update on eds with the sdd and micro-structural characterization

Update on Energy Dispersive X-ray Spectrometrywith the Silicon Drift Detector (SDD) and

Microstructural Characterization with NIST Lispix

D.E. Newbury*, D.S. Bright*, J.H. Scott*,N. Ritchie*, J.R. Michael**, and P.G. Kotula**

*National Institute of Standards and Technology, Gaithersburg, MD 20899-8370**Sandia National Laboratory, Albuquerque, NM 87185-0886

[email protected]

Certain commercial equipment, instruments, or materials are identified in this talk to foster understanding. Such identification does not imply recommendation or endorsement by the National Institute of Standards and Technology, nor does it imply that the materials or equipment identified are necessarily the best available for the purpose.

Advanced Instrumental Techniques and Software Algorithms in EPMA Workshop

X-ray Mapping in the Spectrum Image Modeat Output Count Rates above 100 kHz with

the Silicon Drift Detector (SDD)

Dale E. NewburySurface and Microanalysis Science Division

National Institute of Standards andTechnology

Gaithersburg, MD 20899-8370

[email protected]

50 years of x-ray mapping: a tribute to Peter Duncumb

At the 2006 M&M Conference in Chicago, I gave the following paper:

SDD

Si(Li)

The need for speed!

The new EDS: Silicon Drift Detector (SDD)

This paper examined the performance of the SDD forvarious parameters but especially for high speed X-ray Spectrum Imaging (XSI).

SDD

Si(Li)

The need for speed!

The new EDS: Silicon Drift Detector (SDD)

This paper examined the performance of the SDD forvarious parameters but especially for high speed X-ray Spectrum Imaging (XSI).

But as I gave the paper, Iwas already aware (thanks toJuly ‘06 communication fromJoe Michael at Sandia) that myresults were stale: 100 kHz OCRwas already way behind the limiting performance envelopefor SDD technology.

What had happened?

ElectronsHoles

X-rays

Au electrode, ~20 nm

Window: Be, BN, C (diamond), or polymer 0.1 - 7 μm

Active silicon (intrinsic), 3 mm

- 1000 V

Inactive silicon (n-type), ~100nm

Rear Au electrode, ~20 nm

Si 'dead' layer (p-type), ~100 nm

Al reflective coating, 20 - 50 nm

Ice

The Si(Li) Energy Dispersive X-ray SpectrometerFitzgerald, R, Keil, K. and Heinrich, K. Science v 159 (1968) 528“Solid-State Energy-Dispersion Spectrometer for Electron-Microprobe X-ray Analysis”

EDS MappingAdvantages:1. Detects entire x-ray spectrum enabling mapping of allelements simultaneously2. The unexpected can bediscovered!3. No defocusingDisadvantages1. Poor resolution: interferences2. Lower P/B; more susceptibleto continuum effects; poorer limits of detection3. OCR: 2 kHz to 20 kHz

Al Fe Ni

50 μm

X-ray detection

Photon Energy (keV)

Inte

nsity

Tiesc

TiKα +TiKα

1.74 keV Coincidence peak

EDS Artifacts: Si escape peak and coincidence peak

Only for majorelements at highcount rate

Escape peakalways formsat same fraction

Spectrum appears to be continuously measured at all energies,but it is actually one photon at a time: deadtime

SiK-edge

Drastic decreasein detector efficiency

0

500

1000

1500

2000

0 2000 4000 6000 8000 10000 12000

Conventional EDS (time constant 50 microseconds)O

utpu

t Cou

nt R

ate

(cou

nts/

seco

nd)

Input Count Rate (counts/second)

Current State-of-the Art Si(Li)-EDS (30 mm2)

Resolution = 134 eV (MnKα)

Target: Mn E0 = 20keV

OC

R

ICR

“Best resolution”Max OCR ~ 1.9 kHz

“Paralyzable deadtime”

0

10000

20000

30000

40000

50000

60000

70000

100

120

140

160

180

200

220

0 20 40 60 80 100

Manganese (Beam energy 20 keV)

OCR at 40%DT

FWHM (MnKa)

Out

put C

ount

Rat

e at

40%

dea

dtim

eR

esolution (FWH

M at M

nKα)

Time Constant (microseconds)

Current digital signal processing with conventional Si-EDS enables count rates > 30 kHz (40% DT) with FWHM of ~180 eV (MnKα)

