Документ взят из кэша поисковой машины. Адрес оригинального документа : http://zebu.uoregon.edu/ccd.html
Дата изменения: Fri Mar 9 01:22:02 2012
Дата индексирования: Mon Oct 1 19:24:36 2012
Кодировка:
Поисковые слова: п р п п р п п р п р п п р п р п п р п р п п р п р п п р п р п р п п р п п р п р п п р п п р п р п р п п р п р п п р п п р п р п п р п р п п р п п р п р п п р п

CCDs for Material Scientists

Evolving Towards The Perfect CCD

Note: This material is from a 1994(!) Talk about CCDs - the physics remains the same, only the detector size has increased since then.

Characteristics of an Ideal Detector:

100% QE

Perfectly Uniform Response

Noiseless

Unlimited Dynamic Range

Completely Understandable Characteristics

Why Astronomers Love CCDs:

High QE compared to photographic media

High Linearity

Large Dynamic Range

Fairly Uniform Response

Relatively Low Noise

Its Digital !

Early Limitations:

Low areal coverage

poor blue response

Read Out Noise Dominated for Spectroscopy

Low light level deferred Charge Transfer Problems

Most of these limitations have now been overcome and have lead to todays 2048x2048 CCD detector. Charge-coupled devices (CCDs) have been moving closer to becoming an ideal detector and are almost there now

HISTORY:

Conceived in 1970 at Bell Labs

Electronic Analogue to Bubble Memory --> a bit = packet of charges (e-) or holes (h+)

Charge detection amplifier gives an external voltage

Charge packets read one at a time --> Serial Device

1973: JPL initiates Scientific Grade large array CCD program

1974: Fairchild 100x100 on an 8-inch telescope produces first astronomical CCD image

1974--1979: NASA's Traveling CCD Camera System --> always made new scientific discoveries

1979: RCA 320x512 LN Cooled System sees first light at KPNO 1-m

THEORY AND OPERATION:

CCD must perform 4 tasks to generate an image:

Generate Charge --> Photoelectric Effect

Collect Charge --> pixels: an array of electrodes (called gates)

Transfer Charge --> Apply a differential voltage across gates. Signal electrons move down vertical registers (columns) to horizontal register. Each line is serially read out by an on-chip amplifier.

Detect Charge --> individual charge packets are converted to an output voltage and then digitally encoded.

Metal Insulator Capacitor (MIS) . Application of Positive voltage means that majority carriers (holes) are repelled in region underneath the oxide layer --> this forms a potential well for electrons.
A CCD is an array of closely spaced MIS capacitors separated by channel stops (implanted potential barriers)

Initial design used surface channels --> but charge can become trapped by fast surface states As a result CTE is 98% which is horrible.

To circumvent the surface state trapping problem requires a buried channel CCD (electrodes buried into the bulk). CTE's of up to 99.999% can be achieved in this manner.

n-dopant reshapes the potential well and forces electrons to collect below the oxide interface.

electrons are now the majority carriers

charge carrying capacity is now less due to increased gate-channel separation (lowers the capacitance)

all buried channel devices are susceptible to charge spill over

So now we have our basic CCD

CCD MANUFACTURING

Main limitation is number of gate shorts or opens. These defects are usually caused by contamination. Minimum clean room level is 100.

Kinds of shorts

Interlevel short --> shorts between clock phases when aluminum gates are used in a 3 phase detector. Very poor CTE transfer in vicinity of short lowers functional yield to about 0.5%

Aluminum gate shorts would limit array size to 512x512

Poly silicon gates developed in early 80's at TI

Doped poly silicon is much less conductive than aluminum --> interelectrode shorts are no longer catastrophic. Also poly silicon is transparent to incoming photons allowing for frontside illumination. Aluminum is opaque which requires thinning and backside illumination .

Substrate shorts --> column blemish as current leaks from the conductive gate into the signal channel

Advanced Silicon processing occurs at the same time as CCD development
--> LOCOS significantly reduces substrate shorts.

Today 2048x2048 devices are standard and mosaics are being planned. As late as 1990 the largest arrays were still 800x800 (HST)

Pixels are laid down via photolithography --> precise positioning of doped regions and interconnections. 3-phase pixels with sizes of 5 microns can now be made.

QUANTUM EFFICIENCY

Frontside Illumination:

Limited by absorption of blue photons by relatively thick (5000 A) poly-silicate gates. Absorption depth for 4000 A photon is only 2000 A. Also, surface reflectivity increases with decreasing wavelength. Hence thick frontside illuminated devices have good QE only in the red.

Backside Illumination:

Thin the substrate from 300 microns down to about 10 microns. (this took about a decade to learn how to do it)

QE is much better in backside illuminated devices

Thinning process was non-uniform, corners thinner than the center --> leads to serious nonuniform response (the potato chip factor)

non-flatness produced poor focus when illuminated with beams faster than f5

After thinning the silicon oxides forming a native oxide layer about 20 A in thickness which, through a process too complex to be understood by astronomers, causes the surface to become positively charged.

For high and stable QE the backside of the CCD must be negatively charged to drive signal electrons towards the front surface. Three backside treatments:

UV flooding --> creates lots of free electrons. Some of these electrons have enough energy to escape to the back surface thus creating a small surface voltage which in turn attracts and acculumates a thin layer of holes; hole gradient sets up an intense electric field in the silicon (100,000 V/cm). But, doesn't last unless the detector stays at -95 C (or lower).

