and analyze for a real sensor and is
determined experimentally
EE 392B: Spatial Resolution 9-4
 • By making several simplifying assumptions (as we shall see), the sensor
can be modeled as a 1-D linear space-invariant system with impulse
response h(x) that is real, nonnegative, and even
In this case the transfer function H(f ) = F[h(x)] is real and even, and the
signal at x
S(x) = F (x, f ) ∗ h(x)
= Fo(1 + cos(2πf x)) ∗ h(x)
= Fo (H(0) + H(f ) cos(2πf x))
Therefore
Smax = Fo(H(0) + |H(f )|)
Smin = Fo(H(0) − |H(f )|),
and the sensor MTF is given by
|H(f )|
MTF(f ) =
H(0)

EE 392B: Spatial Resolution 9-5

 Simplified MTF Derivation

• We consider a 1-D doubly infinite image sensor

p w
x
depletion
region n+ Ld

quasi-neutral L
p-region
z

• To model the sensor response as a linear space-invariant system

◦ We assume n+/p-sub photodiode with very shallow junction depth,
and therefore we can neglect generation in the isolated n+ regions
and only consider generation in the depletion and p-type
quasi-neutral regions
◦ We assume a uniform depletion region (from −∞ to ∞)

EE 392B: Spatial Resolution 9-6

 • The monochromatic input photon flux F (x) to the pixel current iph(x)
can be represented by the linear space invariant system (iph(x) is sampled
at regular intervals p to get the pixel photocurrents)

jph (x) x
F (x) d(x) w ⊓ (w ) iph (x)

Photogeneration Integration

where (
x 1 |x| < w2
⊓ =
w 0 otherwise,

d(x) is the (spatial) impluse response corresponding to the conversion

from photon flux to photocurrent density (we will derive it soon), and we
assume a square photodetector

EE 392B: Spatial Resolution 9-7

 • The impulse response of the system is thus given by
x
h(x) = d(x) ∗ w ⊓
w
and its Fourier transform (transfer function) is given by
H(f ) = D(f )w 2 sinc(wf ),
where
sin(πx)
sinc(x) =
πx
• Note that D(0) = qQE(λ), so D(f )/q can be viewed as a generalized
quantum efficiency (function of spatial frequency as well as wavelength)

EE 392B: Spatial Resolution 9-8

 Finding D(f )

• To find D(f ), note that the photocurrent density

sc n
jph(x) = jph (x) + jph (x),
sc
where jph (x) is the photocurrent density due to generation in the
n
depletion region and jph (x) is the photocurrent density due to generation
in the quasi-neutral p-region
• Assuming an incident 1-D photon flux F (x) (at z = 0) with Fourier
Transform F(f ), then
F[jph]
D(f ) =
F(f )
so we need to find F[jph]
• First note that the generation rate at (x, z) is
g(x, z) = αe−αz F (x)

EE 392B: Spatial Resolution 9-9

 • Assuming that all photogenerated charge in the depletion region is
collected, we obtain
Z Ld
sc
jph (x) = q g(x, z)dz = qF (x)(1 − e−αLd )
0
sc
and F[jph ] = q(1 − e−αLd )F(f )
n
• Finding F[jph ] is more involved, first recall that

n ∂np(x, z)
jph (x) = qDn ,
∂z z=Ld

where np(x, z) is the photogenerated electron concentration at (x, z), so

n ∂Np (f, z)
F[jph ] = qDn
∂z z=Ld

where Np(f, z) is the Fourier Transform w.r.t. x of np(x, z)

EE 392B: Spatial Resolution 9-10

 • To find np(x, z), we need to solve the 2-D continuity equation (in steady
state)  2
∂ np ∂ 2 np

np
0 = Dn + − + g(x, z),
∂x2 ∂z 2 τn
where the np/τn term is the recombination (which cannot be ignored
here), with the assumed boundary conditions
np(x, Ld) = 0 (edge of depletion region)
np(x, Ld + L) = 0 (ohmic contact)

• Since we want to find Np(f, z), we take the Fourier Transform of the
continuity equation w.r.t. x and we obtain
2
 
2 ∂ Np Np
0 = Dn (j2πf ) · Np(f, z) + − + G(f, z),
∂z 2 τn
where
G(f, z) = F[g(x, z)] = αe−αz F(f )

