Silicon Drift Detectors: Understanding the Advantages for EDS Microanalysis Patrick Camus, PhD Applications Scientist March 18, 2010
EDS Detector Requirements Detect whole energy range of x-rays 50 ev (Li-K) to incident beam energy SEM: up to 30 kev (S)TEM: up to 300 kev, realistically up to 50 kev Lower limit set by intrinsic noise of system Good spectral resolution Separate closely spaced energy peaks Varies with x-ray energy Theoretical limit based on sensor material, sensor design and system noise High x-ray detection rates Reduced collection times Fast sensor response to reduce detection overlap Physical geometry to maximize x-ray collection Shorten the sample-to-detector distance Leave enough physical room for other detectors and accessories 2
EDS Sensor Technologies Si(Li) Diode Silicon Drift Entrance window INPUT X-RAYS INPUT X-RAYS Not to Scale Different geometries provide different absorption and electrical pulse characteristics 3
Comparison of SDD to Si(Li) Technology SDD Fabrication Technology Semiconductor Operating Temperature 230~250 K at the sensor Peltier cooled Most have convective cooling Some have fans (vibration) Sensor thickness 0.5 mm Good sensitivity up to 10 kev Reduced sensitivity until 20 kev Electric Field and Electron Path Radial field Low capacitance Si(Li) Fabrication Technology Discrete components Operating Temperature 77~130 K at the sensor LN 2 Peltier+Water or refrigeration Sensor thickness 3 mm Good sensitivity up to 20 kev Reduced sensitivity until 50 kev Electric Field and Electron Path Axial field High capacitance 4
Si(Li) Detector Features Sensitive to ~70 ev (Be-K) x-rays Good sensitivity above 20 kev Parallel-plate contacts High capacitance Medium throughput rates Extreme cooling required Reduce electronic noise Good spectral resolution: 129 ev @ Mn-Kα Degrades substantially with increased input count rate Limits usable throughput maximum 5
Si(Li) Detector Performance Spectral Resolution Shorter shaping times means degraded resolution 100µs, 4µs, 2µs, 1µs 129 ev to >200 ev 5k cps input to 130k cps input 25000 20000 Throughput curve 15000 Shorter shaping times produces more output at same input 10000 Increased dead time % cause peak 5000 Operation past peak is counterproductive 0 Output Count Rate (cps) 4 us 14 us 30 us 50 us 100 us 0 20000 40000 60000 80000 100000 Input Count Rate 6
History of Silicon Drift Detectors (Abridged) 1983 Emilio Gatti and Pavel Rehak (Brookhaven National Lab) Silicon Drift Chamber 1995 Rontec (now Bruker) in cooperation with MPI and Ketek introduce EDS and XRF SDD detectors 1997 Photon Imaging (now Seiko) introduce XRF SDD detector 2000-2007 All major EDS companies introduce SDD technology for electron beam instruments Notable EDS Dates 2004 Peak and resolution stability from Thermo Fisher Scientific 2006 Quad 10mm 2 detector from Bruker AXS 2006 30mm 2 SDD from Thermo Fisher Scientific 2007 Good low-energy (Be-K) performance from Thermo Fisher Scientific 2008 80mm 2 SDD from Oxford Instruments 7
Silicon Drift Detector Features High input count rate capability up to 10 6 cps. Comfortable operation Peltier cooling @ -10 C to -60 C Good energy resolution down to 124 ev @ Mn-Kα Maintains good energy resolution as input count rates increase Size and shape limited only by fabrication technology 8
Silicon Drift Detector Benefits Small capacitance Small electrical contact Low noise for better spectral resolution High input and output count rates Electron Potential and Trajectories Integration of first FET Further noise improvement No pickup, no microphony 9
Types of Silicon Drift Detectors Concentric rings Allows large areas with good resolution Droplet rings For small devices hides the pickup and FET under the collimator Discreet FET Lower noise potential in FET Complex manufacturing and sensitive to microphonics Integrated FET Higher count rates due to lower capacitance 10
11 Silicon Drift Detector Devices
12 TO8 Package for SDD up to 30 mm 2
