scieee Science in your language
[en] (orig)

Total Ionizing Dose Effects on a Delay-Based Physical Unclonable Function Implemented in FPGAs

Abstract

Physical Unclonable Functions (PUFs) are hardware security primitives that are increasingly being used for authentication and key generation in ICs and FPGAs. For space systems, they are a promising approach to meet the needs for secure communications at low cost. To this purpose, it is essential to determine if they are reliable in the space radiation environment. In this work we evaluate the Total Ionizing Dose effects on a delay-based PUF implemented in SRAM-FPGA, namely a Ring Oscillator PUF. Several major quality metrics have been used to analyze the evolution of the PUF response with the total ionizing dose. Experimental results demonstrate that total ionizing dose has a perceptible effect on the quality of the PUF response, but it could still be used for space applications by making some appropriate corrections.

Read accessible full text

Total Ionizing Dose Effects on a Delay-Based Physical Unclonable Function Implemented in FPGAs

Author: Martín, Honorio; Martín Holgado, Pedro; Morilla García, Yolanda; Entrena, Luis; San Millán, Enrique
Publisher: MDPI
Year: 2018
DOI: 10.3390/electronics7090163
Source: https://idus.us.es/bitstreams/00b6356b-92a9-4297-8680-b13dd3b9334b/download
elec onics
A icle
To al Ionizing Dose E ec s on a Delay-Based Physical
Unclonable Func ion Implemen ed in FPGAs
Hono io Ma in 1,* , Ped o Ma in-Holgado 2, Yolanda Mo illa 2, Luis En ena 1
and En ique San-Millan 1
1Depa emen o Elec onics Technology, Uni e sidad Ca los III de Mad id, 28911 Leganés, Spain;
[email p o ec ed] (L.E.); [email p o ec ed] (E.S.-M.)
2Cen o Nacional de Acele ado es (CNA, Uni e sidad de Se illa, CSIC), 41092 Se illa, Spain;
[email p o ec ed] (P.M.-H.); [email p o ec ed] (Y.M.)
*Co espondence: [email p o ec ed]; Tel.: +34-91-624-8390
Recei ed: 6 Augus 2018; Accep ed: 23 Augus 2018; Published: 24 Augus 2018


Abs ac :
Physical Unclonable Func ions (PUFs) a e ha dwa e secu i y p imi i es ha a e inc easingly
being used o au hen ica ion and key gene a ion in ICs and FPGAs. Fo space sys ems, hey a e a
p omising app oach o mee he needs o secu e communica ions a low cos . To his pu pose, i is
essen ial o de e mine i hey a e eliable in he space adia ion en i onmen . In his wo k we e alua e
he To al Ionizing Dose e ec s on a delay-based PUF implemen ed in SRAM-FPGA, namely a Ring
Oscilla o PUF. Se e al majo quali y me ics ha e been used o analyze he e olu ion o he PUF
esponse wi h he o al ionizing dose. Expe imen al esul s demons a e ha o al ionizing dose has a
pe cep ible e ec on he quali y o he PUF esponse, bu i could s ill be used o space applica ions
by making some app op ia e co ec ions.
Keywo ds:
physical unclonable unc ion; FPGA; o al ionizing dose; Co-60 gamma adia ion;
ing-oscilla o
1. In oduc ion
Secu ing sensi i e in o ma ion on low-cos sa elli e applica ions has become a majo challenge
o he space indus y. Typical app oaches ha include e y expensi e c yp og aphic p imi i es,
non- ola ile memo y and analogous blocks canno be a o ded in hese small space sys ems. In his
con ex , comme cial Field P og ammable Ga e A ays (FPGAs) ha e u ned ou o be a good solu ion
due o hei lexibili y and cos . Among hei many uses, FPGAs can be dedica ed o ensu ing secu e
sa elli e da a.
