CA2711188C - Public key cryptography utilizing elliptic curves - Google Patents
Public key cryptography utilizing elliptic curves Download PDFInfo
- Publication number
- CA2711188C CA2711188C CA2711188A CA2711188A CA2711188C CA 2711188 C CA2711188 C CA 2711188C CA 2711188 A CA2711188 A CA 2711188A CA 2711188 A CA2711188 A CA 2711188A CA 2711188 C CA2711188 C CA 2711188C
- Authority
- CA
- Canada
- Prior art keywords
- vector
- shift registers
- register
- representing
- curve
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Lifetime
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
- G06F7/60—Methods or arrangements for performing computations using a digital non-denominational number representation, i.e. number representation without radix; Computing devices using combinations of denominational and non-denominational quantity representations, e.g. using difunction pulse trains, STEELE computers, phase computers
- G06F7/72—Methods or arrangements for performing computations using a digital non-denominational number representation, i.e. number representation without radix; Computing devices using combinations of denominational and non-denominational quantity representations, e.g. using difunction pulse trains, STEELE computers, phase computers using residue arithmetic
- G06F7/724—Finite field arithmetic
- G06F7/726—Inversion; Reciprocal calculation; Division of elements of a finite field
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/30—Public key, i.e. encryption algorithm being computationally infeasible to invert or user's encryption keys not requiring secrecy
- H04L9/3066—Public key, i.e. encryption algorithm being computationally infeasible to invert or user's encryption keys not requiring secrecy involving algebraic varieties, e.g. elliptic or hyper-elliptic curves
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L2209/00—Additional information or applications relating to cryptographic mechanisms or cryptographic arrangements for secret or secure communication H04L9/00
- H04L2209/80—Wireless
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Computational Mathematics (AREA)
- Algebra (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- General Engineering & Computer Science (AREA)
- Complex Calculations (AREA)
- Error Detection And Correction (AREA)
Abstract
An elliptic curve encryption system represents coordinates of a point on the curve as a vector of binary digits in a normal basis representation in F2m. A key is generated from multiple additions of one or more points in a finite field. Inverses of values are computed using a finite field multiplier and successive exponentiations. A key is represented as the coordinates of a point on the curve and key transfer may be accomplished with the transmission of only one coordinate and identifying information of the second. An encryption protocol using one of the coordinates and a further function of that coordinate is also described.
Description
PUBLIC KEY CRYPTOGRAPHY UTILIZING ELLIPTIC CURVES
FIELD OF THE INVENTION
The present invention relates to public key cryptography.
The increasing use and sophistication of data transmission in such fields as telecommunications, networking, cellular communication, wireless communications, "smart card" applications, audio-visual and video communications has led to an increasing need for systems that permit data encryption, authentication and verification.
It is well known that data can be encrypted by utilizing a pair of keys, one of which is public and one of which is private. The keys are mathematically related such that data encrypted with the public key may only be decrypted with the private key and conversely, data encrypted with the private key can only be-decrypted with the public key. In this way, the public key of a recipient may be made available so that data intended for that recipient may be encrypted with the public key and only decrypted by the recipient's private key, or conversely, encrypted data sent can be verified as authentic when decrypted with the sender's public key.
FIELD OF THE INVENTION
The present invention relates to public key cryptography.
The increasing use and sophistication of data transmission in such fields as telecommunications, networking, cellular communication, wireless communications, "smart card" applications, audio-visual and video communications has led to an increasing need for systems that permit data encryption, authentication and verification.
It is well known that data can be encrypted by utilizing a pair of keys, one of which is public and one of which is private. The keys are mathematically related such that data encrypted with the public key may only be decrypted with the private key and conversely, data encrypted with the private key can only be-decrypted with the public key. In this way, the public key of a recipient may be made available so that data intended for that recipient may be encrypted with the public key and only decrypted by the recipient's private key, or conversely, encrypted data sent can be verified as authentic when decrypted with the sender's public key.
The most well known and accepted public key cryptosystems are those based on integer factorization and discrete logarithms in finite groups. In particular, the RSA system for modulus n = p - q where p and q are primes, the Diffie-Hellman key exchange and the ElGamal protocol in Z , (p a prime) have been implemented worldwide.
The RSA encryption scheme, where two primes p and q are multiplied to provide a modulus n, is based on the integer factorization problem. The public key e and private key d are related such that their product e-d equals 1(mod 4)) where 4 = (p-1) (q-1) . A message M is encrypted by exponentiating it with the private key e to the modulus n, [ C Me (mod n) ] and decrypted by exponentiating with the public key mod n [M = Cd (mod n) ] .
This technique requires the transmission of the modulus n and the public key and the security of the system is based on the difficulty of factoring a large number that has no relatively small factors. Accordingly both p and q must be relatively large primes.
One disadvantage of this system is that p and q must be relatively large (at least 512 bits) to attain an adequate level of security. With the RSA protocol this results in a 1024 bit modulus and a 512 bit public key which require significant bandwidth and storage capabilities. For this reason researchers have looked for public key schemes which reduce the size of the public key. Moreover, recent advances in analytical techniques and associated algorithms have rendered the RSA encryption scheme potentially vulnerable and accordingly raised concerns about the security of such schemes. This implies that larger primes, and therefore a larger modulus, need to be employed in order to'maintain an acceptable level of security. This in turn increases the bandwidth and storage requirements for the implementation of such a scheme.
Since the introduction of the concept of public key cryptography by Diffie and Hellman in 1976, the potential for the use of the discrete logarithm problem in public key cryptosystems has been recognized. In 1985, ElGamal described an explicit methodology for using this problem to implement a fully functional public key cryptosystem, including digital signatures. This methodology has been refined and incorporated with various protocols to meet a variety of applications, and one of its extensions forms the basis for a proposed U.S. digital signature standard (DSS). Although the discrete logarithm problem, as first employed by Diffie and Hellman in their public key exchange algorithm, referred explicitly to the problem of finding logarithms with respect to a primitive element in the multiplicative group of the field of integers modulo a prime p, this idea can be extended to arbitrary groups (with the difficulty of the problem apparently varying with the representation of the group).
The discrete logarithm problem assumes that G is a finite group, and a and b are elements of G. Then the discrete logarithm problem for G is to determine a value x (when it exists) such that a" = b. The value for x is called a logarithm of b to the base of a, and is denoted by logb.
The difficulty of determining this quantity depends on the representation of G. For example, if the abstract cyclic group of order m is represented in the form of the integers modulo m, then the solution to the discrete logarithm problem reduces to the extended Euclidean algorithm, which is relatively easy to solve. However, the problem is -made much more difficult if m+1 is a prime, and the group is represented in the form of the multiplicative group of the finite field Fm+1. This is because the computations must be performed according to the special calculations required for operating in finite fields.
It is also-known that by using computations in a finite field whose members lie on an elliptic curve, that is by defining a group structure G on the solutions of y2+xy=x3+ax2+b over a finite field, the problem is again made much more difficult because of the attributes of elliptic curves. Therefore, it is*possible to attain an increased level of security for a given size of key.
The RSA encryption scheme, where two primes p and q are multiplied to provide a modulus n, is based on the integer factorization problem. The public key e and private key d are related such that their product e-d equals 1(mod 4)) where 4 = (p-1) (q-1) . A message M is encrypted by exponentiating it with the private key e to the modulus n, [ C Me (mod n) ] and decrypted by exponentiating with the public key mod n [M = Cd (mod n) ] .
This technique requires the transmission of the modulus n and the public key and the security of the system is based on the difficulty of factoring a large number that has no relatively small factors. Accordingly both p and q must be relatively large primes.
One disadvantage of this system is that p and q must be relatively large (at least 512 bits) to attain an adequate level of security. With the RSA protocol this results in a 1024 bit modulus and a 512 bit public key which require significant bandwidth and storage capabilities. For this reason researchers have looked for public key schemes which reduce the size of the public key. Moreover, recent advances in analytical techniques and associated algorithms have rendered the RSA encryption scheme potentially vulnerable and accordingly raised concerns about the security of such schemes. This implies that larger primes, and therefore a larger modulus, need to be employed in order to'maintain an acceptable level of security. This in turn increases the bandwidth and storage requirements for the implementation of such a scheme.
Since the introduction of the concept of public key cryptography by Diffie and Hellman in 1976, the potential for the use of the discrete logarithm problem in public key cryptosystems has been recognized. In 1985, ElGamal described an explicit methodology for using this problem to implement a fully functional public key cryptosystem, including digital signatures. This methodology has been refined and incorporated with various protocols to meet a variety of applications, and one of its extensions forms the basis for a proposed U.S. digital signature standard (DSS). Although the discrete logarithm problem, as first employed by Diffie and Hellman in their public key exchange algorithm, referred explicitly to the problem of finding logarithms with respect to a primitive element in the multiplicative group of the field of integers modulo a prime p, this idea can be extended to arbitrary groups (with the difficulty of the problem apparently varying with the representation of the group).
The discrete logarithm problem assumes that G is a finite group, and a and b are elements of G. Then the discrete logarithm problem for G is to determine a value x (when it exists) such that a" = b. The value for x is called a logarithm of b to the base of a, and is denoted by logb.
The difficulty of determining this quantity depends on the representation of G. For example, if the abstract cyclic group of order m is represented in the form of the integers modulo m, then the solution to the discrete logarithm problem reduces to the extended Euclidean algorithm, which is relatively easy to solve. However, the problem is -made much more difficult if m+1 is a prime, and the group is represented in the form of the multiplicative group of the finite field Fm+1. This is because the computations must be performed according to the special calculations required for operating in finite fields.
It is also-known that by using computations in a finite field whose members lie on an elliptic curve, that is by defining a group structure G on the solutions of y2+xy=x3+ax2+b over a finite field, the problem is again made much more difficult because of the attributes of elliptic curves. Therefore, it is*possible to attain an increased level of security for a given size of key.
5 Alternatively a reduced key may be used to maintain a required degree of security.
The inherent security provided by the use of elliptic curves is derived from the characteristic that an addition of two points on the curve can be defined. as a further point that itself lies on the curve. Likewise the result of the addition of a point to itself will result in another point on the curve. Therefore, by selecting a starting point on the curve and multiplying it by an integer, a new point is obtained that lies on the curve. This means that where P = (x,y) is a point on an elliptic curve over a finite field (E(Fgn)), with x and y each represented by a vector of n elements then, for any other point R E< P > (the subgroup generated by P), dP = R. To attack such a scheme, the task is to determine an efficient method to find an integer d, 0 s d s (order of P) - 1 such that dP = R. To break such a scheme, the best algorithms known to date have running times no better than O(vrp), where p is the largest prime dividing the order of the curve (the number of points on the curve).
The inherent security provided by the use of elliptic curves is derived from the characteristic that an addition of two points on the curve can be defined. as a further point that itself lies on the curve. Likewise the result of the addition of a point to itself will result in another point on the curve. Therefore, by selecting a starting point on the curve and multiplying it by an integer, a new point is obtained that lies on the curve. This means that where P = (x,y) is a point on an elliptic curve over a finite field (E(Fgn)), with x and y each represented by a vector of n elements then, for any other point R E< P > (the subgroup generated by P), dP = R. To attack such a scheme, the task is to determine an efficient method to find an integer d, 0 s d s (order of P) - 1 such that dP = R. To break such a scheme, the best algorithms known to date have running times no better than O(vrp), where p is the largest prime dividing the order of the curve (the number of points on the curve).
Thus, in a cryptographic system where the integer d remains secret, the difficulty of determining d can be exploited.
An ElGamal protocol of key exchange based on elliptic curves takes advantage of this characteristic in its definition of private and public keys. Such an ElGamal protocol operates as follows:
1. In order to set up the protocol, where a message is to be sent from A to B, an elliptic curve must be selected and a point P= (x,y), known as the generating point, must be selected.
Encryption 2. The receiver, B, then picks a random integer d as his private key. He then computes dP, which is another point on the curve, which becomes his public key that is made available to the sender and the public. Although the sender knows the value dP, due to the characteristic of elliptic curves noted above, he has great difficulty determining the private key d.
3. The sender A, chooses another random integer k,the session seed, and computes another point on the curve, kP which serves as a public session key.
