IEEE Transactions on Nuclear Science, Vol. NS-31, No. 6, December 1984
PRODUCTION OF A "STANDARD" RADIATION ENVIRONMENTTO MINIMIZE DOSIMETRY ERRORS IN FLASH X-RAY PARTS TESTING
C. M. Dozier and D. B. BrownNaval Research Laboratory
Washington, DC 20375
Abs tract
Calculations have been performed to indicatethe effects of differing device caps and of spectrum
filtration on dosimetry errors in flash x-ray parts
testing. Dosimetry typically used for upset or
latch-up testing of microelectronic devices with
flash x-ray sources can be in error by 300% or more
due to interface dose enhancement effects. When the
use of spectrum hardening is permissible, placingthe device to be irradiated within a Pb/Ta box can
reduce the dosimetry errors to less than 20% in most
cases. Obtaining the interface dose enhancementfactor by using the experimentally obtained ratio ofeffects on devices behind Kovar/Au caps to effectsmeasured using alumina caps is shown to give erro-
neous results.
Introduction
It is well known that dosimetry for the flashx-ray (FXR) sources commonly used for the testing ofparts is substantially more difficult than for totaldose testing with Co-60 sources. Flash x-ray
sources typically used for parts testing (selection)have peak electron beam voltages of from to 5 MeV.The x-ray spectra from these machines peak between200 to 500 keV. A major fraction of their dose inmaterials can be due to photons of energies below100 keV. The relative softness of these spectraleads to three significant problems for reliabledosimetry. First, the spectrum may be hardenedby intervening material before it reaches a regionof interest. Thus, the spectral distribution may
vary from place to place within a device package.Second, the beam may be attenuated by interveningmaterial. Thus, the intensity of the beam may vary
from place to place. Third, strong interface doseenhancement effects may occur. It has been demon-strated that such effects are maximized for photonsin the energy range of 20-200 keV [1].
Three general methods have been proposed for
handling the above listed effects. First, the FXRspectrum can be calculated, or measured, and each of
the above effects can be calculated, or measured[2]. This procedure is difficult, but sometimesnecessary. Second, representative test devices can
be exposed on a linac, with the bulk of devices
tested on a FXR source [3]. This procedure allows
the FXR dosimetry to be normalized to the more
manageable linac dosimetry. Third, it has been
suggested that the spectra from FXR sources be
hardened in order to minimize the undesirable
effects of the low energy photons [41.
The third of these methods is the topic of this
paper. It will be demonstrated that the use of
heavy metal filters with FXR sources allows the use
of relatively simple dosimetry. It is here en-
visioned that the device under test and the key
*
These results have been supported by the Defense
Nuclear Agency through its Hardness Assurance
Program.
dosimetry (e.g. TLD's) are both totally enclosed ina Pb/Ta walled test box. This box will allow for asimpler conversion from the dosimeter dose to thatin Si and also eliminate much of the dose enhance-ment that can occur. It will be shown also thatthis method does not work as well for FXR sourceswith peak voltages below 2 MeV because of excessiveloss of dose delivered to the device. In contrastto a related earlier publication [4], this paperconcentrates on the dose delivered to the Sisubstrate where the resulting photocurrents maycause device upset or latchup. Finally, the effectsof different device caps on the deposited dose willbe discussed.
Background, Assumptions, and Model
The earlier related paper, reference 4, hasconsidered dosimetry errors in the irradiation ofMOS devices using FXR and Co-60 sources. It was
concluded that the worst case dosimetry errors are
caused by interface dose enhancement. A model fora reasonable worst case device geometry was intro-duced. This model was based on the assumption thatdose to the SiO2 gate oxide was the critical dose.
Calculations were presented showing the magnitude ofdosimetry errors which can occur (in a reasonableworst case). Examples were presented showing thereduction of these errors which can be obtained byhardening of the spectrum using a Pb filter.
