a.d. short european space agency · a.d. short european space agency radiation damage & gaia...

Radiation Damage & Gaia CCDs A.D. Short – ESTEC, Leiden 2007

Radiation Damage & Gaia CCDsA.D. Short

European Space Agency


• e2v are providing some of the largest CCDs they can make

• We will operate ~100 of them in parallel in a large focal plane which we cannot shield effectively

• We will launch close to solar maximum and place the instrument outside the Earth’s geomagnetic field where it will be exposed to a hard spectrum of solar flare protons

• Within the first year, we expect the CCDs to receive a radiation dose similar to the XMM CCD lifetime dose.

Will we be able to extract astrometric centroidingperformances equivalent to a few thousandths of one CCD pixel per CCD transit?

The Challenge


Required astrometric performance

V magnitude 10 15 20arcseconds 7 24 300

pixels in focal plane 1.19×10-4 4.07×10-4 50.9×10-4

• Gaia requirements for end of mission parallax standard errors (G2V)

• Assuming that each object is observed ~ 75 times in 9 CCDs(= 675 CCD transits), then the residual centroiding error per CCD transit is equivalent to…

V magnitude 10 15 20pixels in focal plane 0.003 0.011 0.132

Plate scale = 169.685×10-4 m. arcsec-1, 1 pixel = 10 m


Is this possible?

• It is possible to centroid to a few thousandths of a pixel given sufficient signal to noise. This has been demonstrated in the lab using a single virgin CCD.

• Since the Global Iterative Solution is designed to “solve”the entire data-set, it is a very powerful way to remove random and systematic errors. Gaia is sometimes called “self-calibrating”.

• Without the ugly reality of CCD radiation damage, Gaia performance requirements are achievable on paper….


The problem

• During the proposal and selection phases, radiation damage to the CCDs was identified as a potential major problem

• However, the real magnitude of the problem was appreciated by very few people and could not be demonstrated prior to dedicated testing

• The environment at L2 was considered by many to be relatively benign. Instrument proposals largely neglected the radiation issue


Effects of radiation damage on PSF

EXAGERATED


• Increased RMS centroiding errors– Charge loss reduces signal to noise and thereby degrades fitting– PSF is distorted so that its shape no longer matches the fitting

function• Variable centroid shift (or bias)

– Since the PSF is distorted and translated, even perfect fitting would give a centroid which is shifted

• Photometry– Directly affected by charge loss

The real problem is that these effects are all highly variable with trap occupancy

Effects of radiation damage on Gaia


TDI Motion

Most traps full following charge

injection

Lots of traps are empty prior to

charge injection

Stars close behind charge injectionmeet fewer empty traps than stars

far behind charge injection

Trailing stars meet fewer empty traps than leading stars

Effects are highly variable


Effects are highly variable as f()

•Time in the mission (radiation dose)•Position in the focal plane (per CCD or even per column)•Magnitude of each source•Distance of source behind line of charge injection•Distance behind another star, magnitude of that star and degree of overlap•Overall star density•Phasing of the PSF centroid within the TDI pixels•Across scan position of the star within each CCD•History of charge read through the readout register•Diffuse optical background•Particle rate (GCR and solar proton) during data taking•CCD temperature•Other….


So how bad is it?

• In 2003, started producing Monte-Carlo models of the Gaia CCDs which included electron trapping and de-trapping due to radiation damage.

• Values for trap parameters were taken from literature and from the results of testing conducted by Sira Electro-Optics during the demonstration phase [254.DO.28 iss 2]

• Initial estimates of the effects of radiation damage were obtained. However, there was very limited TDI mode data with which to constrain the model.