A popula solu ion o p o ide secu i y in esou ce cons ained applica ions, such as hose using
FPGAs, is on-chip Physical Unclonable Func ions (PUFs). PUFs a e a e y p omising secu i y p imi i e
used o au hen ica ion and key gene a ion in IC and FPGAs. These secu i y p imi i es a e based
on he impossibili y o c ea ing wo physically exac ly iden ical ICs due o he in luence o andom
and uncon ollable e ec s du ing he manu ac u ing p ocess. These uncon ollable in luences lea e
measu able andom ma ks on some ea u es which possess he po en ial o gene a e enc yp ion keys
di ec ly associa ed o a de ice [
1
]. Thus, PUFs wo k as an unclonable speci ic ea u e ha can iden i y
a ci cui , jus as a inge p in can iden i y a human being. Among he a ious de ice p ope ies ha
can be used o his pu pose, delay-based PUFs dese e special a en ion due o hei s aigh o wa d
implemen a ion. A delay-based PUF exploi s he delay dependency on he andom p ocess a ia ions [
2
].
Well-known examples o his ype o PUF a e he A bi e PUF [3] and he Ring Oscilla o PUF [2].
PUFs can be used o sol e an impo an issue ela ed o he gene a ion o secu e enc yp ion keys
in sa elli e communica ions, emo ing he necessi y o key s o age. None heless, as PUF esponse
Elec onics 2018,7, 163; doi:10.3390/elec onics7090163 www.mdpi.com/jou nal/elec onics
Elec onics 2018,7, 163 2 o 11
depends on some ci cui ea u es ha may be a ec ed by he ope a ional condi ions; i is impo an o
asse he sui abili y o hese p imi i es in ha sh en i onmen s subjec ed o ionizing adia ion. Ionizing
adia ion induces cha ges in he semiconduc o ma e ial ha can be apped in he oxide, al e ing
he elec ical cha ac e is ics o elec onic de ices. This e ec , known as To al Ionizing Dose (TID),
is cumula i e and p oduces a g adual deg ada ion o majo elec ical pa ame e s, such as h eshold
ol age and leakage cu en , ha can e en ually esul in de ice ailu e a a ce ain dose. To he bes
o ou knowledge, he e ec s o TID in delay-based PUFs ha e no been s udied be o e. The e is one
wo k ha s udies he TID e ec s in a CMOS silicon PUF based on ansis o b eakdown [4].
In his wo k, we e alua e he e ec s o ionizing adia ion on a well-known delay-based PUF [
2
]
implemen ed in a SRAM-FPGA. To ha end, we ha e pe o med an ex ensi e es wi h wo di e en
de ices exposed o a adia ion sou ce, pe iodically collec ing he PUF esponse as TID inc eased.
All he ex e nal in luences ha can a ec he PUF esponse ( empe a u e, humidi y, ol age, e c.) we e
con olled. Se e al majo quali y me ics ha e been used in o de o assess he impac o adia ion in
he PUF esponse.
The emainde o his pape is o ganized as ollows. Sec ion 2in oduces he RO-PUF unde s udy
and he ypical e ec s o ionizing adia ion on SRAM-FPGAs. Sec ion 3desc ibes he implemen a ion
o he RO-PUF and he TID es se up. Sec ion 4 epo s he impac o he ionizing dose on he
RO-PUF. Se e al me ics a e p esen ed and analyzed in his sec ion. Finally, Sec ion 5summa izes he
conclusions o his wo k.
2. Backg ound
2.1. Ring Oscilla o Based PUF
A Ring Oscilla o (RO) is a delay loop ha oscilla es a a pa icula equency. Thanks o i s
s aigh o wa d implemen a ion in FPGAs, ROs ha e been widely used in he implemen a ion o secu e
p imi i es such as ue andom numbe gene a o (TRNG). RO-PUFs a e delay-based PUFs ha use
Challenge-Response scheme as a chip au hen ica ion mechanism. A adi ional RO-PUF [
2
] makes use
o many iden ically laid ou ROs o quan i y he manu ac u ing a iabili y. RO oscilla ion equencies
depend on (i) ixed condi ions es ablished a he design phase (i.e., numbe o s ages, place& ou e, e c.);
(ii) andom p ocess a ia ions ( ha once manu ac u ed a e ixed o each single de ice); (iii) dynamic
condi ions de i ed om he ope a ion en i onmen (i.e., supply ol age, empe a u e, su ounding
logic, e c.). Figu e 1depic s he adi ional RO-PUF scheme [
2
] ha consis s o many iden ical ROs,
coun e s and compa a o s.
MUX MUX
Inpu s
C
Coun e
Coun e
Ou pu s
Figu e 1. Ring Oscilla o Physical Unclonable Func ion (RO-PUF) gene al scheme.
