Method for strengthening the implementation of ECDSA against power analysis

Abstract

A method of inhibiting the disclosure of confidential information through power analysis attacks on processors in cryptographic systems. The method masks a cryptographic operation using a generator G. A secret value, which may be combined with the generator G to form a secret generator is generated. The secret value is divided into a plurality of parts. A random value is generated for association with the plurality of parts. Each of the plurality of parts is combined with the random value to derive a plurality of new values such that the new values when combined are equivalent to the secret value. Each of the new values is used in the cryptographic operation, thereby using the secret generator in place of the generator G in the cryptographic operation. The introduction of randomness facilitates the introduction of noise into algorithms used by cryptographic systems so as to mask the secret value and provide protection against power analysis attacks.

Description

FIELD OF THE INVENTION

This invention relates to a method for minimizing the vulnerability of cryptographic systems to power analysis-type attacks.

BACKGROUND OF THE INVENTION

Cryptographic systems generally owe their security to the fact that a particular piece of information is kept secret. When a cryptographic algorithm is designed, it is usually assumed that a potential attacker has access to only the public values. Without the secret information it is computationally infeasible to break the scheme or the algorithm. Once an attacker is in possession of a piece of secret information they may be able to forge the signature of the victim and also decrypt secret messages intended for the victim. Thus it is of paramount importance to maintain the secrecy and integrity of the secret information in the system. The secret information is generally stored within a secure boundary in the memory space of the cryptographic processor, making it difficult for an attacker to gain direct access to the secret information. Manufacturers incorporate various types of tamper-proof hardware to prevent illicit access to the secret information. In order to decide how much tamper proofing to implement in the cryptographic system, the designers must consider the resources available to a potential attacker and the value of the information being protected. The magnitude of these resources is used to determine how much physical security to place within the device to thwart attackers who attempt to gain direct access to the secure memory. Tamper-proof devices can help prevent an attacker who is unwilling or unable to spend large amounts of time and money from gaining direct access to the secret information in the cryptographic system. Typically, the amount of work that is required to defeat tamper proof hardware exceeds the value of the information being protected.

However, a new class of attacks has been developed on cryptographic systems that are relatively easy and inexpensive to mount in practice since they ignore the tamper-proof hardware. Recent attacks on cryptographic systems have shown that devices with secure memory may leak information that depends on the secret information, for example in the power usage of a processor computing with private information. Such attacks take advantage of information provided by an insecure channel in the device by using the channel in a method not anticipated by its designers, and so render redundant any tamper proofing in the device. Such insecure channels can be the power supply, electromagnetic radiation, or the time taken to perform operations. At particular risk are portable cryptographic tokens, including smart cards, pagers, personal digital assistants, and the like. Smart cards are especially vulnerable since they rely on an external power supply, whose output may be monitored non-intrusively. Access to the power supply is required for proper functioning of the device and so is not usually prevented with tamper-proof hardware.

Further, constrained devices tend not to have large amounts of electromagnetic shielding. Since the device is self-contained and dedicated, the power consumption and electromagnetic radiation of the smart card may be monitored as the various cryptographic algorithms are executed. Thus in a constrained environment, such as a smart card, it may be possible for an attacker to monitor an unsecured channel that leaks secret information. Such monitoring may yield additional information that is intended to be secret which, when exposed, can significantly weaken the security of a cryptographic system.

In response to the existence of such unsecured channels, manufacturers have attempted to minimize the leakage of information from cryptographic devices. However, certain channels leak information due to their physical characteristics and so it is difficult to completely eliminate leakage. A determined attacker may be able to glean information by collecting a very large number of samples and applying sophisticated statistical techniques. In addition, there are severe restrictions on what can be done in hardware on portable cryptographic tokens that are constrained in terms of power consumption and size. As a result, cryptographic tokens are particularly vulnerable to these types of attacks using unsecured channels.

The more recent attacks using the power supply that can be performed on these particularly vulnerable devices are simple power analysis, differential power analysis, higher order differential power analysis, and other related techniques. These technically sophisticated and extremely powerful analysis tools may be used by an attacker to extract secret keys from cryptographic devices. It has been shown that these attacks can be mounted quickly and inexpensively, and may be implemented using readily available hardware.

The amount of time required for these attacks depends on the type of attack and varies somewhat by device. For example it has been shown that simple power analysis (SPA) typically takes a few seconds per card, while differential power analysis (DPA) can take several hours. In order to perform SPA, the attacker usually only needs to monitor one cryptographic operation. To perform DPA, many operations must be observed. In one method used, in order to monitor the operations, a small resistor is connected in series to smart card's power supply and the voltage across the resistor is measured. The current used can be found by a simple computation based on the voltage and the resistance. A plot of current against time is called a power trace and shows the amount of current drawn by the processor during a cryptographic operation. Since cryptographic algorithms tend to perform different operations having different power requirements depending on the value of the secret key, there is a correlation between the value of the secret key and the power consumption of the device.

Laborious but careful analysis of end-to-end power traces can determine the fundamental operation performed by the algorithm based on each bit of a secret key and thus, be analyzed to find the entire secret key, compromising the system. DPA primarily uses statistical analysis and error correction techniques to extract information that may be correlated to secret keys, while the SPA attacks use primarily visual inspection to identify relevant power fluctuations. In SPA, a power trace is analyzed for any discernible features corresponding to bits of the secret key. The amount of power consumed varies depending on the executed microprocessor instructions. For example, in a typical “square-and-multiply” algorithm for exponentiation, a bit 1 in the exponent will cause the program to perform both squaring and multiply operations, while a bit 0 will cause the multiply operation to be skipped. An attacker may be able to read off the bits of a secret exponent by detecting whether the multiply operation is performed at different bit positions.