FXR tests are commonly used to test for suchtransient radiation failures such as device upset or
latchup. For these failure modes the dose to thegate oxide (as was the case in reference 4) is not
the critical dose. The driving mechanism isphotocurrents produced primarily within a criticalregion of the Si. Thus the dose which contributesto the primary photocurrent pulse is that to the Sisubstrate within the depletion layer distance of the
SiO2/Si interface, for MOS devices, or to the
Si within a depletion depth Qf the PN junction, for
bipolar devices. It follows that the dose in the
top 10 to 30 uim of the Si is the critical dose.
As was done in reference 4, a reasonable worst
case will be assumed for device geometry. Energydeposition within the critical region of this worst
case device will be calculated. The reasonable
worst case assumed for this paper is that of a
device mounted behind a Kovar cap flashed with 1 vim
of gold. The photoelectrons produced during photo-electric absorption of a FXR spectrum in the Kovar
and Au can travel tens of microns, resulting in a
significant enhancement of the dose deposited in
the Si of the device. It will be assumed that the
FXR radiation is collimated, and that its direction
of travel causes it to pass through the Kovar/Au capfirst and then enter into the Si. Irradiation in
the opposite direction (radiation entering the Si
first) is of course possible but would be bad
practice because it would result in a significantincrease of interface enhancement due to the scat-
tering of Compton electrons. For further detail on
this subject see, for example, the work of Wall
and Burke [5]. It is emphasized that here it is the
testing of individual parts which is under con-sideration, not boards or larger assemblies ofparts. For such larger assemblies it may not alwaysbe possible to face the device caps toward theradiation source.
The effect on devices of six FXR spectra willbe reported. These six spectra include three peakenergies (1,2 and 3 MV). They also contain twotypes of spectral distribution, relatively softand relatively hard spectra. These are calculatedspectra designed to span the full range of spectraldistributions of FXR machines currently in use forparts testing. The three soft spectra are basedon the output characteristics of Gamble II [6]. Thethree hard spectra are based on the output charac-teristics of the Febetron 705 source [7,8]. Notethat there is no intent to imply that Gamble IIshould be used for parts testing or that the Febe-tron 705 can be run at 3 MeV. These are simply usedas the prototypical soft and hard machines. Thesesix representative spectra will be used to testthe effects of the FXR's peak electron voltage onthe deposition behavior and to demonstrate theability to provide a similar or "standard" environ-ment for parts testing with different sources.Further details on the determination of thesespectra can be found in an appendix to this paper.
Calculation Method
The energy deposition calculations used in thispaper are (a) the calculation of photon absorptionin the device and in intervening materials and then,(b) the calculation of electron transport in thedevice. These methods have been discussed inprevious papers [4,9] and thus will be reviewed onlybriefly here. The materials being irradiated aremodeled as a sandwich, i.e., a stack of planarslabs. Photon energy loss in intervening materials(filtering) and energy deposition within regionsof interest in the device are calculated usingenergy deposition coefficients [10]. Interface doseenhancement effects are treated in two parts. Thetransport of the generated electrons with energiesless than about 500 keV (primarily photoelectronsand Auger electrons) are treated using a numericalsolution of the Boltzmann equation [9]. Thiselectron transport program has been used extensivelyand has been checked against experimental data insuch fields as x-ray microbeam analysis [11], theoutput of x-ray tubes [12,13], and the output offlash x-ray machines [6,14]. The low energy con-tribution to interface dose enhancement has beentested in MOS capacitors [15]. The interface doseenhancement effects of higher energy electrons(predominantly Compton electrons) are treatedapproximately using experimental and calculated datafrom the literature [5,15]. Because the majoreffects are caused by the lower energy electrons thedegree of approximation thus introduced has anegligible effect on the conclusions to be drawn.
