Secure EPC Gen2 compliant Radio Frequency
Identification
Mike Burmester
1
, Breno de Medeiros
2
, Jorge Munilla
3
, and Alberto Peinado
3
1
Department of Computer Science
Florida State University, Tallahassee, FL 32306, USA
burmester@cs.fsu.edu
2
Go ogle, Inc.
1600 Amphitheatre, Parkway Mountain View, CA 94043, USA
breno@brenodemedeiros.com
3
Departamento de Ingenier´ıa de Comunicaciones
Universidad de M´alaga, Spain
munilla@ic.uma.es, apeinado@ic.uma.es
Abstract. The increased functionality of EPC Class1 Gen2 (EPCGen2)
is making this standard a de facto specification for inexpensive tags in
the RFID industry. Recently three EPCGen2 compliant protocols that
address security issues were proposed in the literature. In this paper we
analyze these protocols and show that they are not secure and subject to
replay/impersonation and statistical analysis attacks. We then propose
an EPCGen2 compliant RFID protocol that uses the numbers drawn
from synchronized pseudorandom number generators (RNG) to provide
secure tag identification and session unlinkability. This protocol is opti-
mistic and its security reduces to the (cryptographic) pseudorandomness
of the RNGs supported by EPCGen2.
Keywords: EPCGen2 compliance, security, identification, unlinkability.
1 Introduction
Radio Frequency Identification (RFID) is a promising new technology that is
widely deployed for supply-chain and inventory management, retail operations
and more generally for automatic identification. The advantage of RFID over
barcode technology is that it is wireless and does not require direct line-of-sight
reading. Furthermore, RFID readers can interrogate tags at greater distances,
faster and concurrently.
One of the most important advantages of RFID technology is that tags have
read/write capability, allowing stored tag information to be altered dynamically.
Typically an RFID system consists of tags, one or more readers, and a back-end
server. The communication channel between the reader and the back-end server
is assumed to be secure while the wireless channel between the reader and the
tag is assumed to be insecure.
2 Burmester, M., de Medeiros, B., Munilla, J., and Peinado, A.
To promote the adoption of RFID technology and to support interoperability,
EPCGlobal [10] and the International Organization for Standards (ISO) [12] have
been actively engaged in defining standards for tags, readers, and the communi-
cation protocols. A recently ratified standard is EPC Class 1 Gen 2 (EPCGen2).
This defines a platform for the interoperability of RFID protocols, by support-
ing efficient tag reading, flexible bandwidth use, multiple read/write capabilities
and basic reliability guarantees, provided by an on-chip 16-bit Pseudo-random
Number Generator (RNG) and a 16-bit Cyclic Redundancy Code (CRC16).
EPCGen2 is designed to strike a balance between cost and functionality, with
little attention paid to security.
In this paper we are concerned with the security of EPCGen2 compliant
protocols. Clearly one has to take into account the additional cost for intro-
ducing security into systems with restricted capability. It is important therefore
to employ lightweight cryptographic protocols that are compatible with the ex-
isting standardized specifications. Several RFID authentication protocols that
address security issues using cryptographic mechanisms have been proposed in
the literature. Most of these use hash functions [16, 21, 2, 8, 19, 9, 15], which are
beyond the capability of low-cost tags and are not supported by EPCGen2.
Some protocols use pseudorandom number generators (RNG) [21, 13, 5, 4, 20, 3],
a mechanism that is supported by EPCGen2, but these are not optimized for
EPCGen2 compliance. One can also use the RNG supported by EPCGen2 as
a pseudorandom function (PRF) (as in [3, 11]) to link challenge-response flows,
however it is not clear if such protocols are vulnerable to related key attacks [3].
The research literature for RFID security is extensive. We refrain from a
detailed review, and refer the reader to a comprehensive repository available
online at [1]. Recently three RFID authentication protocols specifically designed
for compliance with EPCGen2 have been prop osed [7, 17, 18]. These combine the
CRC-16 of the EPCGen2 standard with its 16-bit RNG to hash, randomize and
link protocol flows, and to prevent cloning, impersonation and denial of service
attacks. In this paper we analyze these protocols and show that they do not
achieve their security goals. One may argue that, because the EPCGen2 standard
supports only a very basic RNG, any RFID proto col that complies with this
standard is potentially vulnerable, for example to ciphertext-only attacks that
exhaust the range of the components of protocol flows. While this is certainly
the case, such attacks may be checked by using additional keying material and
by constraining the application (e.g., the life-time of tags). We contend that
there is scope for securing low cost devices. Obviously, the level of security may
not be sufficient for sensitive applications. However there are many low cost
applications where there is no alternative.
