recent progresses of visible light image sensors©2018 n. teranishi the detector seminar at cern...
TRANSCRIPT
©2018 N. Teranishi
The Detector Seminar at CERN
Recent Progresses of Visible Light Image Sensors
February 23, 2018
Nobukazu TeranishiUniversity of Hyogo / Shizuoka University
1
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
2
©2018 N. Teranishi
3
What is Image Sensors?
Smart-phone
Movie SecurityIris verificationEndoscopeAutomobile
Image Sensor
Examples of applications
Photodiode
Light
●●
Electron
Microlens
Color filter
Pixel arrayDSC Image sensor Pixel cross-section
- Semiconductor device, which converts light image to electric signal.- Used in cameras.
©2018 N. Teranishi
4
1406 08 10 12040201 03 05 07 09 11 13
Year
1
2
0
3
4
Bpcs
(Source: TSR)
Camera phone
Tablet
PC/WEB camera
Application in amount (2015)
Surveillance
AutomobileGame
Compact DSC
DSLR
Camcoder PMPMedical
OthersBroadcast
CCD image sensorCMOS image sensor
15
Smart glasses
Sal
es
amount
Image Sensor (IS) Market
- IS sales amount has grown by camera phone.
- IS spreads into various applications,
- “Others” includes scientific,
industrial, …
©2018 N. Teranishi
5
Pixel Structure
P+ P+N
P
N
Detective capacitance
(Floating diffusion (FD))
+P+
Silicon
SotrageTransfer gate (TG)
microlens (ML)
Color filter (CF)
Pinning layer
e- e-
e-
Cross-section of frontside illumination type CMOS image sensor pixel
Amplifier (SF)
©2018 N. Teranishi
6
Based on (Piet De Moor (IMEC))
Inner Shell Excitation
Small surfacedead layer isneeded.
Large depletionis needed.
Visible light
Near UV
X-ray
Sensitivity Expansion to Invisible
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
7
©2018 N. Teranishi
8
Classification of Dark Current and Blemish P
ixel
Counts
Dark Signal Level
Average: Dark current・Dark current shot noise・Black level shift by temperature
Dark current FPN(Dark current pixel-by-pixel fluctuation)
3σ
Micro white blemish
(Mid-range)
White blemishes
(Large-range)
Very large level blemish
by radiation damage
©2018 N. Teranishi
9
Metal Contamination (1)
・Metals, especially transition metals, form mid-gap levels, which become GR centers.
kT
EEnpn
npnNvU
iti
itth
cosh2
2
(Sze: “Semiconductor Devices”
Chap. 1 Eq.(59))
Assuming the cross sections, pn
・Generate dark current based on Shockley-Read-Hall (SRH) process
U: Recombination Rate
・Assuming depletion condition, then, n, p = 0,
and Et=Ei because U becomes the largest;
s)1/cm :(Unit 2
3tith NnvU
・One GR center generates 1/s) :(Unit ,2
ithnvU
Et
Valence band
Conduction band
Dark current by mid-gap level
Nt: GR center (trap) density
©2018 N. Teranishi
10
Metal Contamination (2)
・ “Dark Current Spectroscopy” identifies metals and estimatesconcentrations.
・ It can measure very small amounts of metal contamination.
σ(cm2) Density(1/pixel) Density(cm-3)
a 1.8E-15 1.25 1.3E9
b 5.4E-16 0.15 1.5E8
(D. McGrath et al. (TI); “Counting of Deep-Level Traps Using a Charge-Coupled Devices”)
Poisson distribution
a a
b b b
a
Nu
mb
er
of
cou
nts
0 1000 5000400030002000Signal (e-)
0
400
200
300
100
275 K20 Frames60 s integration0 1
2
3
Two series of specific and periodic
peaks, labeled as “a” and “b”.
Assuming metals are distributed
as Poisson distribution, metal
density per pixel can be derived.
σ s are derived from the pitches
using the previous slide formula.
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
11
©2018 N. Teranishi
12
P+
P+
P+ N
TG
P-Well
N-Substrate
N
Storage
1. Grounded P+ pinning layer
prevents interface to be
depleted, and stabilizes PD
electrically.