Si-EDS (30 mm2)

0

5000

10000

15000

20000

25000

0 20000 40000 60000 80000 100000 120000 140000

Conventional EDS (time constant 4 microseconds)O

utpu

t Cou

nt R

ate

(cou

nts/

seco

nd)


Current State-of-the Art Si(Li)-EDS (30 mm2)

Resolution = 178 eV (MnKα)

Target: Mn E0 = 20keV

“Mapping throughput”Max OCR ~ 22.5 kHz

ICR

OC

R

SDD Backsurface

Central anode,80 μm diameter

Resistor bridge

Ring electrodes

300 μm

SDDs are thin!

SDDs have a complexback surface electrodestructure. Appliedpotential creates internal collection channel

Active area 5 mm2 to 100 mm2

X-rays

The anode of an SDDis ~ 0.005 mm2 for a 50 mm2 detector, about1/10,000 the area of EDS

Silicon Drift Detector: Same detection physics as Si(Li)-EDS, but a radically different design X-ray absorption and ejection of

photoelectron; inelastic scatteringcreates electron-hole pairs.

SDD EDS with thermoelectric cooling to ~ -20 to -50 oC

Conventional Si(Li) EDS withLiquid nitrogen cooling to~ -190 oC

100

150

200

250

0

50

100

150

200

250

300

350

0 1 2 3 4 5 6 7 8

Radiant Detectors SDD Performance

FWHM (MnKa)

OCR(40%DT)

y = 173.93 * x^(-0.1418) R= 0.9744

y = 94.748 * x^(-1.0105) R= 0.98328

FWH

M (M

nKa)

Output C

ount Rate (40%

DT)

Time Constant (microseconds)

134 eV

3rd generation Radiant “Vortex” 50 mm2 SDD

0

5000

10000

15000

20000

0 20000 40000 60000 80000 100000

SDD with 8 microsecond time constant

Out

put c

ount

rate

(cou

nts/

seco

nd)

Input count rate (counts/second)

MnKα FWHM = 134 eV50 mm2 8 μs shaping time

Limit of conventional Si(Li) EDSat optimum resolution

0

500

1000

1500

2000

0 2000 4000 6000 8000 10000 12000

Conventional EDS (time constant 50 microseconds)

Out

put C

ount

Rat

e (c

ount

s/se

cond

)


Peak OCR = 13.6 kHz

Si(Li)

0 10 0

5 10 4

1 10 5

1.5 10 5

2 10 5

0 10 0 2 10 5 4 10 5 6 10 5 8 10 5 1 10 6 1 10 6

SDD 1 microsecond time constantO

utpu

t cou

nt ra

te (c

ount

s/se

cond

)

Input count rate (counts/second)

MnKα FWHM = 163 eV50 mm2 1 μs shaping time

Peak OCR = 160 kHz

0 10 0

1 10 5

2 10 5

3 10 5

4 10 5

5 10 5

0 10 0 5 10 5 1 10 6 2 10 6 2 10 6

SDD Time constant = 500 nanosecondsO

utpu

t cou

nt ra

te (c

ount

s/se

cond

)


MnKα FWHM = 188 eV50 mm2 500 ns shaping time

Peak OCR = 280 kHz

0 10 0

1 10 5

2 10 5

3 10 5

4 10 5

5 10 5

6 10 5

0 10 0 5 10 5 1 10 6 2 10 6 2 10 6 3 10 6 3 10 6 4 10 6

SDD Time constant = 250 nanosecondsO

uput

Cou

nt R

ate

(cou

nts/

seco

nd)


MnKα FWHM = 217 eV50 mm2 250ns shaping time

Peak OCR = 510 kHz

10 3

10 4

10 5

10 6

10 3 10 4 10 5 10 6 10 7

Input Count rate (c/s)

Out

put C

ount

rat

e (c

/s)

8 μs

1 μs500 ns

250 ns

Radiant Detectors 50 mm2 VORTEX SDD

Detector solid angle plus processing speed revolutionizes x-ray mapping!