An ElGamal protocol of key exchange based on elliptic curves takes advantage of this characteristic in its definition of private and public keys. Such an ElGamal protocol operates as follows:
1. In order to set up the protocol, where a message is to be sent from A to B, an elliptic curve must be selected and a point P= (x,y), known as the generating point, must be selected.
Encryption 2. The receiver, B, then picks a random integer d as his private key. He then computes dP, which is another point on the curve, which becomes his public key that is made available to the sender and the public. Although the sender knows the value dP, due to the characteristic of elliptic curves noted above, he has great difficulty determining the private key d.
3. The sender A, chooses another random integer k,the session seed, and computes another point on the curve, kP which serves as a public session key.
This also exploits the characteristic of elliptic curves mentioned above.
4. The sender, A, then retrieves the public key dP of receiver B and computes kdP, another point on the curve, which serves as the shared encryption key for that session.
5. The sender, A, then encrypts the message M with the encryption key to obtain the ciphertext C.
6. The sender then sends the public session key kP and the ciphertext C to the receiver B.
Decryption 7. The receiver, B, determines the encryption key kdP
by multiplying his private key d by kP.
4. The sender, A, then retrieves the public key dP of receiver B and computes kdP, another point on the curve, which serves as the shared encryption key for that session.
5. The sender, A, then encrypts the message M with the encryption key to obtain the ciphertext C.
6. The sender then sends the public session key kP and the ciphertext C to the receiver B.
Decryption 7. The receiver, B, determines the encryption key kdP
by multiplying his private key d by kP.
8. The receiver, B, can then retrieve the message M by decrypting the ciphertext C with the encryption key kdP.
During the entire exchange, the private key d and the seed key k remain secret so that even if an interloper intercepts the session key kP he cannot derive the encryption key kdP from B's public key dP.
Elliptic curve cryptosystems can thus be implemented' employing public and private keys and using the ElGamal protocol.
The elliptic curve cryptography method has a number of benefits. First, each person can define his own elliptic curve for encryption and decryption, which gives rise to increased security. If the private key security is compromised, the elliptic curve can be easily redefined and new public and private keys can be generated to return to a secure system. In addition, to decrypt data encoded with the method, only the parameters for the elliptic curve and the session key need be transmitted.
One of the drawbacks of other public key systems is the large bandwidth and storage requirements for the public keys. The implementation of a public key system using elliptic curves reduces the bandwidth and storage requirements of the public key system because the parameters can be stored in fewer bits. Until now, however, such a scheme was considered impractical due to the computational difficulties involved and the requirement for high speed calculations. The computation of kP, dP and kdP used in a key exchange protocol require complex calculations due to the mathematics involved in adding points in elliptic curve fields.
During the entire exchange, the private key d and the seed key k remain secret so that even if an interloper intercepts the session key kP he cannot derive the encryption key kdP from B's public key dP.
Elliptic curve cryptosystems can thus be implemented' employing public and private keys and using the ElGamal protocol.
The elliptic curve cryptography method has a number of benefits. First, each person can define his own elliptic curve for encryption and decryption, which gives rise to increased security. If the private key security is compromised, the elliptic curve can be easily redefined and new public and private keys can be generated to return to a secure system. In addition, to decrypt data encoded with the method, only the parameters for the elliptic curve and the session key need be transmitted.
One of the drawbacks of other public key systems is the large bandwidth and storage requirements for the public keys. The implementation of a public key system using elliptic curves reduces the bandwidth and storage requirements of the public key system because the parameters can be stored in fewer bits. Until now, however, such a scheme was considered impractical due to the computational difficulties involved and the requirement for high speed calculations. The computation of kP, dP and kdP used in a key exchange protocol require complex calculations due to the mathematics involved in adding points in elliptic curve fields.
Computations on an elliptic curve are performed according to a well known set of relationships. If K
defines any field, then an equation of the form y2 + alxy + a3y = x3 + a2x2 + a4x + a6 , where each of the coefficients ai lie in K, defines an elliptic curve over K. If E is the set of points on this curve, then an abelian group can be defined on the set E U(O), where 0 is a special element not occurring in E. 0 acts as the zero element of the group. If P = (x, y) , then -P = (x, -y) in the case of an odd characteristic, and for,two points P and Q on the curve where Q # P, the sum P + Q is the third point on the curve where the line joining P and Q
again meets the curve. If P = Q, then the tangent line is used. As in any abelian group, we use the notation nP
to denote P added to itself n times if n is positive, and -P added to -itself + n I times if n is negative, and oP=O.
If Fq is a finite field, then elliptic curves over Fq can be divided into two classes, namely supersingular and non-supersingular curves. If F4 is of characteristic 2, i.e. q=21 , then the classes are defined as follows.
i) The set of all solutions to the equation y2 + ay=X3 + bx + c where a, b, c E Fq , a* 0, together with a special point called the point at infinity 0 is a supersingular curve over Fq.
ii) The set of all solutions to the equation 5 y2 + xy = x3. + axe + b where a, b E Fq , b*0 , together with a special point called the point at infinity 0 is a nonsupersingular curve over Fq.
By defining an appropriate addition on these points, 10 we obtain an additive abelian group. The addition of two points P(xl, yl) and Q(x2, y2) for the supersingular elliptic curve E with y2 + ay = x3 + bx + c is given by the following:-If P = (xl, yl) E E; then define -P = (xl, yl + a) , P + 0 =0 + P = P for all PEE.
If Q = (x2, y2) E E and Q # -P, then the point representing the sum of, P + Q, is denoted (x3, y3) , where z x3 = y? xl X2 ( P # Q) ( i z) or x3 = xl b (P=Q) a and y3 = yl xl x3) yl (P # Q) i z or Y3 _ {(xb)(XX) y3. a (P = Q) The addition of two points P(xl, yl) and Q(xz, y2) for the nonsupersingular elliptic curve y2 + xy = x3 + axz + b is given by the following:-If P = (xl, yl) E E then define -P = (x1, yl + xz) For all PEE, 0+P=P+0 =P. If Q= (xz,yz) E E and Q0 -P, then P + Q is a point (x3, y3) , where z x3 C yl yz ED yl yz x1 xz a(P
X:1 (D X2 X, ED X2 or -x3 X1 =z (P = Q) and 1(y, y2 ) (x1 x3) x3 y1 (P # Q) y3 =
xx2 or y3 = Xi (x, (D -V1) X3 X3 (P = Q) Accordingly it can be seen that computing the sum of two points on E requires several multiplications, additions, and inverses in the underlying field Fq. In turn, each of these operations requires-a sequence of elementary bit operations When implementing an ElGamal or Diffie-Hellman scheme with elliptic curves, one is required to compute kP = p + p + ... +P (P added k times) where k is a positive integer and P E E. This requires the computation of (X31Y3) to be computed k-i times. Even if alternative techniques such as "double and add" are utilised, it is still necessary to compute the addition of two points several times, each of which requires multiplications, additions and inverses in the underlying finite field. For large values of k which are typically necessary in cryptographic applications, this has previously been considered impractical for data communication.
defines any field, then an equation of the form y2 + alxy + a3y = x3 + a2x2 + a4x + a6 , where each of the coefficients ai lie in K, defines an elliptic curve over K. If E is the set of points on this curve, then an abelian group can be defined on the set E U(O), where 0 is a special element not occurring in E. 0 acts as the zero element of the group. If P = (x, y) , then -P = (x, -y) in the case of an odd characteristic, and for,two points P and Q on the curve where Q # P, the sum P + Q is the third point on the curve where the line joining P and Q
again meets the curve. If P = Q, then the tangent line is used. As in any abelian group, we use the notation nP
to denote P added to itself n times if n is positive, and -P added to -itself + n I times if n is negative, and oP=O.
If Fq is a finite field, then elliptic curves over Fq can be divided into two classes, namely supersingular and non-supersingular curves. If F4 is of characteristic 2, i.e. q=21 , then the classes are defined as follows.
i) The set of all solutions to the equation y2 + ay=X3 + bx + c where a, b, c E Fq , a* 0, together with a special point called the point at infinity 0 is a supersingular curve over Fq.
ii) The set of all solutions to the equation 5 y2 + xy = x3. + axe + b where a, b E Fq , b*0 , together with a special point called the point at infinity 0 is a nonsupersingular curve over Fq.
By defining an appropriate addition on these points, 10 we obtain an additive abelian group. The addition of two points P(xl, yl) and Q(x2, y2) for the supersingular elliptic curve E with y2 + ay = x3 + bx + c is given by the following:-If P = (xl, yl) E E; then define -P = (xl, yl + a) , P + 0 =0 + P = P for all PEE.
If Q = (x2, y2) E E and Q # -P, then the point representing the sum of, P + Q, is denoted (x3, y3) , where z x3 = y? xl X2 ( P # Q) ( i z) or x3 = xl b (P=Q) a and y3 = yl xl x3) yl (P # Q) i z or Y3 _ {(xb)(XX) y3. a (P = Q) The addition of two points P(xl, yl) and Q(xz, y2) for the nonsupersingular elliptic curve y2 + xy = x3 + axz + b is given by the following:-If P = (xl, yl) E E then define -P = (x1, yl + xz) For all PEE, 0+P=P+0 =P. If Q= (xz,yz) E E and Q0 -P, then P + Q is a point (x3, y3) , where z x3 C yl yz ED yl yz x1 xz a(P
X:1 (D X2 X, ED X2 or -x3 X1 =z (P = Q) and 1(y, y2 ) (x1 x3) x3 y1 (P # Q) y3 =
xx2 or y3 = Xi (x, (D -V1) X3 X3 (P = Q) Accordingly it can be seen that computing the sum of two points on E requires several multiplications, additions, and inverses in the underlying field Fq. In turn, each of these operations requires-a sequence of elementary bit operations When implementing an ElGamal or Diffie-Hellman scheme with elliptic curves, one is required to compute kP = p + p + ... +P (P added k times) where k is a positive integer and P E E. This requires the computation of (X31Y3) to be computed k-i times. Even if alternative techniques such as "double and add" are utilised, it is still necessary to compute the addition of two points several times, each of which requires multiplications, additions and inverses in the underlying finite field. For large values of k which are typically necessary in cryptographic applications, this has previously been considered impractical for data communication.
It is an object of the present invention to provide a method of encryption utilizing elliptic curves that facilitates the computation of additions of points while providing an adequate level of security in an efficient and effective manner.
The applicants have developed a method using a modified version of the Diffie-Hellman and ElGamal protocols defined in the group associated with the points on an elliptic curve over a finite field. The method involves formulating the elliptic curve calculations so as to make elliptic curve cryptography efficient, practical and viable, and preferably employs the use of finite field processor such as the Computational Method and Apparatus for Finite Field Multiplication as disclosed in U.S. Patent 4,745,568. The preferred method exploits the strengths of such a processor with its computational abilities in finite fields. The inventive method structures the elliptic curve calculations as finite field multiplication and exponentiation over the field r2~ In the preferred method, a normal basis representation of the finite field is selected and the calculations which can readily be performed on a finite field processor.
The inventors have recognized that the computations necessary to implement the elliptic curve calculations can be performed efficiently where a finite field of characteristic 2 is chosen.
When computing in a field of characteristic 2, i.e. &2. , squaring is a linear operation, i.e. (A+B)2 is A2 + B2. By adapting appropriate representations, the computation of the squared terms required in the addition of two points is greatly simplified. In particular, if a normal basis representation is chosen, squaring can be achieved through a cyclic shift of the binary vector representing the field element to be squared.
Moreover, computing inverses in elm can be implemented with simple shift and XOR operations by selection of an appropriate representation. In some implementations, the computation of an inverse can be arranged. to utilize multiple squaring operations and thereby improve the efficiency of the computation.
When such computations are performed using a normal basis representation of the finite field, the inventors have also recognized that the elliptic curve calculations are further simplified with the computations presented in this form, the applicants have realized that specialized semiconductor devices can be fabricated to perform the calculations. With the calculations presented in such a form, 'additions in the field F2m can be efficiently performed in one clock cycle utilizing a simple XOR
operation.
Multiplications can be performed very efficiently in 5 only n clock cycles where n is the number of bits being multiplied. Furthermore, squaring can be efficiently performed in 1 clock cycle as a cyclic shift of the bit register. Finally, inverses can easily be computed, requiring approximately log2n multiplications rather than 10 the approximately 2n multiplications required in other arithmetic systems.
