Content uploaded by Lilian Bossuet
Author content
All content in this area was uploaded by Lilian Bossuet on Aug 27, 2015
Content may be subject to copyright.
New Paradigms for Access Control in Constrained
Environments
A. Cherkaoui∗, L. Bossuet∗, L. Seitz†, G. Selander‡and R. Borgaonkar§
Hubert Curien Lab∗(France), SICS Swedish ICT†(Sweden), Ericsson Research‡(Sweden), TU Berlin / T-Labs§(Germany)
Emails: abdelkarim.cherkaoui@univ-st-etienne.fr, lilian.bossuet@univ-st-etienne.fr, ludwig@sics.se,
goran.selander@ericsson.com, ravii@sec.t-labs.tu-berlin.de
Abstract—The Internet of Things (IoT) is here, more than 10
billion units are already connected and five times more devices
are expected to be deployed in the next five years. Technological
standarization and the management and fostering of rapid
innovation by governments are among the main challenges of
the IoT. However, security and privacy are the key to make
the IoT reliable and trusted. Security mechanisms for the IoT
should provide features such as scalability, interoperability and
lightness. This paper adresses authentication and access control
in the frame of the IoT. It presents Physical Unclonable Functions
(PUF), which can provide cheap, secure, tamper-proof secret keys
to authentify constrained M2M devices. To be successfully used
in the IoT context, this technology needs to be embedded in a
standardized identity and access management framework. On
the other hand, Embedded Subscriber Identity Module (eSIM)
can provide cellular connectivity with scalability, interoperability
and standard compliant security protocols. The paper discusses
an authorization scheme for a constrained resource server taking
advantage of PUF and eSIM features. Concrete IoT uses cases
are discussed (SCADA and building automation).
I. INTRODUCTION
We evolve because we communicate. Data is interpreted as
information, from which we derive a knowledge that translates
into wisdom. Considering the impact the Internet already has
had on education, communication, business and science, it
certainly appears to be one of the most important creations
in human history. Now, the evolution of the Internet is leading
to a global network of objects, which is commonly refered to
as the Internet of Things (IoT). The IoT enables the Internet to
reach out to the real world of physical objects by combining
their ability to sense, collect data, transmit it, analyze it, and
distribute it on a massive scale.
Ultimately, everything would be connected anytime, at
anyplace. This convergence of virtual and physical worlds can
drastically improve the user’s experience, but it also presents
several challenges regarding security and privacy, which are
among the main barriers for the deployment of IoT on a
broad scale [1]. IoT security needs to be garanteed on several
levels. Communication with the IoT needs to be encrypted
with proven algorithms using keys with high entropy that
are securely exchanged between the user and the IoT. The
communication channel is usually secured using a symmetric
encryption algorithm like AES, while public key encryption is
prefered for low data rate communication (e.g. key exchange).
Ideally, an IoT chip would embed a crypto-processor which
performs the encryption tasks and provides secure key man-
agement (key generation, key storage, etc).
Authentication using secret keys stored in non volatile mem-
ories presents nowadays many vulnerabilities, principaly due
to the development of active attacks (e.g. probing) and passive
attacks (e.g. differential power analysis). Protection mecha-
nisms against these attacks are expensive and not adapted
to devices with constrained size and energy. On the other
hand, future IoT scenarios would involve billions of heteroge-
neous devices which some of them maybe reprogrammable.
In many cases, non expert users would define policy and
permissions for the use of their own resources. Therefore,
security mechanisms for the IoT should also provide features
such as scalability, interoperability while still being sufficiantly
light.
This paper discusses two technologies that, combined to-
gether, provide most of the building blocks that meet these
requirements: Physical Unclonable Functions (PUF) provide
secure, low-cost authentication means in constrained devices
while eSIM provide cellular connectivity with the flexibility
to change operator or late binding of subscription needed in
many IoT use cases. Section II presents the PUF technology
and discusses a few design candidates that seem suitable for
the IoT use cases. Section III presents eSIM and its features.
Section IV describes the identity and access management
framework to show how these technologies fit together in
a security architecture for IoT devices. Finally, Section V
concludes the paper.