Flash Gates --> deposit mono-atomic layer of gold or platinum; causes silicon electrons to tunnel through the native oxide layer and generate a surface potential equal to the work function difference between the metal and the silicon. Unfortunately, in a vacuum positive charge is observe to buildup in the native oxide layer. So invent, biased flash gate - now you have a shutter and control of the QE! Positive bias, charge is swept towards the backside - no QE; negative bias, charge is swept to the frontside --> lots of QE

Boron doping --> ion implants provide a permanent hole layer (i.e. thin layer of hole accumulation). Manufacturing process requires heating of the device to anneal the damage down during the implantation process. This is best accomplished using a scanning high energy pulsed laser directly on the immediate backside surface

Phosphor Coatings:

Converts incoming UV photons to longer wavelength photons.

QE Notches

Can evaporate under conditions of high vacuum.

Scattering of light from Target pixel to adjacent pixels in the UV --> mostly a problem for front illuminated devices as on the backside the phosphor is in direct contact with the silicon separated only by the thin native oxide layer

multiple e-h pair generation is not possible; Phosphor only emits one visible photon per incoming photon

PERFORMANCE LIMITATIONS:

CTE -- theoretically limited by three effects

self-induced drift --> caused by mutual electrostatic repulsion of the carriers within a packet; dominates for large packets

thermal diffusion

fringing field For Roger

But for transfer rates of less than 100 kpixels/sec, theory doesn't matter and CTE is limited by four other factors that are all related to electron traps within the device:

Design Traps --> narrowing of signal channel which produces a potential barrier in which charge can be trapped. Usually happens at transfer gate between array and the horizontal register as transfer gates are tapered in order to properly transfer charge to the horizontal register. Hence a trap (potential barrier) is created at the front of the gate. This is most noticeable when low-level point sources are imaged. As the packet passes through the transfer gate, the charge is redistributed into a "tail" of trailing pixels. This is fatal for HST detectors. Problem is temporarily solved by a fat-zero which fills the trapping regions thus allowing signal charge to pass through unimpeded. But this increases the read noise of the detector. Note, about 50,000 HST CCDs were fabricated and tested before this problem was noticed and characterized. Now CCDS are designed so that the signal carrying channel is constricted at the end of beginning of the transfer gate as opposed to mid-phase.

Process Induced Traps --> Localized Defects which are usually randomly distributed. Okay if the defect is in a vertical column but is fatal if its in the horizontal register. This traditionally has been a main cause of low sensor yield.

Bulk Traps --> lattice defects or deep-level metallic impurities in the substrate. If these traps lie in the charge transfer channel then trapping occurs. Ultimately, CTE is bulk state limited. Bulk traps usually become active at low operating temperature when emission time constant of the trap is equal to the charge transfer time from one phase to the next. So, speed up this time either by operating at higher temperature or adjusting clockrates.

Radiation Induced Traps: Energetic particles displace silicon atoms from the lattice structure. These then act like bulk traps. This is mostly a problem for space based CCDs.
Testing CTE --> this is kind of Cool

Use Fe-55 which produces a 5.9 Kev photon. When this interacts with the CCD, 1620 electrons are generated in a volume much smaller than a pixel. Then see if 1620 electrons come out of your amplifier. For a 2048x2048 and a CTE of .999999, 41 electrons will be lost and this, in fact, can be measured now!

READ NOISE:

On-chip Amplifier Noise: combination of thermal white noise, 1/f noise and sense node sensitivity.

Sense node is the final collecting point at the end of the horizontal register. This converts charge to voltage. Typical sensitivities are 1--4 microvolts per electron

Voltage to outside world = charge/capacitance

White and 1/f noise depend on amplifier size (getting lower with larger sizes). But, as amplifier size grows its input capacitance (input comes from the sense node) also grows which lowers sense node sensitivity and increases the net read noise

After a decade of experimentation, optimal design was reached and typical readout noise is now 4-5 electrons (full well capacity is at least 100,000 electrons these days).

Note that 1/f noise is thought to be associated with surface states and therefore should be absent in buried channel CCD amplifiers, but its not. This is a mystery

Generated where current channel does interact with the surface (maybe within the amplifier electronics)

Note that if 1/f noise were not present, white noise could go to zero just by increasing the electrical bandwidth of the CCD signal processor and increasing the sampling period for each pixel.

Still the current world record is 1.5 electrons rms !

DARK CURRENT NOISE:
Thermal generation of electrons by the finite temperature of the device.

Thermal generation and diffusion in the bulk

in the depletion region

in the surface states at the Si-SiO(2) interface

Surface dark current is 2--3 orders of magnitude higher than the bulk dark current. Surface dark current depends on:

density of interface states

density of carriers (e or h) that populate the interface
Electron thermal hopping from the valence band to an interface state and then the conduction band produces an e-h pair that collects in the potential well. This can be minimized by the presence of free carriers that will fill the interface states and inhibit hopping.

In an inverted CCD (substrate potential is larger than interface potential), holes from the channel stop region migrate to the interface and eliminate hopping conduction

Multi-pinned phase (MPP) CCDs take advantage of this.

Requires special processing:

To have well capacity while inverted requires the potential of one of the phases to be offset from the others. This can be done by doping with boron beneath phase 3, which now acts as a barrier phase as it will attain inversion before phases 1 and 2. In practice, total inversion occurs at -6.5 V for phase 3 and -8V for phases 1 and 2. The 2.5 V offset is determined by the amount of boron which is implanted.

Well capacity is reduced by a factor of 2 in this case. Increasing implant dose can lead to larger well capacity

Result of MPP is to eliminate surface dark current and therefore achieve low dark count at temperatures well above 77K (also helps eliminate condensation on dewar window).

MPP devices can operate at room temperature now for 8 minutes before the dark current fills the well!

After Two Decades of R&D, CCDs aren't quite the perfect detector but they are PFG (e.g. damn close)

Standard Astronomer Disclaimer

___________________________________________________________________

The Electronic Universe Project

e-mail: nuts@moo.uoregon.edu