The Real Promise of the SDD: High Speed Spectrum Image Mapping

Radiant SDD/SAMx spectrum imaging software128x128x10ms (1.3 ms overhead) = 185 s total

50 mm2 detector; 500 ns time constant; Res = 188 eV(MnKα)

E0 = 20 keV i = 10nAICR: 320 kHz OCR: 220 kHz ~40%DT

“Three-minute eggx-ray spectrum image”

X-ray spectrum imagingA complete spectrum is recorded at each pixel!

X-ray E

nergy

X

Y

Channel (X-ray Energy)

Cou

nts

Cou

nts

Gorlen et al.,Rev.Sci.Inst. v55(1984) 912.

20μm

Raney Nickel AlloyE0 = 20 keV 10 nATC = 500 ns (188 eV at MnKα)128x128 10 ms per pixel Mapping 185 sec

H

I

Al

Al

Ni

Fe

BSEPhasesAl 99.5 Ni 0.5Al 71.2 Ni 24.6 Fe 4.2 Al 60 Ni 40 “I”Al 46.5 Ni 53.5 “H”

Update on Silicon Drift Detector (SDD) Performance

• Processing speed

Bruker SDD Performance

Detector rise time <100nsOptimum shaping time 700 ns (90 kcps max. throughput)Maximum cooling -25°C (at Iopt=2,5A, ta = 23°C, still air)Thermostat set point -20°C

Values for sum spectrum over all 4 detectors (measured at MnKa)Peak shift (20 ... 800 kcps)<5eVPeak drift (24h) <10eV Peak/Background 2200:1 Energy resolution and processed countrate of 5.899 keV, Mn k

Joe Michael’s news: he had obtained a quad detector with this performance!

Source: Bruker

100x fasterthan Si(Li)!

0

200

400

600

800

1000

1200

1400

0 200 400 600 800 1000 1200 1400

220 ns

400 ns

680 ns

Input Count Rate (kcps)

Out

put C

ount

Rat

e (k

cps)

0

10

20

30

40

50

60

70

80

0 200 400 600 800 1000 1200 1400

220 ns

400 ns

680 ns

Input Count Rate (kcps)%

Dea

d Ti

me

The latest SDD (Bruker 4th generation)

Bruker Quad SDDData: Joe Michael, Sandia National LabAlbuquerque, NM

Bruker: single SDD performance

Four 10 mm2 SDDs with multiplexing givesOCR > 1 MHz with resolution < 140 eV.

127 eV

131 eV

150 eV

0

200

400

600

800

1000

1200

1400

0 200 400 600 800 1000 1200 1400

220 ns

400 ns

680 ns


Out

put C

ount

Rat

e (k

cps)

0

10

20

30

40

50

60

70

80

0 200 400 600 800 1000 1200 1400

220 ns

400 ns

680 ns

Input Count Rate (kcps)%

Dea

d Ti

me

But wait! The latest SDD (Bruker 4th generation)

Bruker 10 mm2 single SDD (from P-N Sensors)

127 eV

131 eV

145 eV

Si(Li) EDS “best resolution”OCR vs ICR performance disappears on this plot!!(covering only 1% of the0 – 200 unit spacing)

Data: Joe Michael, Sandia National LabAlbuquerque, NM

Four 10 mm2 SDDs with multiplexing givesOCR > 1 MHz with resolution < 140 eV.

.Even thisperiod istoo big!

0

500

1000

1500

2000

0 2000 4000 6000 8000 10000 12000

Conventional EDS (time constant 50 microseconds)

Out

put C

ount

Rat

e (c

ount

s/se

cond

)


Si(Li)

Bruker “Tear–drop” Shaped SDDOCR speed: On-chipintegrated chargeamplifier eliminatesmicrophonics; off-centerlocation protects amplifierfrom x-rays

Why is this SDD so much faster?

M&M2005_Optimized Readout Methods of Silicon Drift Detectors for High Resolution Spectroscopy in Micro-Beam AnalysisH.Soltau*, P.Lechner*, A. Niculae*, G. Lutz**, L. Strüder**, C. Fiorini*** and A. Longoni ***R. Eckhard,* G. Schaller,** and F. Schopper**

0 1 2 3 4 5 6 7 8 9

FeK

α

FeK

β NiK

α

NiK

β

CrK

α

CrK

βM

nKα

TiK

α

CaK

α

CaK

β

KK

α

SKα

PKα

SiK

αA

lKα

MgK

αN

aKα

OK

αFe

L

CK

α

NiL

FeS Si

keV

This is a 12 second x-ray spectrum image (128x96 pixels)!