II. PHYSICAL UNCLONABLE FUNCTIONS - PUF
Privacy is an important prerequisite in most IoT use cases,
especially when the managed data is sensitive. This is true
even for a simple device like a sensor: the consequences
of a compromised temperature sensor in a power plant can
range from costly to disastrously. PUF provide a promising
framework for authentication in IoT architectures especially
in reconfigurable and/or constrained devices.
A. Privacy by design
Nowadays, traditionnal authentication methods based on
secret digital keys often require additionnal protection mecha-
nisms for the key storage. In fact, numerous active and passive
attacks which aim at extracting these keys have been developed
and reported over the past several years. On the other hand,978-1-4799-5810-8/14/$31.00 c
2014 IEEE
FPGA-based reconfigurable devices are increasingly growing
in the market of embedded and mobile applications. Integrat-
ing secure non-volatile memories in FPGA significantly raises
the fabrication overhead and production costs: in fact, most
commercial FPGA do not include it. The storage of secret keys
in FPGA therefore requires external memory with additionnal
contermeasures to protect it against attacks.
The concept of PUF was first introduced by Pappu in
[2]. PUF introduce a new paradigm shift from explicitely
programmed digital identity to unclonable physical identity.
They are mostly electrical constructions that extract a unique
secret key from physical parameters of the device: the chal-
lenge/response procedure is based on a physical interaction
which is theoretically unclonable. Entropy is derived from
a physical random variable as the mismatch in transistor
attributes (length, width, oxide thickness, etc) due to manu-
facturing process variability (MPV). The founding principle
is that MPV are not controllable (they are not predictable)
and not reproducible. Therefore and ideally (if the extraction
mechanism is properly designed), a PUF extracts secret keys
which are unique (each device has a unique, non reproducible
ID based on its unique physical characteristics), random (it
is impossible to predict the response of a device to a given
challenge), reliable (each device reproduces the same response
to a give challenge) and tamper resistant (probing the PUF
changes its physical behavior and thus the obtained response).
Therefore, a PUF can be seen as a function returning a
fingerprint of the device in which it is implemented, or even
of a specific part inside the device.
B. A closer look at PUF designs
Traditionnally, silicon based PUF rely on manufacturing
process variability (MPV) to generate unique, reliable and
unpredictable identifiers. There exist many silicon PUF archi-
tectures, but there are two main approaches to extract secrets
from MPV: methods based on delay measurements and other
methods related to the resolution of a metastability situation.
SRAM-PUF [4] and butterfly PUF [5] rely on the settling state
of a couple of cross-coupled elements. At the initialization
of an SRAM, most cells outputs are biased toward ’1’ or
’0’ depending on MPV. The arbiter PUF [3] relies on the
race of two events (electrical transitions) in two symmetrical
delay lines. The Ring Oscillator based PUF [6] (RO-PUF)
leverages the frequency mismatch between several identically
designed ring oscillators (RO). Most recent PUF architectures
are based on differential measurements in order to improve
the responses stability against environmental changes (mainly
temperature and voltage). PUF designs are often characterized
in terms of intra-device variation (a value close to 0% means
that PUF responses are reliable) and inter-device variation (a
value close to 50% means that PUF responses are unique).
Some PUF designs can provide an additionnal True Random
Number Generator (TRNG) function with little design over-
head since the entropy extraction methods are very similar to
those used usually in PUF, the variable being the source of
entropy targetted (MPV for PUF against noise for TRNG).
This feature is particulary interesting in constrained devices
because it allows to provide high entropy keys for encryption
mechanisms with very little design effort (implementing each
feature independently would be much more expensive). Table
I presents a comparison between four PUF designs in terms of
uniqueness, reliability, mathematical unclonability, the ability
to provide an additionnal TRNG function and their imple-
mentation effort. Intra-device and inter-device variations are
provided in [7] for ASIC and FPGA implementations.
