A NEW ALGORITHM TO CONSTRUCT S-BOXES WITH HIGH DIFFUSION

International Journal of Soft Computing, Mathematics and Control (IJSCMC),Vol. 4, No. 3, August 2015
DOI : 10.14810/ijscmc.2015.4303 41
A NEW ALGORITHM TO CONSTRUCT S-BOXES
WITH HIGH DIFFUSION
Claudia Peerez Ruisanchez
Universidad Autonoma del Estado de Morelos
ABSTRACT
In this paper is proposed a new algorithm to construct S-Boxes over GF(28) with Branch Number value at
least 3. This is an important property that guarantees a high diusion in the S-Box [12]. Also are introduced
some defnition and properties that show the be- havior of S-Boxes under the composition with affine
functions. Finally is presented a comparision between this algorithm and the method pro-posed by Tavares
[12].
1 INTRODUCATION
In his paper Communication Theory of Secrecy Systems", Claude Shannon defines confusion and
diffusion as two essential properties for block cipher design. S-Boxes to be used in this
cryptographic algorithms require the compliance of criteria such as Nonlinearity, Branch
Number, Algebraic Degree and Differential Characteristic, which bring security against
differential and linear cryptanalysis [9][10].
In 1994 Tavares [12] proposes a method to and S-Boxes with a desired value of Branch Number.
The principal problem of this proposal is that the obtained S-Boxes can have a big number of
fixed points and his Nonlinearity values were small to be used in Block Ciphers.
In this paper we propose a new algorithm to construct S-Boxes with Branch Number at least 3. In
this method the obtained S-Boxes doesn't have fixed points and his Nonlinearity values are
acceptable. Also we present the definition of Parameter Set and introduce some properties that
allow show when the parameter set of a S-Box is invariant under the composition with affine
transformations. This have an special importance because we can obtain dynamic S-Boxes for
Block Ciphers constructing one with good Parameter Set and making the composition with a_ne
functions.
2 THEORETICAL ASPECTS
In this section are presented some definitions related with S-boxes design that are very important
for a better understanding of the paper.

42
Definition 1. A function II : GF(2)n
→ GF(2)m
is called an n × m S-Box, in which any input of n
bits is transformed into output of m bits
There are many criteria dedicated to measure the cryptographic security of S-Boxes. However,
does not exists an universal measure that allow affirm which of this are the most important to take
into account in S-Box design. We will work with the four criteria presented bellow [11]:
Definition 2 (Nonlinearity). Let f : (GF(2))n
→ (GF(2))n
the nonlinearity of this function denoted
by nl(f) is the magnitude:
when An is the set of affine functions and wt is the Hamming weight.
Definition 3 (Algebraic Degree). Let f(x1; x2…… xn) a boolean function of n variables, then f
could be represented as (ANF):
when the coefficients a0, ai, aij,…….,a12…..n € {0,1}. The number of variables in the product
of higher order with non-zero coefficients is known as algebraic degree of f.
The algebraic degree of a vectorial boolean function is the minimum algebraic degree of all non-
zero linear combinations of his component functions.
Definition 5 (Branch Number). The branch number (diffusion coefficient) of f : GF(2)n
→GF(2)n
is defined as the value obtained from the expression:
This four criteria: branch number, algebraic degree, nonlinearity and differential characteristic are
very important because they provide the necessary elements against the linear and differential
cryptanalysis.

43
Linear Cryptanalysis The main objective of linear cryptanalysis if to find a linear probabilistic
relation for a part of the cipher equation that connects the input and the output data. Also this
relation must be complied with a probability significantly different of 1/2.
If this probability is denoted by pL is necessary minimize the magnitude to obtain a
good security against this kind of attack. If we assume that the X was selected randomly, then; the
better linear approximation of an S-Box in the form
be obtained with probability pe. Also
For example if we have a SP network with 10 rounds and an 8 × 8 S-Box with NL(II) = 112 like
in AES, will be necessary at least 262
known plain-text to obtain a bit of the key.
Differential Cryptanalysis In many of his papers Biham and Shamir show the witness of some
ciphers against the differential cryptanalysis for this reasonis fundamental guaranties the
protection against this kind of attack.
In [12] is defined the probability of the difference par (∆X, ∆Y) in an S-Box to the probability
that ∆Y occurs when one of the input values is selected randomly and the other is deduced from
the relation ∆X. The probability of the most frequently par was denoted as pδ.
Many authors affirm that the minimization of pδ brings resistance against differential
cryptanalysis . O'Connor [13] shows that for a values of n, The probability of the difference par in
an S-Box is expected less than the value . For example in an 8 × 8 S- Box pδ. ≤ 2-4
.
Another frequently cause of high probability characteristic is the pour diffusion in the S-Box bits
[13][12].
3 S-BOX CHARACTERIZATION
In this section we introduce some definitions that allow us characterize S-Boxes from the values
obtained in the four criteria shown before.
Definition 6. Let II : GF(2n
) → GF(2n
) we called Parameter Set of II to the vector
when the parameters pe, bn, pδ, deg denote the static probability, branch number, characteristic
probability and the algebraic degree of II respectively.