Monte-Carlo Model

• Models included trapping in both the image section and the readout register

• Could be operated in imaging mode or TDI mode• Several optical inputs could be selected including

gaussian spots, Airy disks and Gaia polychromatic PSFs• Mirror transmission curves, source spectra, CCD QE,

CCD noise etc. were all consistent with Gaia values• Diffuse optical background could be added• Charge injection could be added• Prompt particle events could be added


Treatment of trapping &de-trapping in early models

• In the presence of some density (ne) of electrons, a trap has an associated capture time constant:

• Where is the trap capture cross section and vth is the electron thermal velocity. However, in many cases it may be assumed that the electron density is high enough that trapping may be considered instantaneous. Most treatments of trapping in CCDs have assumed this and this was also the assumption in early ESTEC models.

• In addition each trap species will have an associated release time constant, rwhich is a strong function of temperature. This is used to calculate the probability per unit time that a trapped electron will be released.


electronconfinement

length

Supplementary Buried Channel

electronconfinement width

(2 for a 4 phase CCD)(2 for a 4 phase CCD)

•Traps distributed randomly throughout CCD volume

•Electrons confined to a volume with grows in proportion to their number

•Any trap which finds itself within the electron volume will instantly capture an electron

Trapping depends only upon “electron cloud”

volume and not on interaction time



• Prior to revelations from the CCN10 testing, all models developed at ESTEC were based upon this volume driven approach with various refinements to treat:

– Trapping during each of the four CCD phases individually– The action of the Supplementary Buried Channel– Trapping during transfer between phases, as well as whilst

static in a phase

etc….



Is MC model consistent with data?


MC model and Sira data (log)

MC model reproduces..

1. Observed signal loss

2. Exponential tail


Applying model to Gaia (2004/5)

Mag 13 Mag 15

Monte-carlo model for un-reddened G2V star, Gaia field point 18


After 4E9 proton equivalent dose

Mag 13 Mag 15

Monte-carlo model for un-reddened G2V star, Gaia field point 18


Model after 4e9 proton dose


First estimates of the magnitude of radiation effects

1. Run monte-carlo model a number of times for each set of conditions (typically 100 repetitions)

2. Fit each resulting LSF with calibrating Line Spread Function

3. Calculate mean centroid and standard error (running average to assess convergence)


Run model 100 times

field point=18

100x identical starting conditions


First predicted radiation effects

RMS errors of 10s to 100s of as

milli-arcsecondbiases to be

calibrated out somehow


But should we believe the model?

Statistics of non-TDI data are quite good. Gives good confidence in the model for

imaging mode, but TDI mode is different story

TDA phase gave:Trap parameters

Optimum temperature

Tests of charge injection

Centroiding accuracy

But not:Much TDI data

Direct measurement of bias in TDI

Direct measurement of charge-loss in TDI


Testing is much more difficult in TDI mode

?

Model is not well constrained by TDI mode data from the TDA phase

Results of Sira testing

(note different conditions){


…and Astrium TDI results gave different trend

Modelling suggests this downward slope may be caused by diffuse optical background


Centroid shift data told similar story

Astrium TDI data

Sira TDI data


Astrium instructed to conduct further tests

CCN10 testing conducted in the

Spring/Summer of 2006

•All in TDI mode•All -110oC•All without CI


CCN10 test configuration

Part of CCD irradiated to 4×109 protons (10MeV

equivalent)

Un-irradiated part of CCD

Part of CCD irradiated to 4×109 protons (10MeV

equivalent)

Un-irradiated part of CCD

Transitionregion

TDI motion

Optical mask~ 15 pixels

Spot 1 Spot 10


Analysis of CCN10 data at ESTEC

• The data-set comprises ~ 380 TDI passes or 3800 individual LSFs to fit

• In order to compare various different fitting methods, code was written to fit all of the data in one pass

• Apart from the fitting itself, the main problem is how to normalize and combine the data from each run to extract centroid shift and charge loss results with minimum errors


Examples of best fit to LSFs


Example of normalized mean centroids

TDI motion

Centroidshift is obvious

D.O.B. = 0


Example of centroid shift results

17700

11800

5900

0

centroidshift in 4e9 region (

arcsecondsapprox.)

G ~ 20

G ~ 18

G ~ 16

Effect of D.O.B.


Example of charge-loss results

Greater effect of D.O.B.