Calculation Results
Dosimetry Difficulties with FXR Sources -- Repre-sentative Calculations
The calculations reported here were performedfor two experimental geometries. The first (theworst case geometry) is a device behind a gold-flashed Kovar cap. The Kovar thickness is 10 milsand the gold flashing is 1 pm thick. The secondgeometry (designed to have low interf ace enhance-ment) is a device behind an alumina cap with athickness of 10 mils. In both cases the device is
1085
simply represented by a layer of Si because (a) itis the dose deposited in the Si substrate which wewish to study, and (b) the intervening layers ofgate oxides, etc., have a negligible effect onphoton and electron transport for this problem. Thecalculations also include the effects of selectedfilters placed between the FXR source and the twoexperimental geometries. These "filters" includeseveral materials which may serve as parts of theFXR source front end or which may serve as a spec-trum-hardening filter box. It should be noted thatthe device located behind the two types of caps canbe regarded conceptually as any Si-based device. Itcould be thought of as an MOS or bipolar device tobe tested, or it could just as well be thought of asa 30 vim PIN detector used for dosimetry purposes.
An example of the results of the energy deposi-tion calculations for a device having a Kovar/Au capare shown in Fig. 1.
50
co
I 101
CO 50
13
-AuKovar Si
/A I,% -A -I
Graphite/Fiberglass
filter
50 0 50DISTANCE -- microns
100
Fig. 1. Energy deposition in a Kovar/Au cappeddevice. Doses from photoelectron depositionnear the Kovar/Au cap interface for 2 MVspectrum with and without 1/16" Pb filtra-tion are shown.
This calculation was performed using a relativelysoft calculated spectrum of the "Gamble" type (seeAppendix) with a 2 MV peak voltage. The results offiltration of the incident beam with 1/16" of Pb canbe seen clearly. For this example, the region ofinterest for charge collection was considered to bethe first 30 Vim of Si. Note that the overall doseand the interface dose enhancement effect are bothreduced significantly by the Pb. It should bereiterated that calculations of the type presentedin Fig. 1 treat well the relatively large inter-face dose enhancement effects due to photelectricabsorption, but do not treat the interface doseenhancement effects characteristic of high energyphotons (due largely to the scattering of electronsgenerated by the Compton scattering of the pho-tons). The work of references 4 and 10 suggest thatfor Kovar/ Au caps there should be a slight reduc-tion in dose in Si due to this high energy doseenhancement effect. Accordingly, in the interfacedose enhancement results which follow, we havereduced that portion of the dose in Si which is dueto Compton processes by 10% of the equilibrium dose(10% of the dose far from the interface). Thiscorrection, though clearly approximate, is estimatedto introduce here no more than a few percent pos-
-
I I
1086
sible error. This adjustment was only used for thecase of Kovar/Au caps. There are expected to be nosignificant high energy interface dose enhancementeffects in the case of alumina caps.
Additional results obtained using the samespectrum are presented in Tables I through III andin Fig. 2. A summary of interface enhancementfactors calculated for both cap types and for threeforms of filtration is given in Table I. Note inthe context of Fig. 1 that the interface enhancementfactor is defined as the average dose in the first30 npm (or 10 p m) of Si divided by the equilibriumdose (the dashed line).
Table I
Calculated Dose Enhancement in the First 30 Micronsof Devices Behind Front-End Filters
Cover on Carbon & 1/4" 1/16"I Package Fiberglass(*) Al Pb
I Kovar/Au 1.96 1.96 1.12 1I Alumina 0.98 0.98 0.98
CaF2 TLD chip in 30 mil Al equilibrium housing.Dose using a Kovar/Au cap divided by TLD dose.Dose using an alumina cap divided by TLD dose.Dose using a Kovar/Au cap divided by the doseusing an alumina cap.
Table III
Loss in Dose With Increasing Filtration*
Measuring Carbon & 1/4" 1/16"Device l Fiberglass Al Pb
TLD 1.0 0.79 0.21Kovar/Au Cap 1.0 0.91 0.23 1Alumina Cap 1.0 0.78 0.28
(*) The value tabulated is the dose in the measuringdevice with a filter divided by the dose withouta filter.