44
The definition shown before brings the necessary tools to introduce the next definition from
whom will be characterize the S-Boxes.
Another important approach would be posses dynamic S-Boxes in the cipher. However in
practice is very difficult find S-Boxes with an acceptable parameter set. With this porpoise is
presented the next definition.
Definition 8. Let II1 : GF(2n) → GF(2n
) and II2 : GF(2n
) → GF(2n
). We say that II1 y II2 are left
affine equivalents if exists an affine function A : GF(2n
) → GF(2n
) so that:
Affine functions are very easy to construct, for this reason we could maybe think about obtain S-
Boxes with an acceptable parameter set from the composition of one with an acceptable
parameter set selected a priory and an affine function. Take this into account is important to know
the be haviour of this composition. For this reason we consider the next example.
The above example serves as a basis to demonstrate the following properties, which are an
essential contribution to the development of this study.
Proposition 1. Let II1 : GF(2n
) → GF(2
n
) and II2 : GF(2n
) → GF(2n
)S-Boxes left affine
equivalent. Then we have that:

45
For static probability a similar result could be obtain:
Proposition 2. Let II1 : GF(2n
) → GF(2n
) and II2 : GF(2n
) → GF(2n
) S-Boxes left affine
equivalents. Then we have that:
In the other side we have that:

46
To see how is the behaviour of branch number of an S-BOX under left affine equivalents
transformations we consider the next example:
And

47
4 PRESENTATION OF THE NEW ALGORITHM
As we show before S-Boxes with good diffusion in the output bits posses a better resistance
against differential cryptanalysis. The branch number is an S-Box design criterion that results as a
diffusion measure. Unfortunately experimental results show that S-Boxes with non-trivial
diffusion orders cant be determinate by randomly search (bn > 2).
Table 1. Probability of randomly selection of S-Boxes with bn≥3

48
A new algorithm for the construction of S-Boxes with bn = 3 is presented below.
For a better understanding of the algorithm we consider the next sets:
B(i) is the set formed by the images of the elements less equal than i.
Then the construction is performed as follows:
A random value different to 0 is assigned to the S-box for input 0.
In the other way if ᴲ K< i so that (i, k) € L1.

49
4.1 Comparative results
In [12] Tavares propose an algorithm for the construction of S-Boxes with bn ≥ 3, however this
algorithm present some inconvenient. The 8 × 8 constructed S-Boxes could posses a high quantity
of fixed points. Also the nonlinearity values for this constructions are always less equal than 96 as
we can see in the table 3.
Table 2. Nonlinearity of 8 × 8 S-Boxes in the new construction.
We generate a set of 50 S-Boxes in which we find his nonlinearity values and the results are
showed in the table 3.
Is easy to see that if we compare the result of table 3 the output S-boxes of the new algorithm
posses nonlinearity values higher than the S-boxes obtained by the Tavares algorithm.
Table 3. Comparision between the new algorithm and the Tavares proposal.
5 CONCLUSIONS
In this paper is presented an algorithm to generate S-boxes with high diffusion. Also was shown
that these outputs could have acceptables values of nonlinearity and do not have fixed point wich
allows that these S-boxes could be used as part of the cipher in a secure way.
As a theoretical result was proved that the Parameter Set of an S-box is not invariant under left
affine transformations. Take this into account is not possible obtain S-boxes with a good
Parameter Set by the left composition with an affine function and a good S-box.
Finally was shown that the outputs of the new algorithm have better results than the outputs of the
Tavares algorithm.

50
REFERENCES
[1] Daemen J. and Rijmen V. The Rijndael block cipher. AES Proposal.
http/www.daimi.ai.dk/iran/rijndael.pdf. 1999.
[2] Mister.S. & Adams.C. Practical S-Box Design. Nortel, P.O. Box 3511, Station C, Canada. 1990.
[3] Webster.A.F. & Tavares.S.E. On the design of S-Boxes. Department of Electrical Engineering
Queen's University. 1989.
[4] Heys.H. & Tavares.S.E. Avalanche Characteristics of Substitution-Permutation Encryption Networks.
IEEE. 1991.
[5] Vergili.I & Yucel.M.D. Statistics on satisfaction of Security Criteria for RandomlyChosen S-Boxes.
Middle East Technical University. 2000.
[6] Daemen.J. & Rijmen.V. The Design of Rijndael. Springer. 2002.
[7] Przemyslaw.R. & Mroczkowski.P. How to create goog s-boxes. 1st International Conference for
Young Researchers. 2006.
[8] Nyberg.K.Dierentially uniform mappings for cryptography. Institute of Computer Technology,
Vienna Technical University. 1994.
[9] Shamir.A.&Biham.E.Dierential Cryptanalysis of DES-like Cryptosystems. Journal of Cryptology.
Vol.4, No 1. 1991.
[10] Matsui.M. Linear cryptanalysis method for DES cipher. Advances in Cryptology: Proceedings of
EUROCRYPT 93, Spriger-Verlag, Berlin. 1994.
[11] Perez, Claudia and Gainzo, Felipe. Algunos criterios para determinar la seguridad en las S-cajas.
Memorias del evento Compumat 2013.
[12] Heys. H.M & Tavares. S. E. Substitution-Permutation Networks Resistant to Differential and Linear
Cryptanalysis. Department of Electrical and Computer Engineering, Queens University, Kingston,
Ontario, Canada K7L 3N6. 1994.
[13] OConnor. L. J. On the distribution of characteristics in bijective mappings. Advances in Cryptology:
Proceedings of EUROCRYPT 93, Springer-Verlag, Berlin, pages 360370. 1994.
[14] Youssef. A.M., Mister. S. & Tavares. S.E. On the Design of Linear Transformations for Substitution
Permutation Encryption Networks. Department Of Electrical and Computer Engineering Queens
University, Kingston, Ontario, Canada. 1998.

A NEW ALGORITHM TO CONSTRUCT S-BOXES WITH HIGH DIFFUSION

More Related Content

Similar to A NEW ALGORITHM TO CONSTRUCT S-BOXES WITH HIGH DIFFUSION (20)

Recently uploaded (20)

A NEW ALGORITHM TO CONSTRUCT S-BOXES WITH HIGH DIFFUSION