QUAD SDDat 500 kHz OCR

Examining peaks inThe SUM PIXELSpectrum (filled):

50 μm


• Processing speed• Spectral quality

– Resolution and peak position stability

100

150

200

250

0 10 20 30 40 50

8 microseconds1 microsecond0.5 microsecond0.25 microsecond

y = 133.75 + 0.0030488x R= 0.11043

y = 162.98 + 0.10743x R= 0.97422

y = 187.5 + 0.16057x R= 0.98415

y = 217.48 + 0.38724x R= 0.99124

Res

olut

ion

(FW

HM

)

Deadtime (%)

Slope = 0.003 eV/1%DT




0.25 μs, 217 eV

0. 5 μs, 188 eV

1 μs, 163 eV

8 μs, 134 eV

8 μs1 μs0.5 μs0.25 μs3rd generation

Radiant50 mm2 SDD

Note constancyof peak shape with input count rate!This is vital for spectral studies, includingspectrum imaging.

130

132

134

136

138

140

0 10 20 30 40 50 60

Manganese (Beam energy 20 keV)R

esol

urio

n (F

WH

M M

nKα

)

Deadtime (%)

For a given choice of time constant, the resolutionshould remain constant with deadtime.

Current state-of-the-art Si(Li)-EDS (30 mm2)

3 eV

5.9

5.905

5.91

5.915

5.92

0 50 100 150 200 250 300

8 mus1 mus500 ns250 ns

Peak

Ene

rgy

(keV

)

Input Count Rate (kHz)

Peak Channel Stability3rd generationRadiant50 mm2 SDD

One10-eVchannel

1 eV

1 eV

220 ns

680 ns

400 ns

120

130

140

150

160

0 200 400 600 800 1000 1200 1400

Res

olut

ion

Mn

Kα

(ev)


-20

-15

-10

-5

0

5

10

15

20

0 200 400 600 800 1000 1200 1400

220 ns

680 ns

400 ns


Peak

Shi

ft M

n K

α (e

v)

Bruker Quad SDDData: Joe Michael, Sandia National LabAlbuquerque, NM

Stability of Resolution and Peak Position

Bruker “Tear–drop” Shaped SDDOCR speed: On-chipintegrated chargeamplifier eliminatesmicrophonics; off-centerlocation protects amplifierfrom x-rays

Why is this SDD so much faster?

Peak stability: pulse reset operationwith reset diode integrated into the readout anode on the chip.M&M2005_Optimized Readout Methods of Silicon Drift Detectors for High Resolution Spectroscopy in Micro-Beam AnalysisH.Soltau*, P.Lechner*, A. Niculae*, G. Lutz**, L. Strüder**, C. Fiorini*** and A. Longoni ***R. Eckhard,* G. Schaller,** and F. Schopper**M&M2006_Optimization of the Peak-to-Background Ratio and the Low Energy Response of Silicon Drift Detectors for High Resolution X-ray SpectroscopyA. Niculae*, H.Soltau*, P.Lechner*, A.Liebl*, G. Lutz**, L. Strüder**, A. Longoni***R. Eckhard*, G. Schaller** and F. Schopper**



– Resolution and peak position stability– Low photon energy peaks

BC O

Photon Energy (keV)

Inte

nsity

(cou

nts)

0 0.25 0.5 0.75 1.0 1.25 1.50 1.750

9200Mg

Radiant Vortex SDD8 μs TC

Low photon energy performance

CaF2E0 = 5 keV8 μs shaping time

0 0.25 0.5 0.75 1.0 1.25 1.50 1.75 2.0 2.25 2.5

Photon Energy (keV)

Inte

nsity

(cou

nts)

0

15197F K

Radiant Detectors “VORTEX”


CaCO3E0 = 5 keV8 μs shaping time

O K

C K

0 0.25 0.5 0.75 1.0 1.25 1.50 1.75 2.0 2.25 2.5

Photon Energy (keV)

Inte

nsity

(cou

nts)

0

37347



AlN7 nm carbonE0 = 5 keV8 μs shaping time

0 0.25 0.5 0.75 1.0 1.25 1.50 1.75 2.0 2.25 2.5

Photon Energy (keV)

Inte

nsity

(cou

nts)