・Low dark current
・Large saturation
・High blue sensitivity
x x xx x
Signal
PPD (Pinned PD) Structure and Advantages
PinningLayer
0V
Pote
nti
al
ON
OFF
FD
2. Complete electron transfer
・No image lag,
・No transfer noise
x: GR (generation-recombination) center
Shallow P+ pinning layer(Low energy implantation)⇒ Good electron transfer
©2018 N. Teranishi
13
Dark Current Reduction by PPD
Still big reduction!
7
210~
2
2
Depleted
depletedNot iitth
itth
n
p
pnNv
nNv
U
U
Assuming hole density (p) at the P+ pinning layer is 1017 h+ cm-3
- Theoretical dark current reduction ratio using Shockley-Read-Hall Process:
Non-PPD (1982) PPD (2012) Unit
Scheme CCD FSI CMOS
Pixel size 23 x13.5 1.12 x 1.12 μm
Dark current 1,300 5.6 e-/s/μm2 at 60℃
- Actual dark current reduction ratio:
0.4 %
©2018 N. Teranishi
14
PPDConventional PD
If Dark current is reduced,
dark current FPN, and dark current shot noise are also reduced.
Example of Dark Current Reduction by PPD
Dark current FPN is suppressed, then, picture quality is much improved.
©2018 N. Teranishi
Saturation is Increased by PPD
フォトダイオード電位
Non-PPD
PPD
Sat
ura
tion
(e-
10
11/cm
2)
Storage potential (V)
Saturation by simulation1984 B. C. Burkey et al.(Kodak)
12.5 timesP+ P+P +
N
P
N
Pinning layer
+xxx x x
e-
e-
e-
PPD pixel cross-section
Saturation becoms12 times larger by PPD.・ PN junction area is increased.・ Acceptor density at pinning layer >> Donor density at storage
Storage region
31 4200
2
4
6
8
10
©2018 N. Teranishi
Pixel Shrinkage Trend
・ Shrinkage has been carried out steadily.
・ Lens volume and weight become 1/1000 when pixel size becomes 1/100,
assuming that F-number, view angle and pixel number are constant,
Mass Production Year85 90
1
10
50% shrinkage in 3.5 years
1005009580
100
(um2)
Min
imum
Pix
el A
rea
15
CCDCMOS
16
24 years
1/100
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
17
©2018 N. Teranishi
Photon Counting Image Sensor
SPAD
(Single photon avalanche diode)
4-Tr CMOS + High conversion gain +
CMS (Correlated multiple sampling)
(N. Dutton et al., VLSI Symposium 2014)
n=0
n=1 n=2
n=3
n=4
n=5
n=6n=7
・ QVGA SPAD, 20 fps,
room temperature
・ High avalanche gain makes following
circuit noise negligible.
・ Large dark count. Small fill factor
・ In 2015, several organization reported
low noise < 0.3 e- rms.
・ DEPFET (Max Plank)
18
(MW. Seo, S. Kawahito et al., IEEE EDL 2015)
128 samplings
©2018 N. Teranishi
19
Photon Counting (2)
(Seo, Kawahito et al, IEEE EDL, Dec. 2015)
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
20
©2018 N. Teranishi
Stack (1)21
Approach- Bump bonding (In, solder Au, …)- SOI- Wafer-to-wafer oxide bonding
+ TSV- Wafer-to-wafer hybrid bonding
Sensor
Readout circuit
Bump bonding
Motivation of stack image sensors- Non-silicon material for sensor part- Finer process technology for logic part- Smaller footprint- More flexibility of wiring- Isolation between analog and digital circuits
SOI pixel detector (KEK Arai et al.)
Bump
©2018 N. Teranishi
Stack (2). Wafer-to-Wafer Oxide Bonding
(S. Sukegawa et al., ISSCC 2013)
Conceptual Diagram
SEM Cross-section
22
Wafer-to-wafer oxide bonding + TSV(Through silicon via) at peripheral・Smaller chip size → Smaller camera, lower cost・Optimal processes for both sensor and logic ・More functions can be integrated.
Suitable for large volume applications
©2018 N. Teranishi
Stack (4). Wafer-to-Wafer Hybrid Direct Bonding
K. Shiraishi et al. (Toshiba), ISSCC 2016
・ Oxide bonding
+ TSV(Through silicon via)Interconnections are limited at only peripheral.
・ Hybrid bondingInterconnection can be locatedat pixel array also.