Table II contains a comparison of doses indevices behind Kovar/Au and alumina caps and in CaF2
dosimeters. The values in the Table reflect theeffects of filtration by the caps, interface doseenhancement, and the different spectral response ofSi and CaF . The relatively good match between
2'the Kovar/Au capped device and the CaF TLD's is
2in part fortuitous due to the partial cancellationof the interface dose enhancement and the loss dueto filtration by the Kovar. The alumina covereddevice does not match as well. It is only with theaddition of 1/16" of Pb that the results in allcases are within acceptable agreement.
It has been assumed in the literature that the
ratio of radiation effects on a device behind a
Kovar/Au cap to the corresponding effects using an
alumina cap is a good approximation to the interface
dose enhancement ratio [16]. The validity of such
an assumption for FXR spectra is investigated in
Table II and Fig. 2. In Fig. 2 the ratio of dose
using a Kovar/Au cap to dose using an alumina cap is
referred to as "apparent dose enhancement". Note,
in Fig. 2, the differences between the interface
enhancement ratio and the ratio of doses behind the
two cap types. The dose calculated behind the two
types of caps is shown for three types of filtra-tion. It is clearly not true in general that the
ratio of doses behind the two cap types can be
equated with the interface dose enhancement ratio.
In Table III the dose behind a Kovar/Au cappeddevice is compared to the dose in a CaF2 TLD. A
similar comparison is made for an alumina cappeddevice and CaF2 TLD's. The loss of deliverable
dose due to the use of filtration can be substan-tial. The effect of three types of filtration on thedose in three materials is summarized in Table III.
c - Kovar/Au Capped Devices0
O_2 bo Actual Dose< UZ ~~~~~~~Enhancement
z AnOarent Dose
ment with "apparent" dosEnhancement
z
z
w
w
(01
Graphite + 1/4 Al 1/1e6' Pb
Fiberglass
INCREASING FILTRATION
Fig. 2. Comparison of actual interface dose enhance-
ment with "apparent" dose enhancement.
Apparent enhancement is the ratio of dose
with a Kovar/Au cap to dose with an alumina
cap. Note the differences and the anomalous
"rise" in the apparent dose enhancement with
increasing filtration.
Production of a "Standard" FXR Environment
It has been found that hard and soft FXR
spectra (see Appendix for details) can be made verysimilar (a "standard" radiation environment) byusing a compound filter consisting of 1/16" of Pbfollowed by 10 mils of Ta. The effect of a Pb plusTa filter on two representative FXR spectra is shownin Fig. 3. The Ta filter is required to eliminatethe undesirable photon intensity between the Ta andPb K-absorption edges (67 and 88 keV, respectively)which would not be adequately filtered out bythe Pb alone. (In practice one might in additionwish to use a thin (15-30 mil) layer of Al to
eliminate Ta photoelectrons and to achieve approx-imate electron equilibrium with the usual lowatomic number devices.)
Table V. Hard ("Febetron") Type Spectrum 1087
SiCritical Interface Interface Dose
Peak Layer Enhance- Enhance- Factor*:Voltage Thickness ment ment TLD
(,pm) With With (inside)No Filter Pa/Ta TLD
Filter (outside)
v 10 3.0 1 1.2 33%lV I 30 1 2.0 1 1.2 1 3
2MV 10 1 1.8 1.1 60%1 30 1 1.5 1.0
3MV I lO 1.4 1.0 74%I 1 30 1 1.2 1.0 1
*The fraction of
filtration.
the original dose remaining after
PHOTON ENERGY(keV)
Fig. 3. Comparison of a "hard" FXR spectrum and a"soft" FXR spectrum before (dashed lines)and after (solid lines) filtration with1/16" of Pb and 10 mils of Ta. Note thesimilarity of the filtered spectra.