The rest of this paper is organized as follows. Section 2 introduces the EPC-
Gen2 standard focusing on security issues. Section 3 analyzes three recently pro-
posed EPCGen2 protocols. In Section 4 we propose a novel EPCGen2 compliant
protocol that provides tag identification and session unlinkability. In Section 5
we define a security framework for Radio Frequency Identification, and show
that our protocol is secure in this framework.
Secure EPC Gen2 compliant Radio Frequency Identification 3
2 The EPCGen2 standard
EPC Global UHF Class 1 Gen 2, commonly known as the EPCGen2, was ap-
proved in 2004, and ratified by ISO as an amendment to the 18000-6 stan-
dard in 2006. This standard defines the physical and logical requirements for
a passive-backscatter, Interrogator-talks-first (ITF), radio-frequency identifica-
tion (RFID) system operating in the 860 MHz - 960 MHz frequency range. The
EPCGen2 standard defines a protocol with two layers, the physical and the
Tag-identification layer, which together specify the physical interactions, the op-
erating procedures and commands, and the collision arbitration scheme used to
identify a Tag in a multiple-tag environment.
The system comprises Interrogators, also known as Readers, and Tags. Below
we briefly summarize the EPCGen2 requirements.
1. Physical Layer
– Communications are half-duplex, meaning that Interrogators and Tags
cannot talk simultaneously.
– An Interrogator transmits information to a Tag by modulating an RF
signal. Tags are passive, meaning that they receive all of their operating
energy from the Interrogator’s RF waveform, as well as information.
– An Interrogator receives information from a Tag by transmitting a conti-
nuous wave (CW) RF signal to the Tag; the Tag responds only after
being directed to do so by an Interrogator, by modulating the reflection
coefficient of its antenna, thereby backscattering a weak signal.
2. Tag memory is logically separated into four distinct banks
– Reserved memory that contains a 32-bit kill password (KP ) to perma-
nently disable the Tag, and a 32-bit access password (AP ) used when
the Interrogator wants to write/read the memory.
– EPC memory that contains the parameters of a CRC16 (16 bits), pro-
tocol control (P C) bits (16 bits), and an electronic product code EP C
that identifies the Tag (32-96 bits).
– T ID memory that contains sufficient information to identify to a Reader
the (custom/optional) features of the Tag and tag/vendor specific data.
– User memory that allows user-specific data storage
3. Tag-identification layer
– An Interrogator manages Tag populations using three basic operations:
Select (the operation of choosing a Tag population), Inventory (the op-
eration of identifying Tags) and Access (the operation of reading from
and/or writing to a Tag).
– The Interrogator begins an inventory round by transmitting a Query
command in one of four sessions. An inventory operates in only one ses-
sion at a time, and the Interrogator inventories Tags within that session.
– A random-slotted collision algorithm is used. The Interrogator sends
a parameter Q, that is an integer in the range (0, 15); the Tags load
a random Q-bit number into a slot counter. Tags decrement this slot
counter when they receive a command (QueryRep), and reply to the
Interrogator when their counter reaches zero. When the Interrogator
detects the reply of a Tag, it requests its P C, EP C, and CRC16.
4 Burmester, M., de Medeiros, B., Munilla, J., and Peinado, A.
– Link cover-co ding can be used to obscure information during Reader to
Tag transmissions. To cover-co de data (or a password), an Interrogator
first requests a random number from the Tag. Then, the Interrogator
performs a bit-wise XOR of the data with this random number, and
transmits the result (cover coded or ciphertext) to the Tag.
4. Hardware requirements
– A 16-bit Pseudo-Random number generator (RNG).
– A 16-bit Cyclic Redundancy Code.
2.1 The Pseudo-Random Number Generator
A pseudorandom number generator (RNG) is a deterministic function that out-
puts a sequence of numb ers that are indistinguishable from random numbers by
using as input a random binary string, called seed. The length of the random
seed must be selected carefully to guarantee that the numbers generated are
pseudorandom. The state of the RNG changes each time that a new random
number is drawn. Although EPCGen2 does not specify any structure for the
RNG, it defines the following randomness criteria.