Du ing he au hen ica ion p ocess, a pai o ROs (selec ed by a use challenge) a e quan ized by
measu ing and compa ing he RO equencies ( a, b)and gene a ing a esponse bi :
Elec onics 2018,7, 163 3 o 11
=(1 i a> b
0o he wise (1)
An n-bi signa u e o he chip is compu ed om n di e en compa isons be ween RO equencies.
The quali y o he PUF is e alua ed by analysing he signa u es. In mos o he li e a u e, majo secu i y
me ics like uni o mi y, eliabili y and uniqueness a e used o assess he PUF quali y [5].
•Uni o mi y
is a me ic ela ed o he en opy o he sys em. This me ic es ima es he a io o
‘1’ s. ‘0’
in all he esponse bi s gene a ed by a PUF. The uni o mi y is compu ed as ollows o
an n-bi PUF esponse:
Uni o mi y =1
n
n
∑
l=1
i,l×100% (2)
whe e
i,l
is he l- h bi o an n-bi PUF esponse. An uni o m dis ibu ion o 0’s and 1’s (50%) is
expec ed in PUFs ha ha e ull en opy.
•Reliabili y
is a me ic ha quan i ies how s able he PUF esponse is o e a ying ope a ing
condi ions ( ol age, empe a u e, aging, e c.). To ha end, a speci ic challenge is applied o
he RO-PUF in o de o ob ain an n-bi e e ence esponse
( i)
o no mal condi ions ( oom
empe a u e, ideal powe supply ol age). The same n-bi esponse is collec ed a di e en
ope a ing condi ions
( 0
i)
. Finally, he eliabili y is ob ained using he Hamming dis ance (HD)
analysis o esponses.
Reliabili y =1
x
x
∑
y=1
HD( i, i,y0)
n×100% (3)
whe e xis he numbe o samples o each condi ion and
i,y0
is he y- h sample o
( 0
i)
. A lowe
alue o in a-die HD leads o a highe eliabili y.
•Uniqueness
is a measu e o how di e en he PUF esponses o di e en chips among a se o
chips a e. The uniqueness o a popula ion o k-chips is ob ained by compu ing he in e -HD o
he n-bi esponses o he same challenge:
Uniqueness =2
k(k−1)
k−1
∑
i=1
k
∑
j=i+1
HD( i, j)
n×100% (4)
whe e
i
and
j
a e he PUF esponses o chips iand j(
i6=j
). An ideal uniqueness o 50% is
desi ed o he comple e se o chips.
2.2. TID on SRAM FPGAs
TID e ec s on SRAM-based FPGAs, including non- adia ion ha dened, ha e been widely s udied
in ecen yea s [
6
–
8
]. TID causes a deg ada ion o he ansis o s as ionizing adia ion accumula es on
he componen . This deg ada ion leads o c ea ing apped cha ges ha will slowly a ec he elec ical
pa ame e s o he de ice ( h eshold ol age
(V h)
and leakage cu en ) [
9
]. In his con ex , NMOS and
PMOS ansis o s beha e di e en ly. The apped cha ges will nega i ely a ec he h eshold ol age
inc easing he leakage cu en in NMOS ansis o s. Con e sely, in PMOS ansis o s he h eshold
ol age will be inc eased and he leakage educed. In addi ion, a de e io a ion o noise pa ame e s
can be obse ed [
10
]. All hese e ec s a e dependen on many ac o s such as dose a e, he ype o
adia ion applied, empe a u e, e c.
Elec onics 2018,7, 163 4 o 11
A de ice le el, an inc ease o he p opaga ion delay o he ci cui s ins an ia ed in he FPGA is
he main aspec o conside . Faul s can appea when he iming cons ain s a e iola ed due o his
inc emen [11].
3. Expe imen al Se up
3.1. De ice Unde Tes
The adia ion expe imen s ha e been ca ied ou on wo Xilinx XC3S500E FPGA, manu ac u ed on
90 nm CMOS echnology. In he emainde o he pape , hese De ices Unde Tes (DUTs) a e e e ed
o as FPGA1 and FPGA2. The clock is se using he on-boa d 50MHz c ys al oscilla o . Two high
p ecision ol age sou ces (p og ammable HP 66103A DC powe modules) ha e been used in o de o
se and moni o he co e ol age (1.2 V) and he I/O ol age (3.3 V).