EE 392B: Spatial Resolution 9-11

 Now we define
L2n
=L2f ,
1 + (2πf Ln)2
√
where Ln = Dnτn is the diffusion length of electrons in p-sub (around
600µm for our 0.5µ technology)
Substituting in the previous equation we obtain
∂ 2Np Np G(f, z)
− 2 + 2 =
∂z Lf Dn
The solution w.r.t. z has the form
− Lz z
Lf
Np(f, z) = c1e f + c2 e + c3G(f, z)
Substituting in the PDE we obtain
L2f
c3 =
Dn(1 − (αLf )2)

EE 392B: Spatial Resolution 9-12

 Using the boundary conditions, we obtain
− LL
G(f, Ld + L) − G(f, Ld)e f
c2 = −c3 Ld +L Ld −L , and
Lf Lf
e −e
2Ld Ld
Lf Lf
c1 = −c2e − c3G(f, Ld)e

• Thus
n ∂Np(f, z)
F[jph (x)] = qDn
 ∂z Lz=Ld Ld

c1 − d c2
= qDn − e Lf + e Lf − αc3G(f, Ld)
Lf Lf
 
−αLd −αL − LL
αe e −e f
= qDnc3 (1 − αLf ) −    F(f )

Lf sinh LL
f
 
−αL − LL
−αLd 1 e −e f
= qLf αe −  F(f )
  
1 + αLf (1 − (αL )2) sinh L

f L f

EE 392B: Spatial Resolution 9-13

 • Combining the results we obtain
sc n
F[jph] = F[jph (x) + jph (x)]
  
−αL − LL
−αLd −αLd 1 e −e f
= q 1 − e + Lf αe  −    F(f )
1 + αLf (1 − (αL )2) sinh L
f L f

Thus  
− LL
−αLd

qLf αe−αLd e−αL − e f
q 1 + αLf − e
D(f ) = −   ,
1 + αLf (1 − (αLf ) ) sinh LL
2
f

and the system transfer function is given by

 
−αLd −αLd −αL − LL
1 + αL f − e L f αe (e − e f ) 2
H(f ) = q  −   · w sinc(wf )
1 + αLf (1 − (αLf ) ) sinh LL
2
f

1
• Finally, the modulation transfer function for |f | ≤ 2p is
|H(f )| D(f )
MTF(f ) = = · sinc(wf )
H(0) D(0)
D(f )
◦ D(0) is called the diffusion MTF
◦ sinc(wf ) is called the geometric MTF

EE 392B: Spatial Resolution 9-14

 Example — Diffusion MTF

For the following examples: p = 10µm, Ld = 1.8µm, and L = 10µm

1.05

0.95
MTF

0.9

0.85 Lambda = 400nm

Lambda = 550nm
Lambda = 700nm

0 5 10 15 20 25 30 35 40 45 50
spatial frequency (lp/mm)

EE 392B: Spatial Resolution 9-15

 Example — Geometric MTF

0.95

0.9
MTF

0.85

0.8

w = 4um
w = 6um
w = 8um
0.75
0 5 10 15 20 25 30 35 40 45 50
spatial frequency (lp/mm)

EE 392B: Spatial Resolution 9-16

 Example — Sensor MTF

Here we take w = 6µm

1

0.95

0.9
MTF

0.85

0.8

0.75
Lambda = 400nm
Lambda = 550nm
Lambda = 700nm
0.7
0 5 10 15 20 25 30 35 40 45 50
spatial frequency (lp/mm)

EE 392B: Spatial Resolution 9-17

 Example — Degradation due to MTF

Here λ = 700nm
input pattern

20

40

60

80

100
50 100 150 200 250 300 350 400

output pattern

20

40

60

80

100
50 100 150 200 250 300 350 400

EE 392B: Spatial Resolution 9-18

 Aliasing

• Aliasing occurs when the bandwidth of the input signal exceeds the
Nyquist rate – the high frequency components are “folded” into the band
F(f )

1 1 f
−B − 2p 2p B
H(f )

S(f )

1
S(f ) refers to the Fourier Transform of the sampled photocurrent for |f | ≤ 2p
EE 392B: Spatial Resolution 9-19
 Aliasing Effect

Here we take λ = 700nm

input pattern

20

40

60

80

100
50 100 150 200 250 300 350 400

output pattern

20

40

60

80

100
10 20 30 40 50 60 70 80 90 100

EE 392B: Spatial Resolution 9-20