The inventors have also recognized that the bandwidth and storage requirements of a cryptographic 15 system utilizing elliptic curves can be significantly reduced where for any point P(x,y) on the curve, only the x coordinate and one bit of the y coordinate need be stored and transmitted, since the one bit will indicate which of the two possible solutions is the second coordinate.
The inventors have also recognized when using the ElGamal protocol that messages need not be points on the curve if the protocol is modified such that the message M
is considered as a pair of field elements M1M2 and each is operated on by the coordinates (x1y) of the session encryption key kdP in a predetermined manner to produce new field elements C1C2 that represent the ciphertext C.
The applicants have developed a method using a modified version of the Diffie-Hellman and ElGamal protocols defined in the group associated with the points on an elliptic curve over a finite field. The method involves formulating the elliptic curve calculations so as to make elliptic curve cryptography efficient, practical and viable, and preferably employs the use of finite field processor such as the Computational Method and Apparatus for Finite Field Multiplication as disclosed in U.S. Patent 4,745,568. The preferred method exploits the strengths of such a processor with its computational abilities in finite fields. The inventive method structures the elliptic curve calculations as finite field multiplication and exponentiation over the field r2~ In the preferred method, a normal basis representation of the finite field is selected and the calculations which can readily be performed on a finite field processor.
The inventors have recognized that the computations necessary to implement the elliptic curve calculations can be performed efficiently where a finite field of characteristic 2 is chosen.
When computing in a field of characteristic 2, i.e. &2. , squaring is a linear operation, i.e. (A+B)2 is A2 + B2. By adapting appropriate representations, the computation of the squared terms required in the addition of two points is greatly simplified. In particular, if a normal basis representation is chosen, squaring can be achieved through a cyclic shift of the binary vector representing the field element to be squared.
Moreover, computing inverses in elm can be implemented with simple shift and XOR operations by selection of an appropriate representation. In some implementations, the computation of an inverse can be arranged. to utilize multiple squaring operations and thereby improve the efficiency of the computation.
When such computations are performed using a normal basis representation of the finite field, the inventors have also recognized that the elliptic curve calculations are further simplified with the computations presented in this form, the applicants have realized that specialized semiconductor devices can be fabricated to perform the calculations. With the calculations presented in such a form, 'additions in the field F2m can be efficiently performed in one clock cycle utilizing a simple XOR
operation.
Multiplications can be performed very efficiently in 5 only n clock cycles where n is the number of bits being multiplied. Furthermore, squaring can be efficiently performed in 1 clock cycle as a cyclic shift of the bit register. Finally, inverses can easily be computed, requiring approximately log2n multiplications rather than 10 the approximately 2n multiplications required in other arithmetic systems.
The inventors have also recognized that the bandwidth and storage requirements of a cryptographic 15 system utilizing elliptic curves can be significantly reduced where for any point P(x,y) on the curve, only the x coordinate and one bit of the y coordinate need be stored and transmitted, since the one bit will indicate which of the two possible solutions is the second coordinate.
The inventors have also recognized when using the ElGamal protocol that messages need not be points on the curve if the protocol is modified such that the message M
is considered as a pair of field elements M1M2 and each is operated on by the coordinates (x1y) of the session encryption key kdP in a predetermined manner to produce new field elements C1C2 that represent the ciphertext C.
The receiver can then extract the message M = (m,., m2) by applying the inverse transformation of the predetermined manner. Although this may require an inverse operation in the field, they may be performed efficiently in the field F2155 , and in particular when operating with the processor noted above.
To assist in the appreciation of the implementation of the present invention , it is believed that a review of the underlying principles of finite field operations is appropriate. The finite field F2 is the number system in which the only elements are the binary numbers 0 and 1 and in which the rules of addition and multiplication are the following:
0+ 0= 1+ 1= 0 0+ 1= 1+ 0= 1 0 x 0= 1 X 0= 0 x 1= 0 1 x 1 = 1 These rules are commonly called modulo-2 arithmetic. All additions specified in logic expressions or by adders in this application are performed modulo-2 as an XOR
operation. Furthermore, multiplication is implemented with logical AND gates.
To assist in the appreciation of the implementation of the present invention , it is believed that a review of the underlying principles of finite field operations is appropriate. The finite field F2 is the number system in which the only elements are the binary numbers 0 and 1 and in which the rules of addition and multiplication are the following:
0+ 0= 1+ 1= 0 0+ 1= 1+ 0= 1 0 x 0= 1 X 0= 0 x 1= 0 1 x 1 = 1 These rules are commonly called modulo-2 arithmetic. All additions specified in logic expressions or by adders in this application are performed modulo-2 as an XOR
operation. Furthermore, multiplication is implemented with logical AND gates.
The finite field F2m, where m is an integer greater than 1, is the number system in which there are 2m elements and inwhich the rules of addition and multiplication correspond to arithmetic modulo an irreducible polynomial of degree m with coefficients in F2. Although in an abstract sense there is for each m only one field. F2m, the complexity of the logic required to perform operations in F2m depends strongly on the particular way in which the field elements are represented. These operations may be performed using processors implemented in either hardware or software with dedicated hardware processors generally considered faster.
The conventional approach to operations performed in F2m is described in such papers as T. Bartee and D.
Schneider, "Computation with Finite Fields", Information and Control, Vol. 6, pp. 79-98, 1963. In this conventional approach, one first chooses a polynomial P(X) of degree m which is irreducible over F2m, that is, P(X) has binary coefficients but cannot be factored into a product of polynomials with binary coefficients each of whose degree is less than'm. An element A in FZm is then defined to be a root of P(X), that is, to satisfy P(A)=O.
The fact that P(X) is irreducible guarantees that the m elements A = 1, A, A2, A'"-1 of F2m are linearly independent over F2.
For the purposes of illustration, the example of F23 will be used with the choice of P (X) = X3 + X + 1 for the irreducible polynomial of degree 3. The next step is to define .A as an element of F23 such that A3 + A + 1 ='O.
The following assignment of unit vectors is then made:
A = 1 = [1,0,0]
Al = [0,1, 0]
A2 = [0, 0, 1]
An arbitrary element B of F23 is now represented.by the binary vector [b2, b, , b ] with the meaning that B = [b2, bl, b ] = b2A2 + b1A + b .
If we represent a second element C = [c2, cl, co] , it follows that B + C = [b2 c2, bl c,, b CO]
Thus, in the conventional approach, addition in F23 is easily performed by logic that merely forms the modulo-2 sum of the two vectors representing the elements to be summed component-by-component. Multiplication is, however, considerably more complex to implement.
The conventional approach to operations performed in F2m is described in such papers as T. Bartee and D.
Schneider, "Computation with Finite Fields", Information and Control, Vol. 6, pp. 79-98, 1963. In this conventional approach, one first chooses a polynomial P(X) of degree m which is irreducible over F2m, that is, P(X) has binary coefficients but cannot be factored into a product of polynomials with binary coefficients each of whose degree is less than'm. An element A in FZm is then defined to be a root of P(X), that is, to satisfy P(A)=O.
The fact that P(X) is irreducible guarantees that the m elements A = 1, A, A2, A'"-1 of F2m are linearly independent over F2.
For the purposes of illustration, the example of F23 will be used with the choice of P (X) = X3 + X + 1 for the irreducible polynomial of degree 3. The next step is to define .A as an element of F23 such that A3 + A + 1 ='O.
The following assignment of unit vectors is then made:
A = 1 = [1,0,0]
Al = [0,1, 0]
A2 = [0, 0, 1]
An arbitrary element B of F23 is now represented.by the binary vector [b2, b, , b ] with the meaning that B = [b2, bl, b ] = b2A2 + b1A + b .
If we represent a second element C = [c2, cl, co] , it follows that B + C = [b2 c2, bl c,, b CO]
Thus, in the conventional approach, addition in F23 is easily performed by logic that merely forms the modulo-2 sum of the two vectors representing the elements to be summed component-by-component. Multiplication is, however, considerably more complex to implement.
continuing the example, from the irreducible polynomial it can be seen that A3 = A + 1 and A4 = A2 + A
where use has been made of the fact that -1 = +1 in F(2).
In hardware, multiplication can be simplified by taking advantage of the special feature of a finite field F2m that there always exists a so-called normal basis for the finite field. That is, one can always find a field element N such that N ,N2, N4 ... N2m-1 are a basis for Fem.
Every field element B can be uniquely written as B = bm-1N2m-1 + . . . + b2N4 + b1N2 + bON = [bm-1 , . . . , b2 , b1 , bO]
where bO, b1, b2, ... bm_1 are binary digits.
For example, in the finite field F23, if we let M= [111,01 Element Field Normal.Basis Representation Normal basis Vector 1010,01 - [o, 0, 0]
[1,0,0] N2 + N4 [1,1,1]
1011101 N+N2 + N4 [0, 1, 11 (010,11 N + N2 [1, 0, 1]
[111,01 N [1'0' 0]
[1,0,1] N+N4 [0,1,0]
[0,1, 1] N [111,01 [111,11 N2 10,0"11 Then, if B = [b.,,_3., ... , b2, b-1, bol and C = [Cm_l, , c2, c1, c01 . are any two elements of Fla in normal basis representation, then the product D = B x C = [dm_1, ... , d2, dl, do] has the property 5 that the same logic circuitry which when applied to the components or binary digits of the vectors representing B and C produces dm-1 will sequentially produce the remaining components dm_2, . , d2, dl, do of the product when applied to the components of the successive 10 shifts of the vectors representing B and C.
As illustrated in U.S. Patent 4,745,568 for Computational Method and Apparatus for Finite"Field Multiplication, multiplication may be implemented by 15 storing bit vectors B and C in respective shift registers and establishing connections to respective accumulating cells such that a grouped term of each of the expressions d., is generated in respective ones of m accumulating cells.
By rotating the bit vectors B and C in the shift 20 registers and by rotating the contents of the accumulating cells, each grouped term of a respective binary digit di is accumulated in successive cells. Thus all of the binary digits of the product vector are generated simultaneously in the accumulating cells after one complete rotation of the bit vectors B and C.
where use has been made of the fact that -1 = +1 in F(2).
In hardware, multiplication can be simplified by taking advantage of the special feature of a finite field F2m that there always exists a so-called normal basis for the finite field. That is, one can always find a field element N such that N ,N2, N4 ... N2m-1 are a basis for Fem.
Every field element B can be uniquely written as B = bm-1N2m-1 + . . . + b2N4 + b1N2 + bON = [bm-1 , . . . , b2 , b1 , bO]
where bO, b1, b2, ... bm_1 are binary digits.
For example, in the finite field F23, if we let M= [111,01 Element Field Normal.Basis Representation Normal basis Vector 1010,01 - [o, 0, 0]
[1,0,0] N2 + N4 [1,1,1]
1011101 N+N2 + N4 [0, 1, 11 (010,11 N + N2 [1, 0, 1]
[111,01 N [1'0' 0]
[1,0,1] N+N4 [0,1,0]
[0,1, 1] N [111,01 [111,11 N2 10,0"11 Then, if B = [b.,,_3., ... , b2, b-1, bol and C = [Cm_l, , c2, c1, c01 . are any two elements of Fla in normal basis representation, then the product D = B x C = [dm_1, ... , d2, dl, do] has the property 5 that the same logic circuitry which when applied to the components or binary digits of the vectors representing B and C produces dm-1 will sequentially produce the remaining components dm_2, . , d2, dl, do of the product when applied to the components of the successive 10 shifts of the vectors representing B and C.
As illustrated in U.S. Patent 4,745,568 for Computational Method and Apparatus for Finite"Field Multiplication, multiplication may be implemented by 15 storing bit vectors B and C in respective shift registers and establishing connections to respective accumulating cells such that a grouped term of each of the expressions d., is generated in respective ones of m accumulating cells.
By rotating the bit vectors B and C in the shift 20 registers and by rotating the contents of the accumulating cells, each grouped term of a respective binary digit di is accumulated in successive cells. Thus all of the binary digits of the product vector are generated simultaneously in the accumulating cells after one complete rotation of the bit vectors B and C.
one attribute of operating such a processor is that in the field F2m, is that squaring is a linear operation in the sense that for every pair of elements B and C in F2m, (B + C) 2 = B2 + C2. It is the case for every element B
of F2m that B2M = B.