The results of using a Pb/Ta f ilter are sum-
marized in Tables IV and V. The results for a"sof t" spectrum are given in Table IV; the resultsfor a "hard" spectrum are given in Table V. Thecalculated interface dose enhancement is shown, withand without Pb/Ta filtration. Results are given for3 FXR source peak voltages and for two differentassumptions for the thickness of the Si criticallayer. Note that the interface enhancement issubstantially reduced by Pb/Ta filtration. This isparticularly true for the "soft" spectra. Theimprovement is less dramatic for the "hard" spectrabecause in this case the Ta converter is thickenough to introduce a significant amount of selffilt ration.
Tables IV and V also present the ratio of thedoses to TLDs placed inside and outside of the Pb/Tafilter. This will provide an indication of thedegree of reduction of available dose to a testdevice which is caused by the use of the filter.
Table IV. Soft ("Gamble") Type Spectrum
SiCritical Interface Interface Dose
Peak Layer Enhance- Enhance- Factor*:Voltage Thickness ment ment TLD
(1am) With With (inside)No Filter Pb/Ta TLD
Filter (outside)
I umv I 10 4.3 1.3 8%1 30 1 2.4 1.2 1
2MV 10 3.4 1 1.1 21%1 30 2.1 1 1.1 1
I 3MV I 10 2.9 1 1.0 '32% 1V 1 30 1 1.9 1 1.0
*The fraction of the original dose remaining afterfiltration.
Many engineers and scientists who perform partstests with "sof t" FXR sources are aware of thepotential problems caused by the low energy photonsand introduce additional filtration to their FXRfront ends. However, the amount of filtrationvaries from facility to facility and is probably notas effective as that suggested here. Because ofthis existing filtration the large dose reductionfactors shown in Table IV are somewhat larger thansome facilities would experience. For this reason,the factors shown in Table V for "hard" sources areprobably more representative of the losses thatwould be realized with the addition of a Pb/Ta cell.The losses of less than a factor of 2 shown in TableV for sources of 2 MV or greater voltages may beacceptable in many cases. Below 2 MV the dosereduction factor falls off rapidly, and it willbecome increasingly difficult for some FXR sourcesto generate the desired effects in the devices undertest.
Conclusions
Four conclusions can be drawn from the resultspresented. First, conducting FXR exposures with thedevice under test and the TLD dosimeters within a
test cell of 1/16" thick Pb followed by 10 mils ofTa can produce a "standard" radiation environmentwhich allows relatively simple dosimetry withmoderate accuracy. The errors due to neglecting a
correction for interface dose enhancement are lessthan about 30%. This error may be acceptable formany cases. For comparison, in the absence of Pb/Tafiltration the maximum errors due to neglectinginterface dose enhancement are about a factor of 3.The use of a Pb/Ta test cell to produce a "standard"radiation environment should improve the intercom-parablity of tests on different FXR sources signifi-cantly. It also will simplify the problem of extra-polating the measured results to those which wouldbe expected from a different radiation environment.It should be borne in mind, however, that in some
cases it is precisely the low energy components ofthe spectrum which are of primary interest. Insuch cases the use of a Pb/Ta test cell may beunacceptable.
The second conclusion is that the use of fil-tration suggests the use of relatively high energy
FXR sources. The available energy in photons above100 keV is significantly reduced as the peak elec-tron voltage approaches 1 MV. At 2 MV the typicalloss in deliverable dose caused by the Pb/Ta box isabout a factor of 2. This loss may be acceptable.
011
i-
0
-cn
z
zwH
-jwa:
1088The calculated losses are greater for the "soft"source, but some filtration is currently used forparts testing with these sources in many circum-stances. The losses for sources operated below 2 MVwill be significantly greater. This trend indicatesthat the use of filtration to obtain a standardenvironment is less usable for relatively low energymachines. There will be cases where it is notpermissible to use filtration because the loss ofavailable dose is unacceptable. In such cases itwill be necessary to treat the effect of interfacedose enhancement on dosimetry in a more elaboratemanner than the method suggested in this paper.