・ 2 um pitch interconnectionE. Beyne (IMEC), 2018
23
Cu2Cu optimized CMP
BEOL CMP
CMP optimization Post-bond annealing
< 150 ℃
400 ℃
Y. Kagawa et al. (Sony), IEDM 2016
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
24
©2018 N. Teranishi
NIR --- Motivation
1.To cover shortage of visible light and to increase sensitivityeg. Most illuminations have NIR.
Night glow・ Security use
2.Invisible for mankind and animal. ・ Gesture interface, ToF, ・ Animal observation at night
3.Eye safe laser (1.4-2.6um)・ Longer than Si cutoff wavelength
4.Specific wavelength for each application ・ Window to living organisms ・ Fluorescent protein・ Hemoglobin (Hb) ・ Bill validator (NIR and UV inks are used.)
5. Smaller sunlight background
25
©2018 N. Teranishi
Nightglow (Airglow)
・ Emission in the upper atmosphere- Photoionization by the sun- Luminescence by cosmic ray- Chemiluminescence
・ Light intensity at near IR is larger than that of visible light
http://en.wikipedia.org/wiki/Nightglow
Night glow from ISS Intensity of night glow
©2018 N. Teranishi
Eye-Safe Laser
http://syounanakaribit.sakura.ne.jp/newpage605.htmlTransmittance of young human eye
・ 1.4 - 2.6 μm wavelength light hardly reaches retina.・ Application: Night vision, range finder, obstacle detection・ Note: Silicon cannot detect 1.4 - 2.6 μm wavelength light.
00 0.8 1.0 1.2 1.4
Cormea
Aqueous
Lens
Vitreous
Total
0.60.40.2
Wavelength (μm)
Tra
nsm
itta
nce
©2018 N. Teranishi
・ NIR, 700~900 (1400) nm, has a good transmittance of human body.・ It can pass through even 30 cm thick human body.
Window to Living Organisms
Application・ Vein authentication ・ Optical topography・ Fluorescent protein https://www.an.shimadzu.co.jp/bio/nirs/nirs2.htm
Specific wavelengths ・Hemoglobin (Hb):
Oxygen density in vein ・Myoglobin:
Oxygen density in cardiac muscle and skeleton muscle
・ Cytochrome c oxidase (COX)・ Melanin
Absorptance of hemoglobin and water
Wavelength (nm)
Window to living organisms
Hemoglobin (Hb)Water
©2018 N. Teranishi
NIR Fluorescence Imager (ICG Fluorescence)
https://www.hamamatsu.com/resources/pdf/sys/SMES0020J_pde-neo.pdfICG: Indocyanine-green
ICG
Excitation light(760 nm) Fluorescence
light(830 nm)
NIR fluorescence camera
Visible light image NIR fluorescence image
・ Portable intraoperative diagnosis ・ Both excitation light and fluorescence light are inside the widow of living organisms.
©2018 N. Teranishi
Sunlight Background
http://syounanakaribit.sakura.ne.jp/newpage605.html
・ ToF, for example, uses active illumination or modulated illumination.・ Longer wavelength at NIR, the sunlight background becomes smaller,
and SN ratio will be improved.
Spectral energy distribution of sunlight
1965 Valley
Wavelength [μm]
Energ
y
©2018 N. Teranishi
Absorption of NIR and X-ray in Silicon
0
20
40
60
80
100
0
20
40
60
80
100
0 2010 30 40 500 200100 300 400 500 600
Depth (um)Depth (um)
800nm850nm
950nm900nm
1000nmv
1050nm17.4keV19.7keV
19.7keV
17.4keV1050nm
1000nmv
950nm
・ Large depletion layer is needed to detect NIR or X-ray.For example, 900 nm wavelength NIR (equivalent with 5.4 keV X-ray)is absorbed half by 15 um silicon.
ref: 3 um silicon for 700 nm
Enlarge
8.7keV
8.7keV
5.4keV
13keV
13keV
Intensity (%)
Intensity (%)
©2018 N. Teranishi
NIR QE Increase (1), Pyramid Surface for Diffraction
・ Pyramid Surface, which enhances diffraction.・ DTI(deep trench isolation) between pixels to suppress crosstalk
Increase effective sensor thickness and increase NIR QE.