In particular in a normal basis representation, squaring an element involves a cyclic shift of the vectors representation of the element, i.e. if B = [bm-a, ... , b2, bl, bo] then B2 = [b,112, ... , b2 i b1, bo, bm-i]
Thus when using the processor exemplified above, squaring may be achieved in one cycle. Moreover, this general characteristic of F2m, where squaring is a linear operation, may be exploited in other implementations, such as software, where a normal basis representation is not used.
As noted above, the inventors have' taken advantage of the efficiency of the mathematical operations in F2m in the implementation of an elliptic curve encryption scheme. The applicants have developed a method of formulating the elliptic curve calculations so as to make elliptic curve cryptography efficient, practical and viable. The preferred method employs the use of a finite field processor such as the Computational Method and Apparatus for Finite Field Multiplication as disclosed in U.S. Patent 4,745,568. The method couples the attractive cryptographic characteristics of elliptic curves with the strengths of the field processor through its computational abilities in finite field Fem. The inventive method structures the elliptic curve calculations as operations, such as multiplication and exponentiation, over the field where F2m, which can readily be calculated on a finite field processor.
An embodiment of the invention will now be described by way of example only with reference to the accompanying drawings in which:-Figure 1 is a diagram of the transmission of an encrypted message from one location to another, Figure 2 is a diagram of an encryption module used with the communication system of Figure 1, Figure 3 is a diagram of a finite field processor used in the encryption and decryption module of Figure 2.
Figure 4 is a flow chart showing movement of the elements through the processor of Figure 3 in computing an inverse function.
of F2m that B2M = B.
In particular in a normal basis representation, squaring an element involves a cyclic shift of the vectors representation of the element, i.e. if B = [bm-a, ... , b2, bl, bo] then B2 = [b,112, ... , b2 i b1, bo, bm-i]
Thus when using the processor exemplified above, squaring may be achieved in one cycle. Moreover, this general characteristic of F2m, where squaring is a linear operation, may be exploited in other implementations, such as software, where a normal basis representation is not used.
As noted above, the inventors have' taken advantage of the efficiency of the mathematical operations in F2m in the implementation of an elliptic curve encryption scheme. The applicants have developed a method of formulating the elliptic curve calculations so as to make elliptic curve cryptography efficient, practical and viable. The preferred method employs the use of a finite field processor such as the Computational Method and Apparatus for Finite Field Multiplication as disclosed in U.S. Patent 4,745,568. The method couples the attractive cryptographic characteristics of elliptic curves with the strengths of the field processor through its computational abilities in finite field Fem. The inventive method structures the elliptic curve calculations as operations, such as multiplication and exponentiation, over the field where F2m, which can readily be calculated on a finite field processor.
An embodiment of the invention will now be described by way of example only with reference to the accompanying drawings in which:-Figure 1 is a diagram of the transmission of an encrypted message from one location to another, Figure 2 is a diagram of an encryption module used with the communication system of Figure 1, Figure 3 is a diagram of a finite field processor used in the encryption and decryption module of Figure 2.
Figure 4 is a flow chart showing movement of the elements through the processor of Figure 3 in computing an inverse function.
Figure 5 is a flow chart showing movement of elements through the processor of Figure 3 to compute the addition of two points.
An embodiment of the invention will first be described utilising an ElGamal key exchange protocol and a Galois field F2s5 to explain the underlying principles.
Further refinements will then be described.
System Components Referring therefore to Figure 1, a message M is to be transferred from a transmitter 10 to a receiver 12 through a communication channel 14. Each of the transmitters 10 and receiver 12 has an encryption/decryption module 16 associated therewith to implement a key exchange protocol and an encryption/decryption algorithm.
The module 16 is shown schematically in Figure 2 and includes an arithmetic unit 20 to perform the computations in the key exchange and generation. A
private key register 22 contains a private key, d, generated as a 155 bit data string from a random number generator 24, and used to generate a public key stored in a public key register 26. A base point register 28 contains the coordinates of a base point P that lies in the elliptic curve selected with each coordinate (x, y), represented as a 155 bit data string. Each of the data strings is a vector of binary digits with each digit being the coefficient of an element of the finite field in the normal basis representation of the coordinate.
The elliptic curve selected will have the general form y2 + xy = x3 + ax2 + b and the parameters of that curve, namely the coefficients a and b are stored in a parameter register 30. The contents of registers 22, 24, 26, 28, 30 may be transferred to the arithmetic unit 20 under control of a CPU 32 as required The contents of the public key register 26 are also available to the communication channel 14 upon a suitable request being received. In the simplest implementation, each encryption module 16 in a common security zone will operate with the same curve and base point so that the contents of registers 28 and 30 need not be accessible.
If further sophistication is required, however, each module 16 may select its own curve and base point in which case the contents of registers 28, 30 have to be accessible to the channel 14.
The module 16 also contains an integer register 34 that receives an integer k, the session seed, from the..
generator 24 for use in encryption and key exchange. The module 16 has a random access memory (RAM) 36 that is used as a temporary store as required during computations.
The encryption of the message M with an encryption 5 key kdP derived from the public key dP and session seed integer k is performed in an encryption unit 40 which implements a selected encryption algorithm. A simple yet effective algorithm is provided by an XOR function which XOR's the message m with the 310 bits of the encryption 10 key kdP. Alternative implementations such as the DES
encryption algorithm could of course be used.
An alternative encryption protocol treats the message m as pairs of coordinates -mõm2, each of 155 bit 15 lengths in the case of FZ,.ss, and XOR's the message ml, m2 with the coordinates of the session key kdP to provide a pair of bit strings (m, xo) (m2 (D yo) . For further security a pair of field elements zlz2 are also formed from the coordinates (xoyo) of kdP.
In one embodiment,- the elements z,z2 are formed from the concatenation of part of xo with part of yo, for example, z, = X0111Y02 and Z2 = xa211Yoi where xo, is the first half of the bit string of xo x02 is the second half of the bit string of xo Yoi is the f irst half of the bit string of yo Y02 is the second half of the bit string of yo The first elements z1 and z2 when treated as field elements are then multiplied with respective bit strings (m1 x0) and (m2 yo) to provide bit strings cl c2 of ciphertext c.
i.e. c1 = z1 (m1 E) x0) c2 = z2 (m2 (D yo) In a preferred implementation of the encryption protocol, a function of x0 is used in place of y0 in the above embodiment. For example the function xa is used as the second 155 bit string so that c1= zl (mlexo ) c2 = z2 (m2 exo ) and z1=X01 IXO32 Z2 =Xo2IX031 where xol is the first half of X03 X362 is the second half of X63 This protocol is also applicable to implementation of elliptic curve encryption in a field other than F2., for example ZP or in general Fpm .
Where ZP is used it may be necessary to adjust the values of x0 and yo or Xo to avoid overfow in the multiplication with z1 and z2. Conventionally this may be done by setting the most significant bit x0 and Fpm or yo to zero.
Key generation, exchange and encryption In order for the transmitter 10 to send the message M to the receiver 12, the receivers public key is retrieved by the transmitter 10. The public key is 10. obtained by the receiver 12 computing the product of the secret key d and base point P in the arithmetic unit 20 as will be described more fully below. The product dP
represents a point on the selected curve and serves as the public key. The public key dP is stored as two 155 bit data strings in the public key register 26.
Upon retrieval of*the public key dP by the transmitter 10, it is stored in the RAM 36. It will be appreciated that even though the base point P is known and publicly available, the attributes of the elliptic curve inhibit the extraction of the secret key d.
The transmitter 10 uses the arithmetic unit 20 to compute the product of the session seed k and the public key dP and stores the result, kdP, in the RAM 36 for use in the encryption algorithm. The result kdP is a further point on the selected curve, again represented by two 155 bit data strings or vectors, and serves as an encryption key.
The transmitter 10 also computes the product of the session seed k with the base point P to provide a new point kP, the session public key, which is stored in the RAM 36.
The transmitter 10 has now the public key dP of the receiver 12, a session public key kP and an encryption key kdP and may use these to send an encrypted message.
The transmitter 10 encrypts the message M with the encryption key kdP in the encryption unit 40 implementing the selected encryption protocols discussed above to provide an encrypted message C. The ciphertext C is transmitted together with the value kP to the encryption module 16 associated with receiver 12.
The receiver 12 utilises the session public key kP
with its private key d to compute the encryption key kdP
in the arithmetic unit 20 and then decrypt the ciphertext C in the encryption unit 40 to retrieve the message M.
During this exchange, the secret key d and the session seed k remain secret and secure. Although P, kP
and dP are known, the encryption key kdP cannot be computed due to the difficulty in obtaining either d or k.
The efficacy of the encryption depends upon the efficient computation of the-values kP, dP and kdP by the arithmetic unit 20. Each computation requires the repetitive addition of two points on the curve which in turn requires the computation of squares and inverses in Operation of the Arithmetic Unit The operation of the arithmetic unit 20 is shown schematically in Figure 3. The unit 20 includes a multiplier 48 having a pair of cyclic shift registers 42, 44 and an accumulating register 46. Each of the registers 42, 44, 46 contain M cells 50a, 50b... 50m, in this example 155, to receive the m elements of a normal basis representation of one of the coordinates of e'.g. X, of P. As fully explained in U.S. Patent No. 4,745,568, the cells 50 of registers 42, 44 are connected to the corresponding cells 50 of accumulating register 46 such a way that a respective grouped term is generated in each cell of register 46. The registers 42,44,46 are also directly interconnected in a bit wise fashion to allow fast transfers of data between the registers.
The movement of data through the registers is controlled by a control register 52 that can execute the instruction set shown in the table below:
INSTRUCTION SET
Operation Size Clock Cycles Field Multiplication 155 bit blocks 156 MULT
10 Calculation of Inverse 24 multiplications approx. 3800 INVERSE
I/O 5-32 bit transfers per 10 10 clock cycles WRITE(A,B or C) read/write to registers 2 clock cycles READ(A,B or C) per transfer Elementary Register 155 bit parallel operation (idle) NOP
Rotate (A,B or C) Copy (A4-B) (A-C) (A4-B) (B4-C ) SWAP (AHB) CLEAR (A,B or C) SET (A,B or C) ADD (A B) ACCUMULATE
The unit 20 includes an adder 54 to receive data from the registers 42,44,46 and RAM 36. The adder 54 is an XOR function and its output is a data stream that may be stored in RAM 36 or one of the registers 42, 44.
Although shown as a serial device, it will be appreciated that it may be implemented as a parallel device to improve computing time. similarly the registers 42,44,46 may be parallel loaded. Each of the registers 42,44,46, is a 155 bit register and is addressed by a 32 bit data bus to allow 32 bit data transfer in 2 clock cycles and the entire loading in 5 operations.
The subroutines used in the computation will now be described.
a) Multiplication The cyclic shift of the elements through the registers 42, 44 m times with a corresponding shift of the accumulating register 46 accumulates successive group terms in respective accumulating cells and a.complete rotation of the elements in the registers 42, 44, produces the elements of the product in the accumulating register 46.
b) Squaring By operating in Fe and adopting a normal basis representation of the field elements, the multiplier 48 may also provide the square of a number by cyclically 25, shifting the elements one cell along the registers 42.
After a one cell shift, the elements in the register represent the square of the number. In general, a number may be raised to the power 2g by cyclically shifting g times through a register.
c) Inversion .
Computation of the inverse of a number can be performed efficiently with the multiplier 48 by implementing an algorithm which utilises multiple squaring operations. The inverse X-1 is represented as X2N-2 or X2(2'"-1-1) If m-1 is considered as the product of two factors g, h then X-1 may be written as X2 (29''-1) or R29h-1 where J3 X2.
The exponent 29h-1 is equivalent to h-1 (2g-1) E 2i 91 i=0 g-1 The term 2g-1 may be written as E 2j j=0 2j) (E 219) so that X-1 r-1 23 1+2+2Z+23.....29_i and is denoted y This term may be computed on multiplier 48 as shown in Figure 4 by initially loading registers 42, with the value X. This is shifted 1 cell to represent /3 (i.e. X2) and the result loaded into both registers 42, 44.