Quantum Efficiency Comparison Si(Li) SDD 0 2 4 6 8 10 12 14 16 18 20 kev High energy x-rays: Are detected by the thick Si(Li) sensor. Penetrate through the thin SDD sensor Si(Li) sensors are more sensitive for high energy x-rays (S)TEM applications 13
Spectral Resolution Prediction FWHM = (5.5 F ε E + N 2 ) ½ Where: F = Fano Factor, ~0.1 for Si ε = 3.8 ev for Si E = X-ray Energy of interest N is FWHM of electronic Noise For a given sensor material and x-ray energy, the F ε E term is a constant If N = 0 ev, then limiting resolution for a Silicon-based detector is ~100eV at Mn-Kα Current Mn-Kα resolutions are ~124 ev, placing N ~ 55 ev Reducing the electronic noise is the primary method to improving spectral resolution! 14
Electronic Noise (ENC) Analysis 2 2kT 2 1 2 ENC = α Ctot A 2πa C A qi A 1 + f tot 2 + L 3 g τ m τ Total capacitance C tot 1/f noise coeff. a f C tot : Si(Li) >> SDD (intrinsic design) I L : For Si(Li): thermal noise 1/f noise leakage current Si(Li) << SDD (but dropping) Leakage current is not a factor 1/f noise dominates at high shaping times For SDD: Small capacitance eliminates 1/f contribution Small capacitance reduces thermal noise to very small shaping times Leakage current I L Transconductance g m Filter constants A i Shaping time constant T Electron charge q α = 2/3 for FET 15
Effect of Temperature on SDD Spectral Resolution Decreasing the device temperature makes the resolution better 16
Effect of Temperature on SDD Spectral Resolution - 2 Decreased temperature improves resolution Sensor can be run at Room Temperature, but does not meet specifications. Effect is less dramatic with more cooling 17
Mn Resolution Comparison Both Si(Li) and SDD start at low values Specifications appear similar Si(Li) degrades faster with increasing input count rate Slow response of diode geometry SDD degrades slowly with increasing count rate Fast response of radial field geometry Better resolution at 7x input count rate 200 180 160 Resolution Comparison UD - 10+ NT - 10 SDD maintain their good resolution throughout the input range Mn-Ka Resolution (ev) 140 120 100 80 60 40 20 150 140 130 120 0 10 20 30 40 50 60 70 0 0 100 200 300 400 500 600 700 800 900 Input Count Rate (k cps) Specification resolution is not the primary advantage of SDD 18
New Generation Electronics Offer Spectral Stability As input count rate increases using a constant shaping time: Old self-reset mode produced varying resolution and peak locations New pulse-reset mode produces stable resolution and peak locations FWHM of Mn-Kα increases less than 4 ev Peak shift reduced to less than 8 ev 19
SDD Peak and Resolution Stability for Mn-K Peak location Does not vary with dead time % Does not vary with shaping time Manganese Test - 10mm2 Droplet Resolution Does not vary with dead time % Degrades as shaping time decreases Manganese Test - 10mm2 Droplet 6 150 1uSec 145 1uSec 5.95 2uSec 4uSec 8uSec 140 135 2uSec 4uSec 8uSec 130 ev 5.9 FWHM 125 120 5.85 115 110 105 5.8 0 20 40 60 80 Deadtime 100 0 20 40 60 80 Deadtime 20
SDD Spectral Resolution Stability Superior electronics permit stable resolution values across the spectrum as the count rate increases. Resolution Changes 140 120 Peak Resolution (ev) 100 80 60 40 20 0 C-Ka Si-Ka Mn-Ka Al-Ka 0 20 40 60 80 100 120 Input Count Rate (k cps) 21
SDD Spectral Resolution Display All peak shapes are indistinguishable from 5k to 100k cps input 22
SDD Mn Throughput A variety of electronic shaping time settings permits the best possible spectral resolution at each input count rate. Slower shaping times provide better resolution but lower output rates Faster shaping times provide higher output rates but at lower resolution Resolution Changes - UD UD 10+ Throughput 180 350 Mn-Ka FWHM (ev) 170 160 150 140 130 120 200 ns 400 ns 600 ns 800 ns 1000 ns 1600 ns 2000 ns 3200 ns 4000 ns 6400 ns 0 100 200 300 400 500 600 700 800 900 Input Count Rate (k cps) Output Rate (k cps) 300 250 50% dead time 200 ns 400 ns 200 600 ns 800 ns 150 1000 ns 1600 ns 100 2000 ns 3200 ns 50 4000 ns 6400 ns 0 0 100 200 300 400 500 600 700 800 900 Input Rate (k cps) 23