Pyramid surface ofr diffraction
Wavelength [nm]
(2017IEDM Oshiyama (Sony))
33 %
18 %
Wavelength [nm]
©2018 N. Teranishi
Black Silicon
(2004 Shen, Harvard University)
・ Black silicon is formed on illuminated surface by femtosecond laser・ As well as NIR QE is much improved, the cutoff wavelength is
enlarged beyond 1.2 μm.
Pulse Number dependence( 50 μm squqre)
(2017 Bitran)
Rela
tive
QE
NIR image of IC card
Wavelength [nm]
©2018 N. Teranishi
(B. Ackland et al. 2009)Ge Image Sensor for NIR
Dislocation
Ge epitaxiallayer
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
35
©2018 N. Teranishi
ToF (Time of Flight) Mechanism
Distance image by ToF
ToF setup
(Kawahito Lab., Shizuoka University)
Intensity image
http://www.idl.rie.shizuoka.ac.jp/study/project/tof/index_e.html
Modulated light source
ToF image sensor
Time
Inte
nsi
ty
Emitted lightReceived light
ΔT
Distance = 𝟏
𝟐𝒄∆𝑻 (𝒄: 𝐥𝐢𝐠𝐡𝐭 𝐯𝐞𝐥𝐨𝐬𝐢𝐭𝐲)
36
©2018 N. Teranishi
ToF Applications
Smart Chart
Forward looking
(Obstacle detection)
・ Gesture input: game, medical, digital signage, automobile
・ Range finder, 3D measurement: automobile, dental, robot
Game Digital signage
Dental
(C.Niclass,ISSCC 2013)
(EETimes Japan, 2011)
37
©2018 N. Teranishi
ToF Challenges (Multi-Pass Errors)
・To resolve the multi-pass errors is challenge at ToF.
(a) Normal (b) Reflector
(d) Corner(c) Semi-Transparent
38
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
39
©2018 N. Teranishi
Polarization Image Sensor
CH
R
COOHNH2
CH
R
COOHNH2
Optical Rotation
(Tokuda 2011 IISW)
Metal gridpolarizer
18013590450Polarization angle (degree)
100
101
102
103
Ou
tpu
t
43.7
40
Stress image
(Plastic lens)
Antireflection
R. Muller
ISEurope2017
https://www.4dtechnology.com/
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
41
©2018 N. Teranishi
(S. Koyama et al.@Panasonic, ISSCC 2013)
Camera diagram Pixel cross section Angular response
Parallax
Monocular 3D Image Senosr
Monocular 3D is required instead of binocular 3D・Compact, ・No precise assembling is needed.
42
©2018 N. Teranishi
43
New world of image sensors
Human sees photos and videos.
+
Computer vision (Computers see photos)
Driving assistance
Barcode
Face recognition
Gesture input
+New Functions
Range imageFluorescencelifetime image
(Kawahito, Shizuoka University) Robot(NEC)
(Subaru)
(SoftBank)
(Panasonic)
©2018 N. Teranishi
Contents
1. Basics of Visible Light Image Sensors2. Dark Current and Blemish3. Pinned Photodiode (PPD)
Recent Progresses4. Photon Counting Image Sensor5. Stack6. Near InfraRed (NIR)7. ToF (Time of Flight)8. Polarization Image Sensor9. Monocular 3D Image Sensor
10. Queen Elizabeth Prize for Engineering
44
©2018 N. Teranishi
Queen Elizabeth Prize for Engineering
Purpose: The world’s most prestigious engineering prize,
to celebrates world-changing innovations in engineering
Citation: The creation of digital imaging sensors
Winners: - Eric Fossum: CMOS image sensor
- George Smith: CCD (Charge-Coupled Device)
- Nobukazu Teranishi: Pined photodiode (PPD)
- Michael Tompsett: CCD image sensor
45
©2018 N. Teranishi
Presentation at Buckingham (1)
Reception at Guildhall hosted by the Lord Mayer of London
46
http://www.guildhall.cityoflondon.gov.uk/gallery
N. Teranishi
Invitation letter
©2018 N. Teranishi
From left, Prince Charles, Eric Fossum, Michael Tompsett, Nobukazu Teranishi (Absent: George Smith)
47
Presentation at Buckingham (2)
©2018 N. Teranishi
48
Impression
・ Proud and happyThe best award for engineers
・ LuckyThere are so many important and beneficial technologies.
・ AcknowledgementTo family, teachers, seniors, colleagues, and everybody
・ Step-upI would like to grow up as a person of the Prize.