0

8601AlK

O K

N K

C K



CarbonE0 = 5 keV8 μs shaping time

0 0.25 0.5 0.75 1.0 1.25 1.50 1.75 2.0 2.25 2.5

Photon Energy (keV)

Inte

nsity

(cou

nts)

0

12204

SiKabC dbl

C K



CarbonFill = C (ideal) strippedE0 = 5 keV8 μs shaping time

0 0.25 0.5 0.75 1.0 1.25 1.50 1.75 2.0 2.25 2.5

Photon Energy (keV)

Inte

nsity

(cou

nts)

0

12204

SiKabC dbl

C K



CarbonE0 = 5 keV8 μs shaping time

??? = B K

0 0.25 0.5 0.75 1.0 1.25 1.50 1.75 2.0 2.25 2.5

Photon Energy (keV)

Inte

nsity

(cou

nts)

0

12204

SiKabC dbl

C K



0

6916

0 0.25 0.5 0.75 1.0 1.25 1.5

MnLα

Photon Energy (keV)

Inte

nsity

(cou

nts)

MnE0 = 20 keVSDD (131 eV at MnKα)

MnLl

OCR5 kHz10 kHz20 kHz50 kHz100 kHz150 kHz200 kHz250 kHz

Spectral stability in low photon energy range



– Resolution and peak position stability– Low photon energy peaks– Coincidence peaks

TC = 1 μs 45% DT 107,000 cps OCR28% DT 72,500 cps OCR16% DT 42,300 cps OCR8% DT 21,000 cps OCRResolution = 166 eV at MnKα

Al+

NiL

Al+

Al

Ni+

Al

Al

NiL

FeK

α

NiK

α

NiK

β

ClK

α

Photon Energy (keV)

Cou

nts

3rd Generation SDD (Radiant Detectors LLC)

Al+

NiL

Al+

Al

Ni+

Al

Al

NiL

FeK

α

NiK

α

NiK

β

ClK

α

Photon Energy (keV)

Cou

nts

Oxford Analog 48% DT 30,600 cps OCR38% DT 26,200 cps OCR 22% DT 17,200 cps OCR8% DT 9,900 cps OCRResolution = 166 eV at MnKα

Al+

NiL

Al+

Al

Ni+

Al

Al

NiL

FeK

α

NiK

α

NiK

β

ClK

α

Photon Energy (keV)

Cou

nts

EDAX Genesis DSP TC=4 μs48% DT 38,800 cps OCR36% DT 26,100 cps OCR 20% DT 12,900 cps OCR10% DT 6,600 cps OCRResolution = 173 eV at MnKα

SDD can provide the OCR speed

• To exploit this OCR speed for mapping (e.g.,by x-ray spectrum imaging), we need:

• Suitably intense x-ray production, that is, anSEM that can deliver high beam currents (10 –1000 nA) and a specimen that can withstandhigh dose without degradation.

SDD can provide the OCR speed

• To exploit this OCR speed for mapping (e.g.,by x-ray spectrum imaging), we need:

• Suitably intense x-ray production, that is, anSEM that can deliver high beam currents (10 –1000 nA) and a specimen that can withstandhigh dose without degradation.

• Low (ideally “no”) overhead x-ray eventprocessing (try to avoid wasting time sortingand storing “no information” channels)

Al Ni Fe

10 ms

5 ms

2 ms

1 ms

100 μm

How fast?

217 s192 s 12%

121 s 96 s 21%

63 s 38 s 40%

44 s 19 s 57%

160x120 pixels+1.3 msoverhead

SDD can provide the OCR speed• To exploit this OCR speed for mapping (e.g., by x-ray

spectrum imaging), we need:• Suitably intense x-ray production, that is, an SEM that

can deliver high beam currents (10 – 1000 nA) and aspecimen that can withstand high dose withoutdegradation.