TABLE I
COMPARISON OF THE MAIN PUF ARCHITECTURES IN DIGITAL DEVICES
(*INT RA-DEVICE AND IN TER-DEVICE VARIATIONS HAVE BEEN
CHARACTERIZED IN [7], PP = REQ UIR ES H EAVY PO ST-PRO CE SSI NG, D O =
INVOLVES AN IMPORTANT DESIGN OVERHEAD)
Butterfly
PUF
SRAM
PUF
Arbiter
PUF
RO PUF
Challenge cell selec-
tion
SRAM
adress
delay path
selection
RO selec-
tion
Response settling
state
memory
state
delay
length
oscillation
frequency
Inter-dev.
var.*
50% 50% 38% 46%
Intra-dev.
var.*
6% 12% 10% 0.5%
Math.
clonabil-
ity
no no yes possible
TRNG DO and PP DO and PP no DO
Implemen-
tation
easy easy difficult easy
Arbiter PUF seem nowadays obsolete considering their low
uniqueness and reliability. Their main flaws are the difficulty
to place and route identical delay lines which results in a
low entropy in the subsequent responses, and therefore a high
vulnerability to modeling attacks. This is more especially the
case in FPGA in which routing circuitery is often complex and
uses active elements such as multiplexors. SRAM and butterfly
PUF have remarkable uniqueness properties while being fairly
reliable. They can be easily embedded in most targets (using
SRAM, flip-flops, latches, bus keepers, etc) and are known to
be resistant to modeling attacks. Additional TRNG function
can be obtained but at a high cost. In fact, very few memory
cells would have unpredictable behavior and there is no prior
way to determine which cells should be used for the TRNG.
Also, one main barrier for their usage in many of the IoT use
cases is their low number of challenge/response pairs.
RO-PUF is reliable and has good uniqueness. Its implemen-
tation is straightforward since inverter ring oscillators integrate
very well in all ASIC and FPGA design flows. The number of
pair of challenge/response is potentially very large (2nwhere
nis the number of oscillators) although some of them may be
correlated. An additionnal TRNG feature can be implemented
by integrating a XOR tree at the outputs of the ring oscillators
(the size of the design is approximately doubled in the case
of rings of tens of elements). Until recent years, RO-PUF was
considered as a promising candidate for large scale usage of
PUF. Yet, recent studies highlighted two security issues that
may change this status: the mutual influence of RO frequencies
through supply lines (locking phenomenon) which can be
maliciously exasperated to fault the PUF behavior, and even
worse, the possibility of extracting the RO frequencies through
contactless electromagnetic charecterization [8] without af-
fecting the PUF behavior. The first case would be simply
a denial of service, while the latter could possibly allow to
mathematically clone the identifier (even though such attack
has not been performed yet).
C. TERO-PUF: a promising PUF candidate ?
The Transient Element Ring Oscillator based PUF (TERO-
PUF) is a delay based PUF which has been recently pro-
posed in [9]. The main argument of TERO-PUF is that it
reproduces most of RO-PUF features (good uniqueness and
reliability, straightforward implementation, large number of
challenge/response pairs) while presenting considerably less
security flaws and providing the TRNG feature with a very
low design effort.
A full TERO-PUF architecture is composed of several
TERO loops, whose architecture is described in Fig. 1. Each
TERO loop consists of a bistable circuit composed of two
intialization stages and 2 branches (ideally symmetrical) of
an odd number of inverters. After the initilization, two events
(electrical transitions) start propagating across the TERO loop
which provokes a periodic oscillation of the output. Due to
the charge and discharge phenomena, there occurs a drafting
effect where one event closes the distance to the other until
they ultimately collide which stops the oscillation. Since
the number of oscillations in each TERO loop depends on
manufacturing process variability that affects individually each
logic cell in the device, it appears natural to use a counter
as an entropy extractor. An 8-bit counter is placed at the
output of each TERO loop. Challenges consist of selecting
two TERO loops. Multiplexors allows to select pairs of TERO-
loops whose number of oscillation is compared to provide the
subsequent response to the challenge.
Fig. 1. TERO-PUF basic cell architecture
The number of oscillations in each TERO loop depends on
three parameters:
Intrinsic noise: The effective number of oscillations is directly
affected by intrinsic noise fluctuations (white noise, flicker
noise, etc) in each logic cell of the TERO loop. Therefore, the
reliability of each bit of the output comparison decreases from
most significant bits (MSB) to least significant bits (LSB).
Manufacturing process variability: The mean number of
oscillations depends on manufacturing process variability, it
is independent of noise fluctuations.