The third conclusion refers to the ratio oftest results using a Kovar/Au cap to results usingan alumina cap. This ratio has been used as anexperimental measure of interface dose enhancement[171. Although this ratio can be a valid measure ofinterface enhancement for Co-60 irradiation [171,Fig. 2 shows that it can be misleading when used forFXR irradiation. Note that there is an apparent (anderroneous) indication of increasing dose enhancementwith increasing filtration. These calculationsoffer an explanation of recent anomalous experi-mental results [18].
The fourth conclusion is that the conversionfrom dose in TLD to dose in Si (ignoring interfaceenhancement effects) is simple only if Pb filtrationis used. It can be seen from the second line ofTable II that in the absence of Pb filtration acorrection of about 30% must be made to get fromdose(TLD) to dose(Si). This correction can be doneby calculation only if the spectrum is known.
Appendix
FXR Spectra
Spectra for these calculations were chosen tobracket the characteristics of sources which mightbe used in parts testing. The operating conditionsfor two typical sources were chosen as representa-tive [6,8]. In order to provide spectra for 1, 2,and 3 MV (peak voltage), the voltage trace, the Tatarget thicknesses and the window thicknesses werescaled. The scaled parameters are shown in theTable A-1 below. The first representative sourcewas based on voltage, current, and target parametersof a Febetron 705 operated at 2.2 MV [7,81. Incomparison to other FXR sources, Febetron 705's, astypically configured, produce relatively hard x-rayspectra. The x-ray converter acts as both aconverter and electron beam stop. Much of the lowenergy portion of the x-ray spectrum is filteredout by the extra thickness required to stop theelectrons. Converter thicknesses are approximately0.6 of the range of an electron at the peak outputvoltage. The second representative source wasmodeled on results of measurements on Gamble IIat 1.4 MV using a reflexing electron beam [6]. Thissource uses a converter thickness which is approx-imately 0.1 of the maximum electron range. Reflex-ing of the beam back and forth through the converterincreases the total x-ray output and enhancesthe low energy portion of the spectrum. A thin beamstop of graphite and a thin debris shield of fiber-glass and melamine allows much of the low energyx-rays to escape. Although Gamble II or similarsources are not used for parts testing, theyrepresent a lower bound on the "softness" of spectraobtained from FXR's currently used in this type oftesting.
TABLE A-1. Source Characteristics
I I I I I IISource IVoltageITargeti Target I Window II I IThicknessl 2)
I I (&/cm2) g/)I I I
I I 1 MV ITa 0.50 1 None II"Febe- I I III tron 2 MV I" 1.05 "705" 1 1 1
Itype 13 MV I t 1.50 " II- -- ----
I1 MV I Ta I 0.10 10.50 graphite II"Gamblel I 1 10.30 melamine/glasslI II" 2 MV I " 0.18 11.10 graphite II type I I 1 10.60 melamine/glasslI 1 3 MV I " 0.25 11.70 graphite II I I 1 10.90 melamine/glassl
[10] See for example, "Mass Attenuation and Absorp-tion Coefficients ...", W.J. Viegle, in "Hand-book of Spectroscopy, Vol. I", J.W. Robinson,ed. (CRC Press, Cleveland, 1974), p.28ff.
[17] John G. Kelly, Theodore F. Luera, Lawrence D.Posey, David W. Vehar, Dennis B. Brown andCharles M. Dozier, IEEE Trans. Nuc. Sci.,NS-30, 4388 (1983).
[18] L.D. Posey, D.C. Evans, and D.W. Vehar, DNAFlash X-Ray Dosimetry Workshop, Los Angeles,January 26, 1984.