2. Low (ideally “no”) overhead x-ray event processing:use “position tagged spectrometry” implemented as“event streaming in hardware”, e.g., see:

Efficient handling of data stream from SDD: EventStreaming (No overhead; move units of information)

• X-ray events are outputted directly from the x-raypulse processor’s auxiliary interface bus into the x-yscan generator. The events are then interpreted andcombined with the x-ray position information into pixelevents. The pixel events are packetized andstreamed to a host computer where they are bufferedand stored. The whole process takes place at thehardware level with an extremely high speed and isthus highly efficient. (Pixel overhead μs rather thanms)

High Speed Spectrum Imaging of Raney Nickel Alloy Using a Large AreaSilicon Multi-Cathode Detector (Vortex-EMTM) and Event Streaming TechniqueM&M 2006L. Feng*, V. D. Saveliev*, S. Barkan*, C. R. Tull*, M. Takahashi*, N. Matsumori*, S. D.Davilla**, J. S. Iwanczyk***, D. E. Newbury**** and J. A. Small****

SDD can provide the OCR speed• To exploit this OCR speed for mapping (e.g., by x-ray spectrum

imaging), we need:• Suitably intense x-ray production, that is, an SEM that can deliver

high beam currents (10 – 1000 nA) and a specimen that canwithstand high dose without degradation.

2. Low (ideally “no”) overhead x-ray event processing: use “positiontagged spectrometry” implemented as “event streaming inhardware”,

With these conditions reasonably satisfied:Specimen: Portland St. MeteoriteOCR = 550kHz Quad SDD, resolution MnKα < 140 eV128 x 96 pixels; 1 ms dwell per pixel = 12.3 seconds

0 1 2 3 4 5 6 7 8 9

FeK

α

FeK

β NiK

α

NiK

β

CrK

α

CrK

βM

nKα

TiK

α

CaK

α

CaK

β

KK

α

SKα

PKα

SiK

αA

lKα

MgK

αN

aKα

OK

αFe

L

CK

α

NiL

FeS Si

keV



Examining peaks inThe SUM PIXELSpectrum (filled):

50 μm

S

Si

Fe

S Fe Si

50 μm


0 1 2 3 4 5 6 7 8 9

FeK

α

FeK

β NiK

α

NiK

β

CrK

α

CrK

βM

nKα

TiK

α

CaK

α

CaK

β

KK

α

SKα

PKα

SiK

αA

lKα

MgK

αN

aKα

OK

αFe

L

CK

α

NiL

CaAl Ni

keV

50 μm

Examining peaks inThe MAX PIXELSpectrum (line):



S Ni

Fe

50 μm


Al

P Ca

Cr Ni Mn

Ti

Mg O

50 μm

This is a 12 second x-ray spectrum image!

With x-ray spectrum imaging now possible in10 to 100 seconds with optimized SDD:

• 1. How can we recover information buried in these ~ 100Mbyte – 1 Gbyte files?– Sophisticated statistical approach, see:

- Simple minded, operator intensive approach:NIST LISPIX (available for free at www.nist.gov; search “LISPIX”)

LISPIX is a comprehensive image manipulation software engine thatincludes many tools that are useful for the study of x-ray spectrumimages and extraction of information.

LISPIX “Derived Spectrum” Tools

• Two views of the x-ray spectrum image (XSI)– As an array of true spectra, one at each pixel

X-ray spectrum imagingA complete spectrum is recorded at each pixel!

X-ray E

nergy

X

Y


Cou

nts

Cou

nts

Gorlen et al.,Rev.Sci.Inst. v55(1984) 912.


• Two views of the x-ray spectrum image (XSI)– As an array of true spectra, one at each pixel– As a “card deck” of x-ray images, each 10-eV wide

Cube slices are x-raymaps (images)

X-ray E

nergy

X

Y

The x-ray spectrum image can be viewed as a card deck of x-rayimages, each a 10-eV energy slice



• Basis of Lispix approach: “derived spectra” SUM Spectrum– Add all counts on a card to find intensity for that keV

3/5/2004

Sum Spectrum:Add the counts fromevery pixel in eachenergy plane; theSUM from a planebecomes the intensityin the correspondingenergy channel of thederived spectrum.

Derived x-ray spectra are calculated from measured spectra within the cube.

X-ray E

nergy

X

Y

Cou

nts


ReferencesGorlen et al., Rev.Sci.Inst. 55 (1984) 912.

Jeanguillaume and Colliex,Ultramicroscopy, 28 (1989)252.

NiKαNiKbα

AlKα

NiL



• SUM Spectrum– Add all counts on a card to find intensity for that keV– Peaks identify high abundance features

Peaks in the SUM spectrum correspond to x-ray lines of elements.

Single ‘slice’ or x-ray channel

Multichannel image for Ni peak.