1. Probability of RN16: The probability that a pseudorandom number RN16
drawn from the RNG has value RN is bounded by:
0.8/2
16
< P rob(RN16 = RN ) < 1.25/2
16
.
2. Drawing identical sequences: For a tag population of up to 10,000 tags,
the probability that any two or more tags simultaneously draw the same
sequence of RN16s is < 0.1%, regardless of when the tags are energized.
3. Next-number prediction: A RN16 drawn from a tag’s RNG is not pre-
dictable with probability better than 0.025%, given the outcomes of all prior
draws.
We refer the reader to the discussion in [3] regarding the strength of EPCGen2
compliant RNGs.
2.2 The 16-bit Cyclic Redundancy Code
Cyclic Redundancy Codes (CRC) are error-detecting codes that check accidental
(non-malicious) errors caused by faults during transmission. To compute the
CRC of a bit string B = ( B
0
, B
1
, . . . , B
m−1
) we first represent it by a polynomial
B(x) = B
0
+B
1
x+· · ·+B
m−1
x
m−1
over the finite field GF (2), and then compute
its remainder: CRC(B(x)) = (B(x) · x
n
) mod g(x), for an appropriate generator
polynomial g(x) of degree n.
EPCGen2 uses the CRC-CCITT generator: x
16
+ x
12
+ x
5
+ 1, and XORs a
fixed bit pattern to the bitstream to be checked. EPCGen2 specifies the Cyclic
Redundancy Code CRC16 which, for a 16-bit number B is defined by:
CRC(B) = [ B(x) · x
16
+
31
X
i=16
x
i
] mod g(x) = B(x)x
16
mod g(x) + CRC(0),
Secure EPC Gen2 compliant Radio Frequency Identification 5
where CRC(0) =
P
31
16
x
i
mod g(x) is a fixed polynomial. Since the modulo g(x)
operator is a homomorphism, CRC16 inherits strong linearity aspects. More
specifically, if P , Q are 16-bit numbers, then
CRC(P (x) + Q(x)) = CRC(P (x)) + CRC(Q(x)) + CRC(0). (1)
It follows that the CRC16 of a sequence of numbers can be computed from the
CRC16s of the numbers. Consequently CRC16 by itself will not protect data
against intentional (malicious) alteration. Its functionality is to support strong
error detection particularly with respect to burst errors, not security.
3 Weaknesses in recently proposed EPCGen2 compliant
RFID protocols
In this section we consider three recently proposed EPCGen2 compliant pro-
tocols: the Chen-Deng mutual authentication protocol [7], the Quingling-Yiju-
Yonghua minimalist mutual authentication protocol [17], and the Sun-Ting au-
thentication protocol [18]. We show that these protocols fall short of their claimed
security.
In the protocols below we use the following notation: S is the back-end server,
R a Reader, T a tag. We assume that S and R are linked with a secure channel,
and for simplicity, only consider the case when the authentication is online.
3.1 Analysis of the Chen-Deng protocol
In the Chen-Deng mutual authentication protocol [7] each tag T shares three
private values with the back-end server S: a key K, a value (incorrectly called
nonce) N and an EPC identifier. The tag stores these in non-volatile memory
and the server stores them in a database DB. The protocol has three passes:
1. S ⇒ R → T : query, R
r
, a random number, and P = CRC(N ⊕ R
r
).
T : Check that P is correct. If it is correct,
2. T → R ⇒ S : R
t
, a random number, X = (K ⊕ EP C ⊕ R
t
) and
Y = CRC(N ⊕ X ⊕ R
t
).
S : Check that X, Y are correct. If they are correct,
3. S ⇒ R → T : M
resp
, a response message.
This protocol is clearly subject to a replay attack since the flows from the Reader
R and tag T use independent randomness (and hence are independent). In fact
the adversary needs only one interrogation of T : R
t
, X = (K ⊕ EP C ⊕ R
t
)
and Y = CRC(N ⊕ X ⊕ R
t
), to impersonate the tag by computing a valid
(R
a
, X
∗
, Y
∗
), for any random number R
a
, as: X
∗
= X ⊕ (R
t
⊕ R
a
), Y
∗
= Y
(Note that new P
∗
= P ⊕ CRC(R
r
⊕ R
a
) ⊕ CRC(0) can be also computed).