The class was offered by Prof. Arjang Hassibi, my advisor.
The title of the class is "Integrated Sensor".
1. Performance metrics of sensors
1) Transfer function (f(x)): ensemble average of the output
2) Sensitivity: small signal ratio (or change) of the output for a give input
3) NDNL (Normalized differential nonlinear)
NDNL(x) = (f(x) - f`(x))/f`(x), where f`(x): the best linear fit for f(x)
4) Systematic errors and random errors
Systematic errors - e.g., gain error, offset, background interference
Random errors - e.g., thermal, quantization, shot-noise
(Note) The difference between above two errors
--> Systematic errors can be calibrated, but not for random errors
5) SNR: Signal-to-Noise Ratio
If a transfer function has flat response over input, the input noise is large.
6) Accuracy and Uncertainty
Given Yo(specific output) is observed, what is the input??
Typically we are interested in E[x|y=Y0], the representive input signal, and E[x^2|y=Y0] - (E[x|y=Y0])^2, the input noise given y=Y0.
where E[n]=0 (zero-mean) of the noise is assumed!7) Dynamic Range (DR)
The input range where SNR is acceptable!
(DR is not defined as SNR(max) - SNR(min)!!!)
(Note)
X(MDL): Minimum detectable level --> typically limited by noise
X(HDL): Highest detectable level --> typically limited by saturation or quantization
(Note) You might require basic probability and Tylor series approximationknowledge.
2. Noises in Sensors
1) Random noises
i) Thermal noise
- e.g., CMOS, Diode, BJT, resistor
- Signal indepedent, temperature dependent
- white noise and Gaussian amplitude distribution
- originates from the random motion and random scattering of many charged species within a medium much larger than their mean free-path. In other words, Macroscopic view!!
- steady-state (stationary process) case: 4KTR [Vrms^2/Hz] (1-sided)
2KTR[Vrms^2/Hz] (2-sided)
- if a system undergoes a transient response, the noise analysis should be done in time domain
ii) Shot noise
- e.g., BJT, Diode
- Current going through junctions create shot noise
- signal dependent and temperature independent
iii) Substrate noise
(Note) It is good to know 1) Basic stochastic process, convolution, Lapalce transform, LTI system properties...etc
2) Systematic error (noise)
- all passive and active elements (due to process variation, design, and layout...etc)
- greatly reduced by proper design and layout
- Can be calibrated!
3) Noise calculation at the output and input referred-noise power in various ckts
3. Circuit Architectures for Sensors
1) Detecting sinusoidal tones
- Objective: to measure amplitude, freq, and/or phase information
- narrowband detection
- Ex) Brutal fast A/D, Superheterodyne, and Direct conversion...etc
2) Voltage Detection
- Objective: to acquire a voltage signal which has DC or low frequency components
- Useful when parasitic components are in series and the output of the transducer is large so that most of voltage source information is delivered to the detection circuit very well.
- Techniques for reducing 1/f or low frequency noise
i) Chopper stabilizer (CHS)
ii) Auto zero (AZ) - noise should be slow and additive
iii) Switching biasing - done in device level
3) Current Detection
- Objective: to acquire a current signal which has DC or low frequency components
- Useful when parasitic components are in parallel and the output of the transducer is small so that most of current source information is delivered to the detection circuit very well.
- Techniques for reducing 1/f or low frequency noise
i) CHS
ii) Switching biasing
- Transimpedance amplifier(TIA)/Capacitive transimpedance amplifier (CTIA)
4) Parametric Detection
- Objective: to acquire some characteristic of a passive electrical component
- This method typically uses "the output value change from the nominal ouput" as its measurement
- This method requires an excitation signal
- Information lies on "The difference of the outptu from the nominal output"
4. Integrated Image Sensors (IS)
1) Signal path i IS
- Photonflux(ph/cm^2.sec)-->Current density(A/cm^2)-->Charge(Col)-->Voltage(V)
2) Photocurrent and quantum efficiency (QE)
- absorption coefficient(alpha) (=penetration coefficient^-1)
- Phtonflux at depth x in Si
F(x)=F0*exp(-alpha*x) [ph/Cm^2.sec]
where Fo is the photonflux incident on the surface
- Generation rate [e-h pair/Cm^3.sec]
G(x)=alpha*F(x)
- Quantum efficiency
Def: Given F0, how many electrons that contribute to the photocurrent are generated?
QE = (jph/q)/F0
where jph: photocurrent density, q = 1.6X10^19 [Col]
(Note) jph is linearly proportional to qF0
- Miscellaneous
i) shorter wavelength = high frequency
E = hv=hf, thus higher frequency, the higher energy
ii) If a photon energy is smaller than bandgap of a silicon, it will pass through it!
iii) Shorter wavelength photon generate e-h pair easily (shallow depth in the silicon)
3) Dark Current
- Photodetector current with no illumination present
- Bad:
- introduce unavoidable shot noise (qID)
- reduce signal swing
- DSNU (Vary over image sensor array)
- Sources:
- Thermal generation
- Defects - e.g., interface, material defects..etc
4) Photodiode Structure in CMOS Process
- P+/nwell
- nwell/Psub
- N+/Psub
5) Direct Integration
- Vo(t)=Vref - (Id/Cj)*Tint = Vref - ( (Iph+Idc)/Cj)*Tint
where Id = photodector current = Iph + Idc
Iph = photocurrent
Idc = dark current
Tint = Integration time
- Well capacity(Qmax): Max. # of electrons that a photodector can handle [electrons]
HDL = Iph,max = (qQmax/Tint)-Idc, where qQmax <= VDD*Cj where VDD: supply voltage 6) Integration with CTIA
- The junction capacitance of a photodiode is not changing during integration because there is no bias condition change during integration
- Output voltage is no longer a function of Cj, the juction capacitance of the photodiode, but for a feedback capacitor, which is much linear than Cj!!