Register 44 is then shifted to provide (32 and the registers 42, 44 multiplied to provide in the accumulating register 46. The multiplication is obtained with one motion, i.e. a m bit cyclic shift, of each of the registers 42, 44, 46..
The accumulated term [31+2 is transferred to register 44 and register 42, which contains N2 is shifted one place to provide j34 The registers 42, 44 are multiplied to provide X31+2+4 This procedure is repeated g-2 times to obtain y.
As will be described below, y can be exponentiated in a similar manner to obtain h-1 Ya 0 i.e X1 .Y 1+29+229+23g...... 2('b-')g This term can be expressed as As noted above, -y can be exponentiated to the 2g by shifting the normal basis representation g times in the register 42, or 44.
Accordingly, the registers 42, 44 are each loaded with the value y and the register 42 shifted g times to provide y29 The registers 42, 44 are multiplied to provide y . yep . or y1+29 in the accumulating register 46. This value is transferred to the register 44 and-the register 42 shifted g times to provide y 22Y
The multiplication will then provide y1+29+229 Repetition of this procedure (h-1)g-1 times produces the inverse of X in the accumulating register 46.
From the above it will be seen that squaring, multiplying, and'inverting can be effectively performed utilising the finite field multiplier 48.
Addition of point P to itself (P + P) using the subroutines To compute the value of dP for generation of the 5 public key, the arithmetic unit 20 associated with the receiver 12 initially computes the addition of P + P. As noted in the introduction, for a nonsupersingular curve the new point Q has coordinates (X3,Y3) where X =Xi b 3 e 10 = X 1 2 ( D 1 To compute X3, the following steps may be implemented as shown in Figure 5.
15 The m bits representing Xl are loaded into register 42 from base point register 28 and shifted one cell to the right to provide X1 This value is stored in RAM 36 and the inverse of X3 computed as described above.
20 The value of x,-2 is loaded into register 44 and the parameter b extracted from the parameter register 30 and loaded into register 42. The product bx12 is computed in the accumulating register 46 by rotating the bit vectors and the resultant value XOR'd in adder 52 with value of Xi stored in RAM 36 to provide the normal basis representation of X3. The result may be stored in RAM 36.
A similar procedure can be followed to generate Y3 by first inverting X1, multiplying the result by Y, and XORing with X1 in the adder 52.. This is then multiplied by X3 stored in RAM 36 and the result XOR'd with the value of X3 and Xi to produce Y3 .
The resultant value of (X3, Y3) represents the sum of P + P and is a new point Q on the curve. This could then be added to P to produce a new point Q'. This process could be repeated d-2 times to generate dP.
The addition of P + Q requires the computation of (X3,Y3) where .20 x = Yl y2 )'ED Yl '2 e x x a .
3 C x1 x2 xl 1 a ED X2 and y3 = 1 (x1 X3) x3 y1 i 2) This would be repeated d-2 times with a new value for Q at each iteration to compute dP.
Whilst in principal this is possible with the arithmetic unit 20, in practice the large numbers used make such a procedure infeasible. A more elegant approach is available using the binary representation of the integer d.
Computation of dP from 2P
To avoid adding dissimilar points P and Q, the binary representation of d is used with a doubling method to reduce the number of additions and the complexity of the additions.
The integer d can be expressed as d = E,%j2i,XjE(0,1) and dP = EA.i(2'P) i.e.
i=o i=o A.m2mP+) m-12m-1P... ?.323P+X222P+JL12P+,1.0P
The values of X are the binary representation of d.
Having computed 2P, the value obtained may be added to itself, as described above at Figure 5 to obtain 22P, which in turn can be added itself to provide 23P etc.
This is repeated until 2`P is obtained.
At each iteration, the value of 21P is retained in RAM 36 for use in subsequent additions to obtain dp.
The arithmetic unit 20 performs a further set of additions for dissimilar points for those terms where X
is 1 to provide the resultant value of the point (x3,y3) representing dP.
If for example k=5, this can be computed as 22P + P
or 2P + 2P + P or Q + Q + P. Therefore the result can be obtained in 3 additions; 2P = Q takes 1 addition, 2P + 2P
Q + Q = R takes 1 and R + P takes 1 addition. At most t doublings and t subsequent additions are required depending on how many X are 1.
Performance of Arithmetic units 20 For computations in a Galois field F2155 it has been found that computing the inverse takes approximately 3800 clock cycles.
The doubling of a point, i.e. the addition of point to itself, takes in the order of 4500 clock cycles and for a practical implementation of a private key, the .25 computation of the public key dP may be computed in the order of 1.5 x 105 clock cycles. With a clock rate typically in the order of 40 mHz, the computation of dP
will take in the order of 3 x 10-2 seconds. This throughput can be enhanced by bounding the seed key k with a Hamming weight of, for example, 20 and thereby limit the number of. additions of dissimilar points.
Computation of session public key kP and encryption key kdP
The session public key kP can similarly be computed with the arithmetic unit 20 of transmitter 10 using the base point P from register 28. Likewise, because the public key dP is represented as a point, (x3,y3), the encryption key kdP can be computed in similar fashion.
Each of these operations will take a similar time and can be completed prior to the transmission.
The recipient 12 is similarly required to compute dkP as he received the ciphertext C which again will take in the order of 3 x 10..2 seconds, well within the time expected for a practical implementation of an encryption unit.
The public key dP, and the session key kP are each represented as a 310 bit data string and as such require a significantly reduced bandwidth for transmission. At the same time, the attributes of elliptic curves provides a secure encryption strategy with a practical implementation due to the efficacy of the arithmetic unit 20.
Curve selection a) The selection of the field F m The above example has utilised a field of 2155 and a non-supersingular curve. The value 155 was chosen in 10 part because an optimal normal basis exists in F2155.
over F2. However, a main consideration is the security and efficiency of the encryption system. The value 155 is large enough to be secure but small enough for efficient operation. A consideration-of conventional =15 attacks that might be used to break the ciphertext suggests that with elliptic curves over F2m , a value of m of.about 130 provides a very secure system.
Using one thousand devices in parallel, the time taken to find one logarithm is about 1.5 x 1011 seconds or at least 20 1500 years using the best known method and the field F2155 Other techniques produce longer run times.
b) Supersingular v. Nonsupersingular Curves 25 A comparison of attacks on data encrypted using elliptic curves suggests that non-supersingular curves are more robust than supersingular curves. For a field Fgk , an attack based on the method, suggested by Menezes, Okamoto and Vanstone.in an article entitled "Reducing elliptic curve logarithms to logarithms in finite field" published in the Proceeding 22 Annual ACM
Symposium Theory Computing 1991, pp. 80-89, (The MOV
attack) shows that for small values of k, the attack becomes subexponential. Most supersingular curves have small values of k associated with them. In general however, non-supersingular curves have large values of k and-provided k>log2q then the MOV attack becomes less efficient than more conventional general attacks.
The use of a supersingular curve is attractive since the doubling of a point (i.e. the case where P = O) does not require any real time inversions in the underlying field. For a supersingular curve, the coordinates of 2P
4 z are x3 = xl b and y3 (xb)(e) y,_ (D a.
a2 a Since a is a constant, a"' and a2 is fixed for a given curve-and can be precomputed. The values of xi and xi can be computed with a single and double cyclic shift respectively on the multiplier 48. However, the subsequent addition of dissimilar points to provide the value of dP still requires the computation of an z inverse as x3- ~ Y2 x1 X2 and (xY1 1 2 ) a y3 = YX YZ (x2 x3) ED -v:, 1 X2Accordingly, although supersingular curves lead to efficient implementations, there is a relatively small set of supersingular curves from which to choose, particularly if the encryption is to be robust. For a supersingular curve where m is odd, there are 3 classes of curve that can be considered further, namely y2 + y = x3 y2 + y = x3 + x y2+y = x3.+x+ 1 However, a consideration of these curves for the case where m = 155 shows that none provide the necessary robustness from attack.
Enhanced security for supersingular curves can be obtained. by employing quadratic extensions of the underlying field. In fact, in Fq where q = 2310 , i.e. a quadratic extension of r'2,ss , amongst the supersingular curves, there are four which under the MOV
attack require computation of discrete logs in 1"2930 These curves provide the requisite high security and also exhibit a high throughput. Similarly, in other extensions of subfields of '2155 (e.g. l ~ ) other curves exist that exhibit the requisite robustness.
However, their use increases the digits that define a point and hence the bandwidth when they are transmitted.
By contrast, the number of.nonsupersingular curves of Fq,q = 2155, is 2(2155 - 1) . By selecting q = 2 i.e. a field i , the value of a in the representation of the curve, y2 + xy- = x3 + ax2 + b, can be chosen to be either 1 or 0 without loss of generality. This large choice of curves permits large numbers of curves over this field to be found for which the order of a curve is divisible by a large prime factor. In general, determining the order of an arbitrary nonsupersingular curve over Fq is not trivial and one approach is explained further in a paper entitled "Counting Points'on Elliptic Curves" by Menezes, Vanstone and Zuccherato, Mathematics of computation 1992.
In general however, the selection of suitable curves is well known in the art, as exemplified in "Application of Finite Fields", chapters 7 and 8, by Menezes, Blake et al, Kluwer Academic Publishers (ISBN 0-7923-9282-5).
Because of the large numbers of such curves that meet the requirements, the use of nonsupersingular curves is preferred despite the added computations.
An alternative approach that reduces the number of inversions when using nonsupersingular curves is to employ homogeneous coordinates. A point P is defined by the coordinates (x,y,z,) and Q by the point (x2,y2-x2) The point (0 , 1 , 0) represents the identity 0 in E.
To derive the addition formulas for the elliptic curve with this representation, we take points P = (x1, y1, z1) and Q = (x2, y2, z2) -, normalize each to (x1/z1, Y1/z1, 1) , (x,/z2, y2/z2, 1) , and apply the previous addition formulas. If P = (X,, Y" z1) , Q = (x2, y2, z2) , P, Q # O , and P A =Q then P + Q = (x3, y3, z3) where if P # Q, then X3 = AD
y3 = CD +A 2 (Bx1 + Ay1) Z3 = A3.Z1Z2 where A=x2z1 +x122 , B=y2z1+y1z2 , C=A+B and D = A2 (A + az1z2) + z1z2BC.
In the case of P = Q, then x3 = AB
y3 = x14A + B (xi + y1z1 + A) Z3 = A3 where A = xlz, and B = .bzi + xi .
It will be noted that the computation of x3 y3 and z3 does not require any inversion. However, to derive the 5 coordinates x3,y3 in a nonhomogeneous representation, it is necessary to normalize the representation so that X3 y3 This operation requires an inversion that utilizes 10 the procedure noted above. However, only one inversion operation is required for the computation of dP.
Using homogeneous coordinates, it is still possible to compute dP using the version of the double and add 15 method described above. The computing action of P + Q , P 0 Q, requires 13 field multiplications, and 2P
requires-7 multiplications.
Alternative Key Transfer In the example above, the coordinates of the keys kP
kdP are each transferred as two 155 bit field elements for F2iss To reduce.the bandwidth further it is possible to transmit only one of the co-ordinates and compute the other coordinate at the receiver. An identifier, for example a single bit of the correct value of the other coordinate, may also be transmitted. This permits the possibilities for the second coordinate to be computed by the recipient and the correct one identified from the identifier.
Referring therefore to Figure 1, the transmitter 10 initially retrieves as the public key dP of the receiver 12, a bit string representing the coordinate x0 and a single bit of the coordinate yo.
The transmitter 10 has the parameters of the curve in register 30 and therefore may use the coordinate xo and-the curve parameters to obtain possible values of the other coordinate yo from the arithmetic unit 20.
For a- curve of the form y2 + xy = x3 + ax2 + b and a coordinate x0, then the possible values y1,y2 for yo are the roots of the quadratic y2 + xoy = X0 + ax0 + b.
By solving for y, in the arithmetic unit 20 two possible roots will be obtained and comparison with the transmitted bit of information will indicate which of the values is the-appropriate value of y.
The two possible values of the second coordinate (yo) differ by x0, i.e. y1 = y2+x0.