・ Education for engineersGood engineers are needed to evolve industries.So, I would like to introduce respectable engineers to young generation,and help them to learn technologies.
©2018 N. Teranishi
49
Anyone can take a photo
anywhere, anytime!
Thank you !
©2018 N. Teranishi
N N N
P
e
③Charge diffusion
①Angled incident light
②Multi-reflection,
Diffraction
Crosstalk Mechanism
Crosstalk effect: ・Color mixture, degradation of chromatic S/N.
・Degradation of resolution
Pixel Shrinkage -Crosstalk-
~3 um
1um
50
©2018 N. Teranishi
Pixel Shrinkage -DTI (Deep Trench Isolation)-
・BSI is used for small pixel to reduce crosstalk and increase sensitivity.
・DTI suppress both optical cross talk and charge diffusion cross talk.
(Y. Kitamura et al.(Toshiba), IEDM,2012)
B-DTI (Back-side DTI)・DTI is formed from backside.・Metal is buried in DTI.・Sufficient cross talk reduction
(JC Ahn et al.(Samsung), ISSCC,2014)
F-DTI (Front-side DTI)・DTI is formed from front side.・Poly-silicon is buried in DTI.・Better DTI interface
51
©2018 N. Teranishi
P Epitaxial
P Substrate
N
P+
P+
N+N+
-
+
STI
TG RGFD RD
52
Dark Current Possible Causes
PD interface GR center
Large electric field at TG edge TG interface GR center
RG off-leak
Diffusion currentSTI interface GR center Depleted region GR center
Large electric field at P+ /N junction FD dark current
Charge flow from the outside
- Pixels should be carefully designed by using precise ion implantations.
- Implantations occasionally bring metals and damages into silicon.
©2018 N. Teranishi
53
P+ P+P+N
TG
P
N
PPD xxx
Causes of Image Lag in PPDs (1)
0V
Pote
nti
al
FD
(A) Small electric field
(B) Barrier at the PD edge
- -
--
(D) Pump back when the
signal is large
Off
(C) Pocket at the TG edge
--
-
- -
(E) Traps at the TG interface
On the next slide.
+
CBarrier
On
OnOn
On
Off
©2018 N. Teranishi
54
P+ P+P+ N
TG
P
N
PPD xxx FD
(E) Traps at the TG interface
If the electron transfer path touches the interface, some electrons
are captured by traps.
- Some of them are detrapped in the following frames, causing lag.
- Some of them are annihilated, causing non-linearity.
Signal electrons at PPD
Ou
tpu
t el
ectr
on
s
Ideal case
With traps
- The signal electron annihilation exhibits this kind of non-linearity.
- A buried transfer path is needed to suppress these phenomena.
x : Traps at the TG interface
Causes of Image Lag in PPDs (2)
©2018 N. Teranishi
Pixel Shrinkage Trend
・ Shrinkage speed becomes slower recently.
・ In 2017, 0.9 um pixel began to be mass produced.
Mass Production Year85 90
1
10
50% shrinkage in 3.5 years
1005009580
BSI
Microlens
100
(um2)
Min
imum
Pix
el A
rea
Lightpipe
Inner MicrolensShifted Microlens
15
StackDTI
CCDCMOS
55
Pinned PD
©2018 N. Teranishi
Stack (3). Pixel-DRAM-Logic Oxide Bonding56
(WikiChip Fuse, 2018/02/3)
1 Gbit19.3 Mpixel
©2018 N. Teranishi
57
On-chip Optics Functions
Si
Light
Microlens
Lightpipe
BSI (Back Side Illumination)
PD
Photo shield
(1) Light collecting
(2) Color filtering
Color filter
Vertical color separation
(3) New specific functions
Polarization image sensor
Phase detection autofocus
Computational photography
©2018 N. Teranishi
・RGB-IR image sensor is proposed to obtain IR image and RGBimage simultaneously.
・Double bandpass filter in stead of conventional IR-cutfilter is used to detect NIR.
・Color reproductivity is not sufficient.
RGB-IR Image Sensor
R
BG
IR
400 500 600 700 800 900 1000 1100
Wavelength [nm]
Quan
tum
Eff
icie
ncy
Double Bandpass Filter
RGB-IR Image Sensor
IR
58
Conventional IR-cut filter