Peaks in sum spectrum correspond to dominant features.

Cou

nts


LISPIX constructs the image corresponding toan energy plane or group of energy planes.

Developing Derived Spectrum Tools

• SUM spectrum quickly locates spectralfeatures with high abundance

• Problem: How do we find rare, unanticipatedfeatures in the datacube (200 Mbytes):– Rare feature may occur only at one pixel. For a

256x200 scan, one pixel represents 1/51,200– Any element (except H, He, Li) may be of interest

Test every pixel in anenergy plane to findthe MAXIMUMvalue. Thismaximum valuebecomes the intensityin the correspondingenergy channel of thederived MaximumPixel Spectrum.

Maximum Pixel Spectrum

X-ray E

nergy

X

Y

Cou

nts


Bright and Newbury,J. Microscopy, 216 (2004) 186

10 μm

Cr

Al Cr Ni




• MAX PIXEL Spectrum– Find maximum value on a card and plot for that keV– Locates rare, unanticipated features

Time: 123 seconds

Time: 12.3 seconds Of couse, Max Pixel islimited by thestatistics of asingle pixel

LISPIX “Derived Spectrum” Tools• Two views of the x-ray spectrum image (XSI)

– As an array of true spectra, one at each pixel– As a “card deck” of x-ray images, each 10-eV wide


• MAX PIXEL Spectrum– Find maximum value on a card and plot for that keV– Locates rare, unanticipated features– May not be sensitive to dilute constituents

• RUNNING MAX Spectrum– First average over n-m to n+m cards (e.g., m = 3) to create new

naverage card– Then perform maximum value search on naverage card

RUNNING MAX 12.3 s XSI


Gone!




• MAX PIXEL Spectrum– Find maximum value on a card and plot for that keV– Locates rare, unanticipated features– May not be sensitive to dilute constituents

Cl CrK, Ag, or Cd

Maximum pixel spectrum

Log of the SUM SpectrumC

ount

sL

og10

Cou

nts


NiKα

Al

NiL

Fe

CO

SiO

CAl

NiL NiKα

NiKβ

NiKβCuKα

Where is the Iron?

Fe is not visible in single slice

Fe is confined to one type of region.

Finding dilute iron with the Log of the Sum Spectrum

Cou

nts


NiKα

NiKβ

Developing Strategy for Rapid Mapping

• How many counts are “enough” to establish visibilityfor an object in a compositional map?


• The answer depends on several factors:1. The concentration, C (mass fraction)2. The characteristic x-ray energy

a. Overvoltage, U0 = E0/Ec Ich ~ (U0- 1)n

b. Specimen self-absorption I/I0 = exp [-(μ/ρ) ρ s]c. X-ray detector efficiency

3. X-ray background (continuum) Icm~ Z (U – 1)4. The level of compositional contrast, e.g. vs. Bkg or vs. a different concentration.


128x96 1 ms dwell, 10 frames 123 s total 550kHz OCR

Fe

Ni

Al 10 μm

PhasesAl 99.5 Ni 0.5 Al 60 Ni 40Al 71.2 Ni 24.6 Fe 4.2 Al 46.5 Ni 53.5

Almax=3846

Nimax=662

Femax=93

This dose gives sufficient Fe countsto perform a background correction.

Fe raw

BackgroundCorrected;Fe Contrast vsbackground

128x96 1 ms dwell, 10 frames 123 s total 550kHz OCR

Al Fe

Ni Al Fe Ni

10 μm


Femax= 93

Reduce dose by 10; 128x96 1 ms dwell, 1 frame 12.3 s total 550kHz OCR

Al Fe

Ni Al Fe Ni

10 μm


Femax= 9

128x96 512 μs dwell, 1 frame 6.2 s total 550kHz OCR

Al Fe

Ni Al Fe Ni

10 μm


Femax= 5


Al Fe

Ni Al Fe Ni

10 μm


Femax= 3


Al Fe

Ni Al Fe Ni

10 μm


Femax= 2

n1

n2

n


= P= P+B

= B

For an element localized in one phase

• How many counts are “enough” to establish visibility for an object in acompositional map?