Since the two values of yo differ by x0, then yl and y2 will always differ where a "1" occurs in the representation of x0. Accordingly the additional bit transmitted is selected from one of those positions and examination of the corresponding bit of values of y0, will indicate which of the two roots is the appropriate value.
.The receiver 10 thus can generate the coordinates of the public key dP'even though only 156 bits are retrieved.
Similar efficiencies may be'realized in transmitting the session key kP to the receiver 12 as the transmitter 10 need only forward one coordinate, x0 and the selected identifying bit of yo. The receiver 12 may then reconstruct the possible values of yo and select the appropriate one.
In the field rZm it is not possible to solve for y using the quadratic formula as 2a = 0. Accordingly, other techniques need to be utilised and the arithmetic unit 20 is particularly adapted to perform this efficiently.
In general provided x0 is not zero, if y = x0z then x02z2 + xo z = x30 + axx,2 + b.
This may be written as z2 + z = xo + a + b = c_ xo i.e. z2 + z = c.
If in is odd then either z = c+ca+C16 + +C211 C20+C22+C2`+ . +C29+C2m-1 or z = 1+C2........+C2m1 to provide two possible values for yo.
A similar solution exists for the case where m is even that also utilises terms of the form C
This is particularly suitable for use with a normal basis representation in F2s As noted above, raising a field element in Fla to a power g can be achieved by a g fold cyclic shift where the field element is represented as a normal basis.
Accordingly, each value of z can be computed by shifting and adding and the values of yo obtained. The correct one of the values is determined by the additional bit transmitted.
The use of a normal basis representation in F2m therefore simplifies the protocol used to recover the coordinate yo.
If P = (x0 yo) is a point on the elliptic curve E : y2 + xy = x3 + ax2 + b defined over a field '.PZm , then yo is defined to be 0 if xa = 0; if xa ;,-' 0 then 2a is defined to be the least significant bit of the field element yo = xo i-The x-coordinate xo of P and the bit ya are transmitted between the transmitter 10 and receiver 12.
Then the y coordinate yo can be recovered as follows.
1. If xo = 0 then yo is obtained by cyclically shifting the vector representation of the field elementb that is stored in parameter register 30 one position to the left. That is, if b=bm-1bm-2 ... b1bo then Yo =bm-2 ... blb0bm-1 2. If xa ?d 0 then do the following:
2.1 Compute the field element c = xo + a + bxp2 in F.
2.2 Let the vector representation of c be C = Cm-1 Cm_2=. . C1CO.
2. 3 Construct a field element z = zm_1zm_2. .. zlzo by setting zo ya, z1 = co (D za, z2 = C1 Z1, Zm_2 = Cm-3 Zm_3 , zm-i ` CM-2 Zm_2.
2.4 Finally, compute yo = xo Z.
It will be noted that the computation of x0 can be 5 readily computed in the arithmetic unit 20 as described above and that the computation of yo can be obtained from the multiplier 48.
In the above examples, the identification of the 10 appropriate value of yo has been obtained by transmission of a single bit and a comparison of the values of the roots obtained. However, other indicators may be used to identify the appropriate one of the values and the operation is not restricted to encryption with elliptic 15 curves in the field GF(2m) . For example, if the field is selected as Zp p = 3(mod 4) then the Legendre symbol associated with the appropriate value could be -transmitted to designate the appropriate value.
Alternatively, the set of elements in Zp could be 20 subdivided into a pair of subsets with the property that if y is in one subset, then -y is in the other, provided y;;'-O. An arbitrary value can then be assigned to respective subsets and transmitted with the coordinate xo to indicate in which subset the appropriate value of yo is 25 located. Accordingly, the appropriate value of yo can be determined. Conveniently, it is possible to take an appropriate representation in which the subsets are arranged as intervals to facilitate the identification of the appropriate value of yo.
These techniques are particularly suitable for encryption utilizing elliptic curves but may also be used with any algebraic curves and have applications in other fields such as error correcting coding where coordinates of points on curves have to be transferred.
It will be seen therefore that by utilising an elliptic curve lying in the finite field GF2m and utilising a normal basis representation, the computations necessary for encryption with elliptic curves may be efficiently performed. Such operations may be implemented in either software or hardware and the structuring of the computations makes the use of a finite field multiplier implemented in hardware particularly efficient.
An embodiment of the invention will first be described utilising an ElGamal key exchange protocol and a Galois field F2s5 to explain the underlying principles.
Further refinements will then be described.
System Components Referring therefore to Figure 1, a message M is to be transferred from a transmitter 10 to a receiver 12 through a communication channel 14. Each of the transmitters 10 and receiver 12 has an encryption/decryption module 16 associated therewith to implement a key exchange protocol and an encryption/decryption algorithm.
The module 16 is shown schematically in Figure 2 and includes an arithmetic unit 20 to perform the computations in the key exchange and generation. A
private key register 22 contains a private key, d, generated as a 155 bit data string from a random number generator 24, and used to generate a public key stored in a public key register 26. A base point register 28 contains the coordinates of a base point P that lies in the elliptic curve selected with each coordinate (x, y), represented as a 155 bit data string. Each of the data strings is a vector of binary digits with each digit being the coefficient of an element of the finite field in the normal basis representation of the coordinate.
The elliptic curve selected will have the general form y2 + xy = x3 + ax2 + b and the parameters of that curve, namely the coefficients a and b are stored in a parameter register 30. The contents of registers 22, 24, 26, 28, 30 may be transferred to the arithmetic unit 20 under control of a CPU 32 as required The contents of the public key register 26 are also available to the communication channel 14 upon a suitable request being received. In the simplest implementation, each encryption module 16 in a common security zone will operate with the same curve and base point so that the contents of registers 28 and 30 need not be accessible.
If further sophistication is required, however, each module 16 may select its own curve and base point in which case the contents of registers 28, 30 have to be accessible to the channel 14.
The module 16 also contains an integer register 34 that receives an integer k, the session seed, from the..
generator 24 for use in encryption and key exchange. The module 16 has a random access memory (RAM) 36 that is used as a temporary store as required during computations.
The encryption of the message M with an encryption 5 key kdP derived from the public key dP and session seed integer k is performed in an encryption unit 40 which implements a selected encryption algorithm. A simple yet effective algorithm is provided by an XOR function which XOR's the message m with the 310 bits of the encryption 10 key kdP. Alternative implementations such as the DES
encryption algorithm could of course be used.
An alternative encryption protocol treats the message m as pairs of coordinates -mõm2, each of 155 bit 15 lengths in the case of FZ,.ss, and XOR's the message ml, m2 with the coordinates of the session key kdP to provide a pair of bit strings (m, xo) (m2 (D yo) . For further security a pair of field elements zlz2 are also formed from the coordinates (xoyo) of kdP.
In one embodiment,- the elements z,z2 are formed from the concatenation of part of xo with part of yo, for example, z, = X0111Y02 and Z2 = xa211Yoi where xo, is the first half of the bit string of xo x02 is the second half of the bit string of xo Yoi is the f irst half of the bit string of yo Y02 is the second half of the bit string of yo The first elements z1 and z2 when treated as field elements are then multiplied with respective bit strings (m1 x0) and (m2 yo) to provide bit strings cl c2 of ciphertext c.
i.e. c1 = z1 (m1 E) x0) c2 = z2 (m2 (D yo) In a preferred implementation of the encryption protocol, a function of x0 is used in place of y0 in the above embodiment. For example the function xa is used as the second 155 bit string so that c1= zl (mlexo ) c2 = z2 (m2 exo ) and z1=X01 IXO32 Z2 =Xo2IX031 where xol is the first half of X03 X362 is the second half of X63 This protocol is also applicable to implementation of elliptic curve encryption in a field other than F2., for example ZP or in general Fpm .
Where ZP is used it may be necessary to adjust the values of x0 and yo or Xo to avoid overfow in the multiplication with z1 and z2. Conventionally this may be done by setting the most significant bit x0 and Fpm or yo to zero.
Key generation, exchange and encryption In order for the transmitter 10 to send the message M to the receiver 12, the receivers public key is retrieved by the transmitter 10. The public key is 10. obtained by the receiver 12 computing the product of the secret key d and base point P in the arithmetic unit 20 as will be described more fully below. The product dP
represents a point on the selected curve and serves as the public key. The public key dP is stored as two 155 bit data strings in the public key register 26.
Upon retrieval of*the public key dP by the transmitter 10, it is stored in the RAM 36. It will be appreciated that even though the base point P is known and publicly available, the attributes of the elliptic curve inhibit the extraction of the secret key d.
The transmitter 10 uses the arithmetic unit 20 to compute the product of the session seed k and the public key dP and stores the result, kdP, in the RAM 36 for use in the encryption algorithm. The result kdP is a further point on the selected curve, again represented by two 155 bit data strings or vectors, and serves as an encryption key.
The transmitter 10 also computes the product of the session seed k with the base point P to provide a new point kP, the session public key, which is stored in the RAM 36.
The transmitter 10 has now the public key dP of the receiver 12, a session public key kP and an encryption key kdP and may use these to send an encrypted message.
The transmitter 10 encrypts the message M with the encryption key kdP in the encryption unit 40 implementing the selected encryption protocols discussed above to provide an encrypted message C. The ciphertext C is transmitted together with the value kP to the encryption module 16 associated with receiver 12.
The receiver 12 utilises the session public key kP
with its private key d to compute the encryption key kdP
in the arithmetic unit 20 and then decrypt the ciphertext C in the encryption unit 40 to retrieve the message M.
During this exchange, the secret key d and the session seed k remain secret and secure. Although P, kP
and dP are known, the encryption key kdP cannot be computed due to the difficulty in obtaining either d or k.
The efficacy of the encryption depends upon the efficient computation of the-values kP, dP and kdP by the arithmetic unit 20. Each computation requires the repetitive addition of two points on the curve which in turn requires the computation of squares and inverses in Operation of the Arithmetic Unit The operation of the arithmetic unit 20 is shown schematically in Figure 3. The unit 20 includes a multiplier 48 having a pair of cyclic shift registers 42, 44 and an accumulating register 46. Each of the registers 42, 44, 46 contain M cells 50a, 50b... 50m, in this example 155, to receive the m elements of a normal basis representation of one of the coordinates of e'.g. X, of P. As fully explained in U.S. Patent No. 4,745,568, the cells 50 of registers 42, 44 are connected to the corresponding cells 50 of accumulating register 46 such a way that a respective grouped term is generated in each cell of register 46. The registers 42,44,46 are also directly interconnected in a bit wise fashion to allow fast transfers of data between the registers.
The movement of data through the registers is controlled by a control register 52 that can execute the instruction set shown in the table below:
INSTRUCTION SET
Operation Size Clock Cycles Field Multiplication 155 bit blocks 156 MULT
10 Calculation of Inverse 24 multiplications approx. 3800 INVERSE
I/O 5-32 bit transfers per 10 10 clock cycles WRITE(A,B or C) read/write to registers 2 clock cycles READ(A,B or C) per transfer Elementary Register 155 bit parallel operation (idle) NOP
Rotate (A,B or C) Copy (A4-B) (A-C) (A4-B) (B4-C ) SWAP (AHB) CLEAR (A,B or C) SET (A,B or C) ADD (A B) ACCUMULATE
The unit 20 includes an adder 54 to receive data from the registers 42,44,46 and RAM 36. The adder 54 is an XOR function and its output is a data stream that may be stored in RAM 36 or one of the registers 42, 44.
Although shown as a serial device, it will be appreciated that it may be implemented as a parallel device to improve computing time. similarly the registers 42,44,46 may be parallel loaded. Each of the registers 42,44,46, is a 155 bit register and is addressed by a 32 bit data bus to allow 32 bit data transfer in 2 clock cycles and the entire loading in 5 operations.
The subroutines used in the computation will now be described.
a) Multiplication The cyclic shift of the elements through the registers 42, 44 m times with a corresponding shift of the accumulating register 46 accumulates successive group terms in respective accumulating cells and a.complete rotation of the elements in the registers 42, 44, produces the elements of the product in the accumulating register 46.
b) Squaring By operating in Fe and adopting a normal basis representation of the field elements, the multiplier 48 may also provide the square of a number by cyclically 25, shifting the elements one cell along the registers 42.