(slow traps)

G ~ 20

G ~ 18

G ~ 16

Q/ Can we explain this with a volume

driven model?


In a volume driven model

•Trapping depends only upon “electron cloud” volume and not on interaction time

•But in this model, the volume of a 5 electron charge packet due to DOB must be extremely small and could not have a significant effect upon trap occupancy

•Hence, the larger signal electron packets will experience the same charge loss regardless of an additional 5 electrons of DOB

•And what do we mean by the “volume” of a 3, 2 or 1 electron charge cloud anyway?

A/ I don’t think so


What Does The DOB Tell Us?

• Very small levels of DOB can fill significant numbers of traps a small number of electrons is NOT confined to a proportionally small volume

• DOB electrons are more likely to fill traps than an equivalent number of signal electrons exposure time is a key driver (DOB is always present whilst signals are transient)

• DOB is more effective against slow traps and has little effect on fast traps traps reach an equillibrium state of occupancy according to their capture and release time constants under given conditions

• The effect of DOB tends to saturate entire trap species are neutralized implying that DOB electrons can reach all parts of the pixel volume and confirming that time constants are critical


Density Driven Model

• These factors suggest a different trapping mechanism in which capture is driven by electron density rather than electron cloud volume

• For integrating applications, the distinction may be academic since the electron density will generally be high enough to give instant trapping

• However, the distinction is critical for Gaia because in TDI mode, we are concerned with the behaviour of extremely small signal levels (integrating from 0 electrons)


Density driven trapping

NO DOB – All traps are empty when the signal electrons pass through

WITH DOB – Electrondensity is very low, so trap capture time constants are long. Probability of capture per unit time is low, but DOB is present all of the time so some traps will be occupied by DOB electrons when the signal electrons pass through


Now model electron density distributions


Can we model CCN10 data trends?•Presented to GST in Feb 07

•Required 3+ trap species

•Model becoming extremely complex and slow

•Failed to find simultaneous good fit to centroid shift data (manually)

•We need a simpler, faster model to implement in fitting algorithms and then in IDT

This model can reproduce the effect of DOB. However…


We need fast, simple models

• The full Monte-Carlo model is too slow, cumbersome and un-constrained (it’s a mess)

• We need analytical models based on the same physics which are fast enough to employ in fitting algorithms and in Initial Data Treatment

• This is where I believe the ELSA students can really get started


The Holy Grail

Need a simple, reliable model that reproduces LSF shape, charge loss and centroid shift…..

….as f (trap occupancy)


My attempts not too convincing yet

GAIA-CH-TN-ESA-AS-012-1


What about the dispersive instruments?

Good News: In a new CCD Astrium are able to measure features in spectra with signals as small as 1 electron per sample (GRVS = 15.8 for a G2V star)

Spectral feature


At the moment, the limited test results do not look too good

Bad News: Spectral features are absent in a CCD irradiated with 4×109 protons(10MeV equivalent)

It appears that fast traps rather than slow traps are the main problem. Why?....


TDImotion

Prior to transfer through CCD: Initial RVS or photometer spectrum

After transfer through CCD: Effect of slow traps is to remove electrons (mainly) from front

Transfer through CCD: A slow trap holds an electron for a long time and becomes inactive

Spectral absorption features

Slow traps in RVS, RP and BP


TDImotion

Prior to transfer through CCD: Initial RVS or photometer spectrum

After transfer through CCD: The effect of fast traps is to “smooth out” spectral features

Transfer through CCD: A fast trap can only hold an electron for a short time (by definition)

Spectral absorption features

Fast traps in RVS, RP and BP


Conclusions

• We need fast, simple models for trapping and de-trapping effects as a function of trap occupancy which can be verified by fitting test data and then implemented in IDT

• The results of test data (especially small signals and the effects of DOB), give lots of clues regarding the trapping mechanism which are NOT consistent with a simple volume driven approach

• The dispersive instruments have their own special considerations and problems

a.d. short european space agency · a.d. short european space agency radiation damage & gaia...

Documents