• This issue is closely related to the classic Rose (1948) criterion for contrastvisibility threshold in scanned images. For an object relative to a featurelessbackground, Rose requires the signal to exceed the noise by a factor (“Rosefactor”, RF) of at least 5:

Signal = (n2 – n1) > RF * σn (noise) = RF n11/2

Consider the case of a compositional feature containing element A viewedagainst a featureless background that does not contain element A

For x-rays, n1 = B (continuum) while n2 = P (characteristic) + B

Signal = (n2 – n1) = (P+B – B) = P > RF B1/2

P/B > RF B1/2/B

P/B > RF/B1/2

P/B > 5/B1/2

0

5

10

15

20

25

30

35

40

0 2 4 6 8 10 12

P/B

E (keV)

K

Al

SiP

S

ClMg

CaSc

Ti

V CrMn Fe

Co NiCu

Zn GaGe As Se

Measurements of Peak-to-Background (P/B)

Si(Li) EDS 134 eV at MnKαPure elements (or equivalent)E0 = 20 keV Kα+β peaks

0.01

0.1

1

10

100

0 2 4 6 8 10 12

P/B10% Conc1% Conc

P/B

E (keV)

Pure elements (or equivalent)E0 = 20 keV Kα+β peaksDiluted by element Z+1K

AlSiPSClMg

CaSc

TiV Cr

MnFe Co Ni

Cu Zn GaGe As Se Pure

0.1 mass frac

0.01 mass frac

Measured and Estimated of Peak-to-Background (P/B)

Pure

0.01

0.1

1

10

100

0 2 4 6 8 10 12

P/B10% Conc1% Conc

P/B

E (keV)

Pure elements (or equivalent)E0 = 20 keV Kα+β peaksDiluted by element Z+1K

AlSiPSClMg

CaSc

TiV Cr

MnFe Co Ni

Cu Zn GaGe As Se

B = 25 counts

Pure

0.1 mass frac

0.01 mass frac

B = 2500 counts

P/B > 5/B1/2

Measured and Estimated of Peak-to-Background (P/B)

Pure


• The answer depends on several factors:1. The concentration, C (mass fraction)2. The characteristic x-ray energy

a. Overvoltage, U0 = E0/Ec Ich ~ (U0- 1)n

b. Specimen self-absorption I/I0 = exp [-(μ/ρ) ρ s]c. X-ray detector efficiency

3. X-ray background (continuum) Icm~ Z (U – 1)4. The level of compositional contrast, e.g. vs. B or vs.

a different concentration.• A compositional image simulator is needed:

– DTSA for spectral details, such as P/B (soon NIST Monte)– LISPIX to synthesis and process the images for display


K961O 0.470Na 0.030Mg 0.030Al 0.058Si 0.299P 0.002K 0.025Ca 0.036Ti 0.012Mn 0.003Fe 0.035

E0 = 20 keV0-20keV: ~1,000 counts

0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0

Photon Energy (keV)

Cou

nts

57

0

K KαP = 14B = 18P/B = 0.8

Desktop Spectrum Analyzer (DTSA) Simulation

Feature: 1 pixel (Rose criterion) = 0.4% of image width (256 pixels)

Size Matters! Effect of Feature Size on Visibility

P/B = 0.2

P/B = 0.4

P/B = 0.6

P/B = 0.8

P/B = 1

P/B = 2

P/B = 4

Detection

B = 25 counts

P/B = 0.2

P/B = 0.4

P/B = 0.6

P/B = 0.8

Feature: 5 pixels square = 2% of image width (256 pixels)

Detection

Determineshape

P/B = 1


B = 25 counts

Feature: 64 pixels square = 25% of image width (256 pixels)

P/B = 0.2

P/B = 0.4

P/B = 0.6

P/B = 1

Determineshape

Detection


B = 25 counts See: Bright, D. S., Newbury, D. E., and Steel, E. B., “Visibility of objects in computer simulations of noisy micrographs”, J. Micros., 189 (1998) 25-42.

Conclusions• SDD performance enables XSI mapping in 10 to

200 seconds (512x394_1ms) if the specimencan withstand the high beam current necessaryto generate high x-ray flux.

• NIST LISPIX interactive tools enable detectionand recovery of compositional features,including those that are rare and unanticipated.

• Strategy for mapping can be developed with thespectral simulation tool in NIST DTSA and thenew image simulation tool embedded in NISTLISPIX.

[email protected]

update on eds with the sdd and micro-structural characterization

Documents