A con en ional RO-PUF consis ing o 512 iden ically laid-ou ROs has been implemen ed. Each
o he ROs consis s o ou in e e s and a NAND ga e, he la e being able o enable/disable he
oscilla ion. A ha d mac o ha occupies one Con igu able Logic Block (CLB) has been c ea ed in o de
o gua an ee he same placemen and ou ing o all he ROs. This ha d mac o has been eplica ed in
he middle o he FPGA c ea ing a 16
×
32 a ay. Du ing he expe imen , each RO is ac i a ed a a
ime du ing 13,000 clock cycles using i s enable signal. The es o he RO-PUF logic (mul iplexe s,
coun e s, decode s, e c.) ha e been implemen ed in o he FPGA zones in o de o limi he impac o
he su ounding logic on RO equencies. An RS232 communica ion p o ocol has been used in o de o
ans e he measu ed RO equencies o he hos compu e . A CRC has been implemen ed in o de o
ensu e he in eg i y o he communica ion.
The ope a ing condi ions o he oom ( empe a u e, p essu e and humidi y) ha e been con olled
in o de o gua an ee ha hese condi ions do no a ec he RO equencies.
3.2. TID Se up
The TID es s ha e been pe o med a he RADLAB acili y, he Gamma Radia ion Labo a o y
ins alled in he Cen o Nacional de Acele ado es (CNA), Spain. The RADLAB [
12
,
13
] is based on
a Co-60 adioac i e sou ce, placed in o a Gamma beam X200 i adia o . The a e age alue o he
pho ons ene gy is 1.25 MeV, which is usually es ablished o es ing pu poses.
The i adia o has a conical opening which con ains a a iable collima o , p o iding di e en
squa e i adia ion ields. Du ing a i s i adia ion se up, he maximum i adia ion ield was used and
no shielding was applied on he boa d, so all he componen s o he PCB we e exposed o adia ion,
no only he DUT (FPGA). As a consequence, some issues we e obse ed be o e de ec ing any e ec in
he DUT. Fo he subsequen campaigns, he i adia ion ield was educed o ocus he main gamma
beam on he FPGA unde s udy. Mo eo e , he se up was addi ionally imp o ed wi h a cus om pa ial
shielding on he boa d, signi ican ly dec easing he dose a e on he mos sensi i e componen s o he
PCB (Figu es 2and 3).
Since he DUT is loca ed in one speci ic posi ion o he PCB submi ed o adia ion, a dummy
boa d was placed o each se up p epa a ion in he same posi ion as he SAMPLE in o de o ca y ou
he dosime y (Figu e 4), ha is, o measu e he dose a e on he DUT, be o e s a ing he i adia ion es .
The dosime y sys em is composed by a Fa me ioniza ion chambe connec ed o he UNIDOS
Webline elec ome e , bo h o hem by PTW. Fi s ai ke ma a e is ob ained and he dose a e in silicon
(Si) is calcula ed aking in o accoun he con e sion ac o s. The dose a e uni o mi y in he adia ion
ield was be e han 95% and he expanded unce ain y associa ed wi h he measu emen was
±
4.2%.
Elec onics 2018,7, 163 5 o 11
Figu e 2. Shielding o he PCB a ached o he co e o he il e box.
Figu e 3. P o ile iew o he shielding be ween he PCB and he co e o he il e box.
Figu e 4.
PCB cus omized o ca y ou he dosime y measu emen s wi h he e e ence poin o he
ioniza ion chambe placed exac ly in he posi ion o he de ice unde es (DUT).
The dosime y and he i adia ion un we e pe o med using a il e box, wi h he DUT inside,
acco ding o he TID s anda d om Eu opean Space Agency [
12
]. This con aine has 1.5 mm Pb (lead)
wi h an inne lining o 2 mm Al (aluminium). The on co e is made o Al, excep in he egion close
o he DUT, whe e he build-up ma e ial is polyme hilme ac ila e (PMMA). In Figu e 5is depic ed he
inal se up.
The FPGA1 and FPGA2 we e exposed o a o al dose o 500 k ad(Si), wi h he dose a es o
5.2 k ad(Si)/h and 5.3 k ad(Si)/h, espec i ely.