After a one cell shift, the elements in the register represent the square of the number. In general, a number may be raised to the power 2g by cyclically shifting g times through a register.
c) Inversion .
Computation of the inverse of a number can be performed efficiently with the multiplier 48 by implementing an algorithm which utilises multiple squaring operations. The inverse X-1 is represented as X2N-2 or X2(2'"-1-1) If m-1 is considered as the product of two factors g, h then X-1 may be written as X2 (29''-1) or R29h-1 where J3 X2.
The exponent 29h-1 is equivalent to h-1 (2g-1) E 2i 91 i=0 g-1 The term 2g-1 may be written as E 2j j=0 2j) (E 219) so that X-1 r-1 23 1+2+2Z+23.....29_i and is denoted y This term may be computed on multiplier 48 as shown in Figure 4 by initially loading registers 42, with the value X. This is shifted 1 cell to represent /3 (i.e. X2) and the result loaded into both registers 42, 44.
Register 44 is then shifted to provide (32 and the registers 42, 44 multiplied to provide in the accumulating register 46. The multiplication is obtained with one motion, i.e. a m bit cyclic shift, of each of the registers 42, 44, 46..
The accumulated term [31+2 is transferred to register 44 and register 42, which contains N2 is shifted one place to provide j34 The registers 42, 44 are multiplied to provide X31+2+4 This procedure is repeated g-2 times to obtain y.
As will be described below, y can be exponentiated in a similar manner to obtain h-1 Ya 0 i.e X1 .Y 1+29+229+23g...... 2('b-')g This term can be expressed as As noted above, -y can be exponentiated to the 2g by shifting the normal basis representation g times in the register 42, or 44.
Accordingly, the registers 42, 44 are each loaded with the value y and the register 42 shifted g times to provide y29 The registers 42, 44 are multiplied to provide y . yep . or y1+29 in the accumulating register 46. This value is transferred to the register 44 and-the register 42 shifted g times to provide y 22Y
The multiplication will then provide y1+29+229 Repetition of this procedure (h-1)g-1 times produces the inverse of X in the accumulating register 46.
From the above it will be seen that squaring, multiplying, and'inverting can be effectively performed utilising the finite field multiplier 48.
Addition of point P to itself (P + P) using the subroutines To compute the value of dP for generation of the 5 public key, the arithmetic unit 20 associated with the receiver 12 initially computes the addition of P + P. As noted in the introduction, for a nonsupersingular curve the new point Q has coordinates (X3,Y3) where X =Xi b 3 e 10 = X 1 2 ( D 1 To compute X3, the following steps may be implemented as shown in Figure 5.
15 The m bits representing Xl are loaded into register 42 from base point register 28 and shifted one cell to the right to provide X1 This value is stored in RAM 36 and the inverse of X3 computed as described above.
20 The value of x,-2 is loaded into register 44 and the parameter b extracted from the parameter register 30 and loaded into register 42. The product bx12 is computed in the accumulating register 46 by rotating the bit vectors and the resultant value XOR'd in adder 52 with value of Xi stored in RAM 36 to provide the normal basis representation of X3. The result may be stored in RAM 36.
A similar procedure can be followed to generate Y3 by first inverting X1, multiplying the result by Y, and XORing with X1 in the adder 52.. This is then multiplied by X3 stored in RAM 36 and the result XOR'd with the value of X3 and Xi to produce Y3 .
The resultant value of (X3, Y3) represents the sum of P + P and is a new point Q on the curve. This could then be added to P to produce a new point Q'. This process could be repeated d-2 times to generate dP.
The addition of P + Q requires the computation of (X3,Y3) where .20 x = Yl y2 )'ED Yl '2 e x x a .
3 C x1 x2 xl 1 a ED X2 and y3 = 1 (x1 X3) x3 y1 i 2) This would be repeated d-2 times with a new value for Q at each iteration to compute dP.
Whilst in principal this is possible with the arithmetic unit 20, in practice the large numbers used make such a procedure infeasible. A more elegant approach is available using the binary representation of the integer d.
Computation of dP from 2P
To avoid adding dissimilar points P and Q, the binary representation of d is used with a doubling method to reduce the number of additions and the complexity of the additions.
The integer d can be expressed as d = E,%j2i,XjE(0,1) and dP = EA.i(2'P) i.e.
i=o i=o A.m2mP+) m-12m-1P... ?.323P+X222P+JL12P+,1.0P
The values of X are the binary representation of d.
Having computed 2P, the value obtained may be added to itself, as described above at Figure 5 to obtain 22P, which in turn can be added itself to provide 23P etc.
This is repeated until 2`P is obtained.
At each iteration, the value of 21P is retained in RAM 36 for use in subsequent additions to obtain dp.
The arithmetic unit 20 performs a further set of additions for dissimilar points for those terms where X
is 1 to provide the resultant value of the point (x3,y3) representing dP.
If for example k=5, this can be computed as 22P + P
or 2P + 2P + P or Q + Q + P. Therefore the result can be obtained in 3 additions; 2P = Q takes 1 addition, 2P + 2P
Q + Q = R takes 1 and R + P takes 1 addition. At most t doublings and t subsequent additions are required depending on how many X are 1.
Performance of Arithmetic units 20 For computations in a Galois field F2155 it has been found that computing the inverse takes approximately 3800 clock cycles.
The doubling of a point, i.e. the addition of point to itself, takes in the order of 4500 clock cycles and for a practical implementation of a private key, the .25 computation of the public key dP may be computed in the order of 1.5 x 105 clock cycles. With a clock rate typically in the order of 40 mHz, the computation of dP
will take in the order of 3 x 10-2 seconds. This throughput can be enhanced by bounding the seed key k with a Hamming weight of, for example, 20 and thereby limit the number of. additions of dissimilar points.
Computation of session public key kP and encryption key kdP
The session public key kP can similarly be computed with the arithmetic unit 20 of transmitter 10 using the base point P from register 28. Likewise, because the public key dP is represented as a point, (x3,y3), the encryption key kdP can be computed in similar fashion.
Each of these operations will take a similar time and can be completed prior to the transmission.
The recipient 12 is similarly required to compute dkP as he received the ciphertext C which again will take in the order of 3 x 10..2 seconds, well within the time expected for a practical implementation of an encryption unit.
The public key dP, and the session key kP are each represented as a 310 bit data string and as such require a significantly reduced bandwidth for transmission. At the same time, the attributes of elliptic curves provides a secure encryption strategy with a practical implementation due to the efficacy of the arithmetic unit 20.
Curve selection a) The selection of the field F m The above example has utilised a field of 2155 and a non-supersingular curve. The value 155 was chosen in 10 part because an optimal normal basis exists in F2155.
over F2. However, a main consideration is the security and efficiency of the encryption system. The value 155 is large enough to be secure but small enough for efficient operation. A consideration-of conventional =15 attacks that might be used to break the ciphertext suggests that with elliptic curves over F2m , a value of m of.about 130 provides a very secure system.
Using one thousand devices in parallel, the time taken to find one logarithm is about 1.5 x 1011 seconds or at least 20 1500 years using the best known method and the field F2155 Other techniques produce longer run times.
b) Supersingular v. Nonsupersingular Curves 25 A comparison of attacks on data encrypted using elliptic curves suggests that non-supersingular curves are more robust than supersingular curves. For a field Fgk , an attack based on the method, suggested by Menezes, Okamoto and Vanstone.in an article entitled "Reducing elliptic curve logarithms to logarithms in finite field" published in the Proceeding 22 Annual ACM
Symposium Theory Computing 1991, pp. 80-89, (The MOV
attack) shows that for small values of k, the attack becomes subexponential. Most supersingular curves have small values of k associated with them. In general however, non-supersingular curves have large values of k and-provided k>log2q then the MOV attack becomes less efficient than more conventional general attacks.
The use of a supersingular curve is attractive since the doubling of a point (i.e. the case where P = O) does not require any real time inversions in the underlying field. For a supersingular curve, the coordinates of 2P
4 z are x3 = xl b and y3 (xb)(e) y,_ (D a.
a2 a Since a is a constant, a"' and a2 is fixed for a given curve-and can be precomputed. The values of xi and xi can be computed with a single and double cyclic shift respectively on the multiplier 48. However, the subsequent addition of dissimilar points to provide the value of dP still requires the computation of an z inverse as x3- ~ Y2 x1 X2 and (xY1 1 2 ) a y3 = YX YZ (x2 x3) ED -v:, 1 X2Accordingly, although supersingular curves lead to efficient implementations, there is a relatively small set of supersingular curves from which to choose, particularly if the encryption is to be robust. For a supersingular curve where m is odd, there are 3 classes of curve that can be considered further, namely y2 + y = x3 y2 + y = x3 + x y2+y = x3.+x+ 1 However, a consideration of these curves for the case where m = 155 shows that none provide the necessary robustness from attack.
Enhanced security for supersingular curves can be obtained. by employing quadratic extensions of the underlying field. In fact, in Fq where q = 2310 , i.e. a quadratic extension of r'2,ss , amongst the supersingular curves, there are four which under the MOV
attack require computation of discrete logs in 1"2930 These curves provide the requisite high security and also exhibit a high throughput. Similarly, in other extensions of subfields of '2155 (e.g. l ~ ) other curves exist that exhibit the requisite robustness.
However, their use increases the digits that define a point and hence the bandwidth when they are transmitted.
By contrast, the number of.nonsupersingular curves of Fq,q = 2155, is 2(2155 - 1) . By selecting q = 2 i.e. a field i , the value of a in the representation of the curve, y2 + xy- = x3 + ax2 + b, can be chosen to be either 1 or 0 without loss of generality. This large choice of curves permits large numbers of curves over this field to be found for which the order of a curve is divisible by a large prime factor. In general, determining the order of an arbitrary nonsupersingular curve over Fq is not trivial and one approach is explained further in a paper entitled "Counting Points'on Elliptic Curves" by Menezes, Vanstone and Zuccherato, Mathematics of computation 1992.
In general however, the selection of suitable curves is well known in the art, as exemplified in "Application of Finite Fields", chapters 7 and 8, by Menezes, Blake et al, Kluwer Academic Publishers (ISBN 0-7923-9282-5).
Because of the large numbers of such curves that meet the requirements, the use of nonsupersingular curves is preferred despite the added computations.
An alternative approach that reduces the number of inversions when using nonsupersingular curves is to employ homogeneous coordinates. A point P is defined by the coordinates (x,y,z,) and Q by the point (x2,y2-x2) The point (0 , 1 , 0) represents the identity 0 in E.
To derive the addition formulas for the elliptic curve with this representation, we take points P = (x1, y1, z1) and Q = (x2, y2, z2) -, normalize each to (x1/z1, Y1/z1, 1) , (x,/z2, y2/z2, 1) , and apply the previous addition formulas. If P = (X,, Y" z1) , Q = (x2, y2, z2) , P, Q # O , and P A =Q then P + Q = (x3, y3, z3) where if P # Q, then X3 = AD
y3 = CD +A 2 (Bx1 + Ay1) Z3 = A3.Z1Z2 where A=x2z1 +x122 , B=y2z1+y1z2 , C=A+B and D = A2 (A + az1z2) + z1z2BC.
In the case of P = Q, then x3 = AB
y3 = x14A + B (xi + y1z1 + A) Z3 = A3 where A = xlz, and B = .bzi + xi .
It will be noted that the computation of x3 y3 and z3 does not require any inversion. However, to derive the 5 coordinates x3,y3 in a nonhomogeneous representation, it is necessary to normalize the representation so that X3 y3 This operation requires an inversion that utilizes 10 the procedure noted above. However, only one inversion operation is required for the computation of dP.
Using homogeneous coordinates, it is still possible to compute dP using the version of the double and add 15 method described above. The computing action of P + Q , P 0 Q, requires 13 field multiplications, and 2P
requires-7 multiplications.
Alternative Key Transfer In the example above, the coordinates of the keys kP
kdP are each transferred as two 155 bit field elements for F2iss To reduce.the bandwidth further it is possible to transmit only one of the co-ordinates and compute the other coordinate at the receiver. An identifier, for example a single bit of the correct value of the other coordinate, may also be transmitted. This permits the possibilities for the second coordinate to be computed by the recipient and the correct one identified from the identifier.