Charge and discharge parameters: The dependency of the
number of oscillations on the charge and discharge param-
eters is maybe, unexpectedly, the most interesting feature
of TERO-PUF. Probing the output signal of a TERO loop
would necesseraly change its output capacitance, resulting in
a change in its number of oscillations. This features makes
the TERO-PUF strongly tamper evident and resistant against
active attacks that aim at cloning the identifier. Moreover,
TERO-PUF are non vulnerable to contacless electromagnetic
characterization methods as for RO-PUF. In fact, these meth-
ods are based on frequency analysis, they cannot detect brief
transient oscillations in the case of TERO-PUF.
The two MSB of the 8-bit counters were used to build 128-
bit, 189-bit and 252-bit signatures which have been evaluated
in terms of uniqueness in reliability in 36 PUF instances in
Altera Cyclone II FPGA. Results are presented in Table II:
they show that TERO-PUF has good uniqueness and reliability
properties in FPGA implementations.
TABLE II
UNIQUENESS AND STABILITY OF IDS GE NER ATED U SIN G A 64-LO OP
TERO PUF IN AN ALTE RA CY CLO NE II FPGA
ID size (bits) Intra-device variation (%) Inter-device variation (%)
126 1.73 48.07 %
189 2.07 48.99 %
252 2.75 49.27 %
III. EMBEDDED SIM
Innovation regarding IoT is rapidely increasing, a wider
adoption of the IoT requires pressing the capitals and reducing
the operational costs. Classical removable SIM (Suscriber
Identity Module) cards and their logistic are certainly a bar-
rier for the development of M2M wireless communications.
Changing SIM card is problematic in many business cases:
many M2M devices are remotely deployed, often hermetically
sealed, their after sale location is not known during production
and furthermore their product life cycles are lengthy1(network
operator may change during their life time).
To overcome these issues, the GSMA has developed the new
embedded SIM (eSIM) standard to fullfill all the scalability,
interoperability and over-the-air (OTA) connectivity require-
ments for an array of new connected products. eSIM is a
non-removable, standard compliant, physical SIM specially
designed for M2M devices and which can provide secure
connectivity to the IoT. Its main features are:
Secure remote provisioning: provisioning of one or mul-
tiple operator credentials into a SIM, remote enable-
ment/disablement of the operator credentials within the SIM
(which enables a change of active operator), remote deletion
of an operator credentials within a SIM. Remote provisioning
can be performed OTA with encrypted packets or using SMS
or https connection.
New network elements: Subscription Manager - Data Prepa-
ration (SM-DP) used to securely encrypt operator credentials
and Subscription Manager - Secure Routing (SM-SR) used
to securely deliver the credentials to the SIM and remotely
1http://www.gsma.com/connectedliving/embedded-sim/
manage the SIM once they are installed. These network
elements make easier selecting and installing different mobile
operator credentials once the M2M device has been deployed.
eSIM also provide tamper-proof key storage that can be
used to authentify M2M devices, however at higher cost
than PUF (especially when the other eSIM services are not
required). Therefore, PUF seem suitable for a large number of
constrained devices which does not need to be individually
connected to the IoT (sensors, actuators, etc), while eSIM
would be suitable for a smaller number of more powerful
devices (e.g. control devices, gateways).
IV. PRACTICAL USE CASES IN THE FRAME OF THE IOT
While PUF can provide a cheap, tamper-proof, secret
key which can be used for authentication, this technology
needs to be embedded in a standardized identity and access
management framework in order to reap its benefits. The
target platform that would use PUF would be low cost,
mass produced IoT devices with very constrained resources
(battery driven, very small volatile and persistent memory, low
processor power). These devices would typically interact with
more powerful devices, such as gateways, client devices or
control units that would be equipped with eSIM. Standardized
authentication protocols would use the secret keys provided by
either PUF or eSIM to authenticate the different IoT devices in
such a framework. Based on that authentication, access control
can be performed in order to secure access to the data and
functions provided by the IoT devices.