SDD Mn Throughput Display The resolution display does not change significantly until the final shaping time of 200 ns (800k cps) is used. 24
Low Energy Performance Be spectrum BN spectrum Same sensitivity as Si(Li) 25
Sensor Size: Bigger is Better A larger sensor size has the potential to collect more x-rays than a smaller sensor size. However, the real metric of x-ray detection is the solid-angle subtended by the detector. The solid-angle (SA) is defined as: SA = A / D 2 Where: A = active area of the sensor in mm 2 D = sample-to-sensor distance For the same D, a larger A detector is preferred. HOWEVER.. 26
Sensor Size: When Bigger is Not Always Better The physical geometry of the detector housing may restrict the location of the detector inside a chamber. A larger area sensor may require a larger diameter housing. If that housing conflicts with other structure inside the chamber, then its location may need to be changed. This change may require and increase in D to a clear location. This new D adversely affects the SA value. It is entirely possible that a larger area detector in a large housing mounted at a larger distance may actually collect less x-rays than a smaller area detector in a smaller housing at a shorter distance! 27
Detector Solid-Angle Comparison Solid Angle = Area / Distance 2 30 mm 2 sensor in 19 mm tube @ 43 mm: SA = 16.2 msr 80 mm 2 sensor in 35 mm tube @ 71 mm: SA = 15.9 msr 28
Spectral Resolution Requirements Good spectral resolution permits easy isolation, identification, and measurement of peaks If peak locations are greater than ~2x resolution: Peaks are isolated Peak identification is trivial Net peak counts can be measured manually Software is needed for quantification If peak locations are less than ~2x resolution: Peaks overlap in display Peak identification becomes difficult Software is needed for proper background removal and net counts measurement Software is needed for quantification 29
Spectral Resolution Limits In practice, there are potentials for many peak overlaps Most spectra require software to analyze peaks, even at highest resolution Robust routines were developed in the 1970 s and 1980 s to deal with peak overlaps Peak identification Peak deconvolution Net count measurement These routines were designed for detectors with a best resolution of ~145 ev @ Mn-K 30
Example Carbide Tool Detector: Si(Li) 10mm2 Resolution: 155eV MnK FWHM Accelerating Voltage: 20kV Magnification: 10,000x Map resolution: 256x192 31
Example Carbide Tool Raw Counts Net Counts TaM WM 155eV! 1500 1600 1700 1800 1900 2000 ev 32
Example Mo, S, Ba Multiphase Sample Detector: UltraDry 10mm2 SDD Resolution: 129eV MnK FWHM Accelerating Voltage: 7kV Magnification: 500x Map resolution: 256x192 Acquisition Time: 3 minutes 33
34 Example Mo, S, Ba Raw Count Element Maps
35 Example Mo, S, Ba Net Count Maps
Example Mo, S, Ba Phase Maps MoL SK 2100 2150 2200 2250 2300 2350 2400 2450 2500 ev Distinguishing the three main phases is not possible without robust peak deconvolution 36
Spectral Appearance Low Beam = 10k cps High Beam = 250k cps Longer shaping time produced more visible peaks Al-K resolution: 78 ev vs. 139 ev Not enough resolution to separate all peaks, even at Low Beam (best resolution) Software is required for data analysis With equal data quality, what is more important: data display or acquisition time? 37
SDD Demonstration Spectral Acquisitions Spectral Imaging (Mapping) Acquisitions 38
Summary Silicon-drift technology is relatively young Still being enhanced Silicon-drift detectors have very few limitations for EDS analyses High-energy x-rays SDD have many advantages, especially for high throughput applications Cooling Resolution degradation Maximum storage rate Detector electronics are just as important as the sensor Solid angle is a more important metric than sensor size Spectral resolution is visually appealing, but software processing is still required for spectral (and mapping) analyses 39
Conclusions Analytical results are obtained faster and at the same confidence with an SDD. SDD performance can only get better. 40