Referring therefore to Figure 1, the transmitter 10 initially retrieves as the public key dP of the receiver 12, a bit string representing the coordinate x0 and a single bit of the coordinate yo.
The transmitter 10 has the parameters of the curve in register 30 and therefore may use the coordinate xo and-the curve parameters to obtain possible values of the other coordinate yo from the arithmetic unit 20.
For a- curve of the form y2 + xy = x3 + ax2 + b and a coordinate x0, then the possible values y1,y2 for yo are the roots of the quadratic y2 + xoy = X0 + ax0 + b.
By solving for y, in the arithmetic unit 20 two possible roots will be obtained and comparison with the transmitted bit of information will indicate which of the values is the-appropriate value of y.
The two possible values of the second coordinate (yo) differ by x0, i.e. y1 = y2+x0.
Since the two values of yo differ by x0, then yl and y2 will always differ where a "1" occurs in the representation of x0. Accordingly the additional bit transmitted is selected from one of those positions and examination of the corresponding bit of values of y0, will indicate which of the two roots is the appropriate value.
.The receiver 10 thus can generate the coordinates of the public key dP'even though only 156 bits are retrieved.
Similar efficiencies may be'realized in transmitting the session key kP to the receiver 12 as the transmitter 10 need only forward one coordinate, x0 and the selected identifying bit of yo. The receiver 12 may then reconstruct the possible values of yo and select the appropriate one.
In the field rZm it is not possible to solve for y using the quadratic formula as 2a = 0. Accordingly, other techniques need to be utilised and the arithmetic unit 20 is particularly adapted to perform this efficiently.
In general provided x0 is not zero, if y = x0z then x02z2 + xo z = x30 + axx,2 + b.
This may be written as z2 + z = xo + a + b = c_ xo i.e. z2 + z = c.
If in is odd then either z = c+ca+C16 + +C211 C20+C22+C2`+ . +C29+C2m-1 or z = 1+C2........+C2m1 to provide two possible values for yo.
A similar solution exists for the case where m is even that also utilises terms of the form C
This is particularly suitable for use with a normal basis representation in F2s As noted above, raising a field element in Fla to a power g can be achieved by a g fold cyclic shift where the field element is represented as a normal basis.
Accordingly, each value of z can be computed by shifting and adding and the values of yo obtained. The correct one of the values is determined by the additional bit transmitted.
The use of a normal basis representation in F2m therefore simplifies the protocol used to recover the coordinate yo.
If P = (x0 yo) is a point on the elliptic curve E : y2 + xy = x3 + ax2 + b defined over a field '.PZm , then yo is defined to be 0 if xa = 0; if xa ;,-' 0 then 2a is defined to be the least significant bit of the field element yo = xo i-The x-coordinate xo of P and the bit ya are transmitted between the transmitter 10 and receiver 12.
Then the y coordinate yo can be recovered as follows.
1. If xo = 0 then yo is obtained by cyclically shifting the vector representation of the field elementb that is stored in parameter register 30 one position to the left. That is, if b=bm-1bm-2 ... b1bo then Yo =bm-2 ... blb0bm-1 2. If xa ?d 0 then do the following:
2.1 Compute the field element c = xo + a + bxp2 in F.
2.2 Let the vector representation of c be C = Cm-1 Cm_2=. . C1CO.
2. 3 Construct a field element z = zm_1zm_2. .. zlzo by setting zo ya, z1 = co (D za, z2 = C1 Z1, Zm_2 = Cm-3 Zm_3 , zm-i ` CM-2 Zm_2.
2.4 Finally, compute yo = xo Z.
It will be noted that the computation of x0 can be 5 readily computed in the arithmetic unit 20 as described above and that the computation of yo can be obtained from the multiplier 48.
In the above examples, the identification of the 10 appropriate value of yo has been obtained by transmission of a single bit and a comparison of the values of the roots obtained. However, other indicators may be used to identify the appropriate one of the values and the operation is not restricted to encryption with elliptic 15 curves in the field GF(2m) . For example, if the field is selected as Zp p = 3(mod 4) then the Legendre symbol associated with the appropriate value could be -transmitted to designate the appropriate value.
Alternatively, the set of elements in Zp could be 20 subdivided into a pair of subsets with the property that if y is in one subset, then -y is in the other, provided y;;'-O. An arbitrary value can then be assigned to respective subsets and transmitted with the coordinate xo to indicate in which subset the appropriate value of yo is 25 located. Accordingly, the appropriate value of yo can be determined. Conveniently, it is possible to take an appropriate representation in which the subsets are arranged as intervals to facilitate the identification of the appropriate value of yo.
These techniques are particularly suitable for encryption utilizing elliptic curves but may also be used with any algebraic curves and have applications in other fields such as error correcting coding where coordinates of points on curves have to be transferred.
It will be seen therefore that by utilising an elliptic curve lying in the finite field GF2m and utilising a normal basis representation, the computations necessary for encryption with elliptic curves may be efficiently performed. Such operations may be implemented in either software or hardware and the structuring of the computations makes the use of a finite field multiplier implemented in hardware particularly efficient.
Claims (4)
1. A method of computing, in a finite field GF(2m) and having elements A2i(0<=i<m) that constitute a normal basis, an inverse of an element x of said field using a finite field multiplier having a pair of m-celled re-circulating shift registers connected to an m-celled re-circulating accumulating register to generate, in the accumulating register, a grouped term of a normal basis representation of the product of a pair of elements located in respective ones of the m-celled re-circulating shift registers, the method comprising:
a) representing the element x as a vector of binary digits xi, where each xi is a coefficient, of a corresponding A2i, in a normal basis representation of x;
b) loading into each shift register a vector of binary digits representing a normal basis representation of x2;
c) cyclically shifting the binary digits in a first of the shift registers by one cell, to provide in the first of the shift registers, a vector representing x4;
d) rotating the vectors of binary digits in respective shift registers, and conjointly rotating the accumulating register with an m-fold cyclic shift, to generate, in cells of the accumulating register, m-grouped terms representing a vector of the product of x2 and x4;
e) loading the vector of the product of x2 and x4 from the accumulating register into a second of the shift registers, f) repeating operations c), d), and e) g-2 times, where g is a factor of m-1, to provide in the accumulating register a vector .gamma. of binary digits representing a normal basis representation of an exponentiation ;
g) loading the vector .gamma. into each of the shift registers;
h) performing a g-fold cyclic shift of binary digits of the vector .gamma. in one of the shift registers, where g.h=m-1, to provide a vector representing y2e in the one shift register;
i) rotating bit elements in the shift registers and the accumulating register to generate grouped terms of a vector representing the product of y and y2e;
j) loading the vector representing the product of y and y2e from the accumulating register into the other of the shift registers; and k) repeating operations h), i), and j) a total of g(h-1)-1 times, to provide, in the accumulating register, a vector of binary digits of coefficients of a normal basis representation of the inverse of x.
a) representing the element x as a vector of binary digits xi, where each xi is a coefficient, of a corresponding A2i, in a normal basis representation of x;
b) loading into each shift register a vector of binary digits representing a normal basis representation of x2;
c) cyclically shifting the binary digits in a first of the shift registers by one cell, to provide in the first of the shift registers, a vector representing x4;
d) rotating the vectors of binary digits in respective shift registers, and conjointly rotating the accumulating register with an m-fold cyclic shift, to generate, in cells of the accumulating register, m-grouped terms representing a vector of the product of x2 and x4;
e) loading the vector of the product of x2 and x4 from the accumulating register into a second of the shift registers, f) repeating operations c), d), and e) g-2 times, where g is a factor of m-1, to provide in the accumulating register a vector .gamma. of binary digits representing a normal basis representation of an exponentiation ;
g) loading the vector .gamma. into each of the shift registers;
h) performing a g-fold cyclic shift of binary digits of the vector .gamma. in one of the shift registers, where g.h=m-1, to provide a vector representing y2e in the one shift register;
i) rotating bit elements in the shift registers and the accumulating register to generate grouped terms of a vector representing the product of y and y2e;
j) loading the vector representing the product of y and y2e from the accumulating register into the other of the shift registers; and k) repeating operations h), i), and j) a total of g(h-1)-1 times, to provide, in the accumulating register, a vector of binary digits of coefficients of a normal basis representation of the inverse of x.
2. The method according to claim 1, further comprising.
loading the vector of binary digits xi representing x into one of the shift registers;
performing a 1-cell cyclic shift to provide x2; and copying the resultant vector into the other of the shift registers.
loading the vector of binary digits xi representing x into one of the shift registers;
performing a 1-cell cyclic shift to provide x2; and copying the resultant vector into the other of the shift registers.
3. A computer readable medium comprising computer executable instructions for causing one or more computing systems to execute the method as defined claim 1 or claim 2.
4. A computing system comprising a processor and the computer readable medium according to claim 3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CA2711188A CA2711188C (en) | 1995-07-31 | 1995-07-31 | Public key cryptography utilizing elliptic curves |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CA2640641A CA2640641C (en) | 1995-07-31 | 1995-07-31 | Public key cryptography utilizing elliptic curves |
CA2711188A CA2711188C (en) | 1995-07-31 | 1995-07-31 | Public key cryptography utilizing elliptic curves |
Related Parent Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA2640641A Division CA2640641C (en) | 1995-07-31 | 1995-07-31 | Public key cryptography utilizing elliptic curves |
Publications (2)
Publication Number | Publication Date |
---|---|
CA2711188A1 CA2711188A1 (en) | 1997-02-01 |
CA2711188C true CA2711188C (en) | 2011-11-29 |
Family
ID=42751922
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA2711188A Expired - Lifetime CA2711188C (en) | 1995-07-31 | 1995-07-31 | Public key cryptography utilizing elliptic curves |
Country Status (1)
Country | Link |
---|---|
CA (1) | CA2711188C (en) |
-
1995
- 1995-07-31 CA CA2711188A patent/CA2711188C/en not_active Expired - Lifetime
Also Published As
Publication number | Publication date |
---|---|
CA2711188A1 (en) | 1997-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US6141420A (en) | Elliptic curve encryption systems | |
Koc et al. | Montgomery multiplication in GF (2k) | |
US7961873B2 (en) | Password protocols using XZ-elliptic curve cryptography | |
US6782100B1 (en) | Accelerated finite field operations on an elliptic curve | |
US6202076B1 (en) | Scheme for arithmetic operations in finite field and group operations over elliptic curves realizing improved computational speed | |
Harrison et al. | Software implementation of finite fields of characteristic three, for use in pairing-based cryptosystems | |
US7197527B2 (en) | Efficient arithmetic in finite fields of odd characteristic on binary hardware | |
EP1552382B1 (en) | Efficient arithmetic in finite fields of odd characteristic on binary hardware | |
CA2233120C (en) | Accelerated finite field operations on an elliptic curve | |
Stam et al. | Efficient subgroup exponentiation in quadratic and sixth degree extensions | |
US7519644B2 (en) | Finite field serial-serial multiplication/reduction structure and method | |
US7243292B1 (en) | Error correction using finite fields of odd characteristics on binary hardware | |
Kerins et al. | Fully parameterizable elliptic curve cryptography processor over GF (2 m) | |
Bertoni et al. | Computing tate pairing on smartcards | |
KR100457177B1 (en) | Serial-Parallel Multiplier to Multiply Two Elements in the Finite Field | |
JP4875686B2 (en) | Accelerating finite field operations on elliptic curves | |
CA2129203C (en) | Public key cryptography utilizing elliptic curves | |
Chatterjee et al. | Software Implementation of Curve based Cryptography for Constrained Devices | |
CA2711188C (en) | Public key cryptography utilizing elliptic curves | |
CA2640641C (en) | Public key cryptography utilizing elliptic curves | |
JP4598269B2 (en) | Fast finite field operations on elliptic curves | |
Botes et al. | An implementation of an elliptic curve cryptosystem | |
Beauregard | E cient algorithms for implementing elliptic curve public-key schemes | |
Nabi et al. | Implementation and performance analysis of elliptic curve digital signature algorithm | |
Rostovtsev et al. | Elliptic curve point multiplication |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
EEER | Examination request | ||
MKEX | Expiry |
Effective date: 20150731 |