Fig. 2. Authentication and authorization scheme for constrained M2M
devices using PUF and eSIM features
In Fig. 2 we present such a framework, which involves four
parties: a constrained resource server (the IoT device) which
authenticates using PUF, a client with an eSIM who wants to
access the resource server, a back-end authorization server, and
a resource owner, whose role is limited to the deployment steps
which consist of bootstrapping the devices, registering them in
the authorization server then setting the authorization policies
(steps 1. and 2.). Making access control decisions has been
offloaded to the authorization server (which is unconstrained),
whereas the resource server only needs to enforce these
decisions. This design allows to minimize the functionality
that needs to be implemented on the constrained IoT devices.
A more detailed description of this design is presented in [10].
This scheme can be applied for a number of uses cases such
as building automation and SCADA (Supervisory Control And
Data Acquisition). Table III illustrates how PUF and eSIM can
be effectively utilized in both those use cases.
TABLE III
PUF AN D ESIM UT IL IZATI ON I N TWO IOTP RAC TIC AL U SE CA SE S
Use case PUF eSIM
Building
automa-
tion
heaters, temperature
sensors, smoke detectors,
doorlocks, cameras
gateway, e-car
SCADA sensors and actuators in a
refinery, in an oil platform
gateway of the oil
platform/rafinery,
transport vehicules,
petrol stations
V. CONCLUSION
Security protocols for the IoT need to be flexible and
scalable while still being compliant with communication stan-
dards. This paper adresses authentication and access control
(ACC) in constrained environments connected to the IoT. Two
promising technologies are presented (PUF and eSIM) and
an ACC framework and use cases are discussed. While PUF
are appropriate to authenticate security critical constrained
devices, eSIM provide all the credentials to securely connect
to the IoT and communicate with it. The proposed setup takes
fully advantage of eSIM standard compliance and flexibility
features (remote provisionning, late binding, etc) and PUF
lightness and security features (tamper evidence, the impos-
sibility to physically clone identifiers, etc).
ACK NOW LE DG ME NT
This research work is in the frame of the EIT (European
Institute of innovation and Technology) ICT activity 14056.
REFERENCES
[1] D. Miorandi, S. Sicari, F. Pellegrini, and I. Chlamtac. Internet of
Things: Vision, Applications and Research Challenges. Ad Hoc Networks,
10(7):14971516, September 2012.
[2] Ravikanth S. Pappu. Physical one-way functions. PhD Thesis, MIT, 2001.
[3] B. Gassend, D. Lim, D. Clarke, M. Van Dijk, and S. Devadas. Iden-
tification and authentication of integrated circuits. Concurrency and
Computation: Practice and Experience, 16(11):10771098, 2004.
[4] J. Guajardo, S.S. Kumar, G.J. Schrijen and P. Tulys. FPGA Intrinsic PUFs
and Their Use for IP Protection. In Proc. of Int. Conf. on Cryptographic
Hardware an Embedded Systems (CHES), Springer, LNCS, vol. 4727,
pp. 63-80, 2010.
[5] S.S. Kumar, J. Guajardo, R. Maes, G.-J. Schrijen and P. Tulys. Extended
Abstract: The Butterfly PUF Protecting IP on every FPGA. In Proc. of
Int. Sym. on Hardware-Oriented Security and Trust (HOST), pp. 67-70,
2008.
[6] A. Maiti, J. Casarona, L. McHale and P. Schaumont. A large scale
characterization of RO-PUF. In Proc. of Int. Sym. on Hardware-Oriented
Security and Trust (HOST), IEEE, pp.94-99, 2010.
[7] Roel Maes, Ingrid Verbauwhede. Physically Unclonable Functions: A
Study on the State of the Art and Future Research Directions. Towards
Hardware-Intrinsic Security 2010: 3-37.
[8] P. Bayon, L. Bossuet, A. Aubert, V. Fischer. EM leakage analysis on
True Random Number Generator: Frequency and localization retrieval
method. In Proc. of Asia-Pacific Int. Symp. And Exh. On Electromagnetic
Compatibility (APEMC), 2013.
[9] L. Bossuet, X.T. Ngo , Z. Cherif, V. Fischer. A PUF based on transient
effect ring oscillator and insensitive to locking phenomenon. IEEE Trans.
Emerg. Top. Comput., 2013.
[10] L. Seitz, G. Selander, C. Gehrmann. Authorization Framework for the
Internet-of-Things. In Proc. 4th IEEE International Workshop on Data
Security and PrivAcy in wireless Networks (D-SPAN ’13), 2013.