Elec onics 2018,7, 163 6 o 11
Figu e 5.
Fil e Box wi h he PCB (and DUT) inside and he connec ions o he di e en po s and
powe supply.
4. Expe imen al Resul s and Discussion
This sec ion desc ibes he esul s o he i adia ion expe imen s on he wo DUTs.
4.1. FPGA Pa ame e s and Func ionali y
The co e cu en s and ol ages we e con inuously measu ed du ing he en i e expe imen .
The p e-i adia ion ope a ion cu en s we e measu ed o be 23.9 mA o FPGA1 and 29.09 mA o
FPGA2. A he end o he expe imen [500 k ad(Si)], he cu en s eached 95.97 mA and 83.35 mA,
espec i ely. These in e nal co e cu en s a e wi hin he limi s o he endo ecommenda ions o his
FPGA ( ypical: 25 mA; maximum: 106 mA). Ne e heless, he i s ailu e was egis e ed a 410 k ad(Si)
in FPGA1. This ailu e was ela ed o he RS232 communica ion p o ocol and a ep og amming o he
FPGA was necessa y in o de o eco e no mal unc ionali y. This kind o e o was ep oduced un il
he end o he expe imen . In FPGA2, no ailu es we e egis e ed. Bo h DUTs could no be econ igu ed
any mo e a e he deposi ed dose eached 440 k ad(Si). The aul y beha iou ha appea ed only
in FPGA1 can be explained by he ba ch di e ence on he DUT o he lowe dose a e ( he esul s
ypically show ma ginally highe dose deg ada ion h eshold a lowe dose a es [14]).
Figu e 6depic s he co e cu en s o FPGA1 and FPGA2 h ough he i adia ion expe imen . I is
no ewo hy ha o bo h DUTs he co e cu en inc eased linea ly wi h he dose. This inc ease o leakage
cu en is ully accoun ed o in he c ea ion o elec on-hole pai s due o he ioniza ion o SiO2[15].
Elec onics 2018,7, 163 7 o 11
0 100 200 300 400 500
To al Dose (k ad(Si))
0.02
0.04
0.06
0.08
0.1
0.12
0.14
ICC (A)
FPGA1
FPGA2
Rep og amming E o (440 K ad)
Fi s Communica ion E o (410 K ad)
Figu e 6. In e nal Co e Cu en s. adia ion.
4.2. Delay Va ia ion o he RO Loop
RO equencies play a key ole in he RO-PUF au hen ica ion scheme, he e o e hey ha e been
subjec ed o exhaus i e analysis du ing he whole expe imen . Fo each RO, we ha e ca ied ou
100 equency acquisi ions in o de o imp o e he measu emen e o by a e aging he alues.
A p e-i adia ion condi ions, he a e age RO equencies o FPGA1 and FPGA2 a e 196.12 MHz and
199.16 MHz espec i ely. A he end o he expe imen [500 k ad(Si)], he a e age RO equencies o
FPGA1 and FPGA2 ha e dec eased o 195.1 MHz (0.5%) and 197.92 MHz (0.6%) espec i ely. Figu e 7
shows how he 512 RO a e age equencies o FPGA2 changed du ing he expe imen . I can be
app ecia ed ha a e a i s sha p dec ease a 10 k ad(Si), all he ROs ollow he same endency. I is
also wo hy o no e ha he e a e no many in e sec ions be ween he di e en lines, which indica es a
good equency s abili y. These esul s a e e y simila o hose epo ed in [
16
], whe e he e ec s o
ageing in a RO-PUF we e s udied.
Figu e 8shows he dis ibu ion o he a e age equencies a p e-i adia ion condi ions and a he
end o he expe imen . Once again, he equency s abili y can be highligh ed. The esul s o FPGA1
a e analogous o hose depic ed in Figu es 7and 8.
0 50 100 150 200 250 300 350 400 450 500
To al Dose (k ad(Si))
1.94
1.95
1.96
1.97
1.98
1.99
2
2.01
2.02
2.03
RO equencies (Hz)
108
Figu e 7. RO equencies s. adia ion.