5. Image Sensor Architecture and Noise
- Pixel architecture
PPS, APS, DPS
- Some common senses
After reading (i.e., word signal) the information, the integration cap. should be reset!
Thus, the integration time is always defined the time interval between two consecutive reset pulses (signals).
1) Passive Pixel Sensor (PPS)
- Direct integration, reading charges using CTIA
(Note) Here CTIA is used for reading (amplification) of charges, but not for integration
- Photodiode is reset during reading (word signal)
(Note) Reset is used for resetting CB and CF, not integration capacitor! The integration capacitance, Cj, is reset during word signal!i) Signal transfer function
ii) Noise analysis
- Direct integration noise
- Reset noise
We are not interested in output noise, but the noise stored on the capacitors at the end of the reset. These noise will get transferred to the output during the readout!
- Readout noise
(Note) Noise fed back to the photodiode - noises are stored on Cj (junction cap. of the photodidoe) at the end of readout, which is read out during the following readout. Thus, due to previous fedback noise, the output noise will be doubled!
2) Active Pixel Sensor (APS)
- General advantages of APS over PPS
--> reading is nondestructive, faster, and multiple reading is possible
i) Types of APS
(1) 3T APS: photodiode, direct integration
(2) 4T APS: photogate, direct integration
(3) 1.75T/1.5T APS: Pinned diode, direct integration
(4) CTIA APS: photodiode, integration with CTIA (integration over feedback cap.)
ii) 3T APS
(Note) Reset signal here is used for resetting the integrating capacitor, Cj!- Anti-blooming scheme is possible during integration
- The very bottom current source (transistor with Vbias gate biasing) is used as a current source for all source followers.
(Note) Why is not the reset switch Pmos? Due to the size of Pmos implementation, in order to have large fill factor, we perfer NMOS to PMOS!
iii) 1.75T/1.5T APS
A floating diffusion capacitor in these architecture is the part of the reading action
3) Charge-Coupled Devices
i) Photogate - collect and generate e-h pair
- NMOS(VG>0) and PMOS(VG<0)>Eph))
(2) Low QE due to interface states
- Thich gate oxide is desired
(1) In order to avoid gate leakage due to large gate voltage
(2) In order to make a bump small in a potential well
Smaller a bump, better the transfer efficiency!!
ii) CCD Basic
(1) CCD is a "dynamic" analog (charge) shift register
(2) CCD is clocked and all operations are in "transient mode"
(3) Charge is coupled from one gate to the next gate by fringing electric field, potential and carrier gradient
- Surface potential control in CCD
(1) Doping method
(2) Oxide step
(Note) Higher doping, higher threshold voltage for the same doping material. If the doping materials are different, higher doping region makes threshold voltage lower!
ii) CCD Architectures
(1) FT-CCD (Frame transfer CCD) - most popular
(2) Full FT-CCD - need mechanical shutter
(3) IT - CCD (Interline transfer CCD) - faster, small fill factor
iv) Charge transfer efficiency (CTE)
- Why is CTE 1000%?
--> main causes (1) lack of time to complete tranfer
(2) charge trapping - captured by interface states
(3) due to a bump in a potential well
- transfer mechanisms
: a combination of carrier diffusion and carrier drift
4) Technology
- What are the considerations for image sensor fabrications?
5) Stuff not covered!
- Fixed Pattern Noise (FPN) - nonuniformity
: spatial variation in pixel output values under uniform illumination due to device and interconnnect parameter variations (mismatches) across the sensor
: gain FPN and offset FPN
: Calibrations
6. Integrated Magnetic Field Sensors
1) Detecting alternative magnetic field
2) Detecting DC magnetic field
- Hall sensor (uses Hall effect)
7. Backbone Ckt for MEMS and Packaging
8. Biological and Chemical Sensors
1) General Biosensor Platforms
- Chanllenges
(Biology)
High detection sensitivity
Detection in presence of interference
Parallel detection
Amplification and lable-free
(Electronics)
Low cost
Battery operated
Portable or hand-held
Biocompatible/Implantable
2) Affinity based biosensor
- Many biological molecules, when brought together, can go to a lower energy state by binding
- Molecular affinity
(1) The bindings are inherently probablitistic
(2) Thermal energy results in a postivie probability of detachment
(3) Non-specific binding is also possible, although less probable
- Process procedures
(1) Sample exposure
(2) Incubation (i.e., hybridization)
(3) Detection
- Noise in biosensors
: Noise is dominant by biochemical moise, not by electronic noise
3) Electrochemical Detection
- Molecular binding changes the characteristics of the electrode-electrolyte interface
- Biosensor can detect the changes (parametric detection)
- Study of Prof. Hassibi's circuit (Elecrtrochemical sensor microarray)
------------------ END of Class -----------------------------------------------
No comments:
Post a Comment