Elec onics 2018,7, 163 8 o 11
1.94 1.95 1.96 1.97 1.98 1.99 2 2.01 2.02 2.03
RO equencies 108
0
10
20
30
40
50
60
70 End o he Expe imen
P e-i adia ion
1.96 1.97 1.98 1.99 2 2.01 2.02 2.03
RO equencies a P e-i adia ion condi ions (Hz) 108
1.94
1.95
1.96
1.97
1.98
1.99
2
2.01
RO equencies a he end o he expe imen (Hz)
108
Figu e 8.
FPGA2 equency dis ibu ion and sca e plo a p e-i adia ion condi ions and a 500 k ad(Si).
4.3. RO-PUF Quali y Fac o s
We ha e calcula ed he main me ics ela ed o PUF quali y in o de o e alua e he sui abili y
o he RO-PUF o space applica ions. To ha end, a 511-bi esponse has been gene a ed o each
accumula ed doses. This 511-bi esponse is ex ac ed by compa ing he a e age equencies o adjacen
pai s o ROs in he a ay.
4.3.1. Uni o mi y
Figu e 9depic s he uni o mi y o he PUF esponse du ing he expe imen . Fo bo h DUTs,
he esponse bi s a e ai ly e enly dis ibu ed among ‘0’ and ‘1’, showing almos an ideal dis ibu ion
h oughou he en i e expe imen . The a e age numbe o 1’s in he PUF esponse o FPGA1 and
FPGA2 a e 51.43% and 49.61%, espec i ely.
0 100 200 300 400 500
To al Dose (k ad(Si))
48.5
49
49.5
50
50.5
51
51.5
52
52.5
53
% Uni o mi y
FPGA1
FPGA2
Figu e 9. Uni o mi y s. adia ion.
4.3.2. Reliabili y
Figu e 10 shows he in a-die Hamming dis ance calcula ed using Equa ion (2). The p e-i adia ed
511-bi esponse has been se as he e e ence esponse. In bo h FPGAs, he ini ial in a-die HD is low
(>3%) and i inc eases wi h he accumula ed dose. This means ha he e is an inc easing deg ada ion
o he RO pe o mance ha is p opo ional o he adia ion dose.
Elec onics 2018,7, 163 9 o 11
0 100 200 300 400 500
To al Dose (k ad(Si))
2
3
4
5
6
7
8
9
10
11
12
% In a-die Hamming dis ance
FPGA1
FPGA2
Figu e 10. Reliabili y s. adia ion.
4.3.3. Uniqueness
Figu e 11 p esen s he in e -die Hamming dis ance ob ained using Equa ion (3). Once again, a loss
on he uniqueness can be obse ed and is p opo ional o he accumula ed dose.
0 100 200 300 400 500
To al Dose (k ad(Si))
40
41
42
43
44
45
46
% In e -die Hamming dis ance
Figu e 11. Uniqueness s. adia ion.
4.4. Resul Analysis
The inc ease in he co e cu en s a e consis en wi h o he esul s epo ed in he li e a u e o he
same FPGA [
17
]. In he case o FPGA1, he i s ailu e occu ed a 410 k ad(Si), which is also simila
o he i s ailu e epo ed in [
17
], whe e he DUT wo ked p ope ly un il 345 k ad(Si). Rega ding he
RO- equencies, he expe imen s ha e shown ha he e is a good s abili y on he equencies and he
changes due o he accumula ed dose a e negligible.
On he o he hand, he quali y me ics show ha he accumula ed dose makes he esponses
p oduced by he PUF un eliable. None heless, as he uni o mi y me ic shows, he andomness o
he esponse emains una ec ed by he o al dose. This may be due o he dec ease o he noise
pa ame e s ha ha e a di ec in luence on he andomness. Reliabili y is he key me ic o e alua e
a e he deploymen o PUFs in space. I a p e-i adia ion condi ions all he me ics ha e accep able
alues, only a dec ease in he eliabili y due o ionizing adia ion can a ec he es o he me ics.
In his case, he eliabili y me ic shows a li le deg ada ion ha can be co ec ed using some ypical
coun e measu es such as using only RO pai s wi h maximal equency di e ence [
2
] o using quan ize s
wi h eliabili y gua an ees [
18
]. Rega ding uniqueness, as a colla e al e ec o un eliabili y, he esul s
also show a deg ada ion o he me ic.