A DPA attack attempts to detect more subtle features from the power traces and is more difficult to prevent. To launch a DPA attack, a number of digital signatures are generated and the corresponding power traces are collected. The power trace may be regarded as composed of two distinct parts, namely signal and noise. The patterns that correspond to private key operations tend to remain more or less constant throughout all power traces. These patterns may be regarded as the signal. The other parts of the computation, which correspond to changing data, result in differing patterns in each power trace. These patterns can be regarded as the noise. Statistical analysis can be performed on all the power traces to separate the signal from the noise. The secret value is then derived using the identified signal.

Various techniques for preventing these power analysis attacks have been attempted to date. Manufacturers of smart cards and smart card processors have introduced random wait states and address scrambling. Smart card algorithms avoid performing significantly different operations depending on the value of a secret key and also avoid conditional jump instructions. Hardware solutions include providing well-filtered power supplies and physical shielding of processor elements or the addition of noise unrelated to secrets. However, the vulnerabilities to DPA result from transistor and circuit electrical behaviors that propagate to exposed logic gates, microprocessor operation, and ultimately the software implementations. Cryptographic algorithms to date have been designed with the assumption that there is no leakage of secret information, however with the advent of successful power analysis attacks, it is no longer prudent to assume that a cryptographic device which will leak no secret information can be manufactured. Information stored in constrained environments is particularly difficult to protect against leakage through an unsecured channel during cryptographic operations.

Accordingly, there is a need for a system for reducing the risk of a successful power analysis attack and which is particularly applicable to current hardware environments.

SUMMARY OF THE INVENTION

In accordance with this invention, there is provided a method of inhibiting the disclosure of confidential information through power analysis attacks on processors in cryptographic systems. The method of masking a cryptographic operation using a generator G comprises the steps of:

• • a) generating a secret value, which may be combined with the generator G to form a secret generator;
  • b) dividing the secret value into a plurality of parts;
  • c) generating a random value for association with the plurality of parts;
  • d) combining each of the plurality of parts with the random value to derive a plurality of new values such that the new values when combined are equivalent to the secret value; and
  • e) using each of the new values in the cryptographic operation, thereby using the secret generator in place of the generator G in the cryptographic operation.

The introduction of randomness facilitates the introduction of noise into algorithms used by cryptographic systems so as to mask the secret value and provide protection against power analysis attacks.

BRIEF DESCRIPTION OF THE DRAWINGS

An embodiment of the invention will now be described by way of example only with reference to the accompanying drawings in which:

FIG. 1 is a schematic diagram of a constrained device;

FIG. 2 is a schematic representation of steps of a method performed by the device of FIG. 1; and

FIG. 3 is a flow diagram illustrating an embodiment of the invention.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

A mechanism for protection against power analysis attacks on cryptographic systems involves the introduction of random values into existing algorithms employed by cryptographic systems. These random values are intended to introduce noise into the system.

This technique can be applied to a number of cryptographic systems, including encryption algorithms, decryption algorithms, signature schemes, and the like. In the preferred embodiment, the technique is applied to the ECDSA (elliptic curve digital signature algorithm) on a constrained device, typically a smart card, in order to inhibit the leakage of secret information.

In the ECDSA, as described in the ANSI X9.62 standard, the public values are:

• • The domain parameters: An elliptic curve group E generated by a point G, and a finite field F.
  • The signer's long-term public key D (corresponding to a long-term private key d).
  • The signature (r, s).

FIG. 1 shows generally a smart card (10) for use in a cryptographic system. The smart card incorporates a random number generator (RNG) (11), which may be implemented as hardware or software. The card also includes a cryptographic module (CRYPTO) (14), which may be for example a cryptographic co-processor or specialized software routines. The card includes a memory space (13) for storage needed while making computations, and a parameter storage space (17,18,19,21) for storing the parameters G, G′, β₁, β₂ of the system. The card also includes a secure memory space (15,16) for storing its private key d split into two parts d₁ and d₂, and a processor (12) which may be, for example, an arithmetic logic unit, an integrated circuit, or a general purpose processing unit.

In order to generate a digital signature using an elliptic curve, the signer first computes an elliptic curve point K=kG, where k is a random number and G is the generating point of the elliptic curve group. The value k is selected as a per-message secret key and the point K serves as the corresponding per-message public key. The values k and K are also referred to as an ephemeral private key and an ephemeral public key respectively. These values are used to generate a signature (r, s) wherein:K=kG;

r=K_x mod n, where K_x is the x coordinate of K and n is the order of the generating point G; and

s=k⁻¹(e+dr) mod n, where e is the message to be signed.

The ANSI X9.62 standard provides techniques for interpreting the bit strings corresponding to finite field elements as integers in the above calculations. The standard also provides some guidelines on what elliptic curve groups and finite fields can be used.

Several algorithms, using both direct and indirect methods, may be used to compute kG in order to obtain the elliptic curve point K. Algorithms to compute signature components are potentially vulnerable to power analysis attacks since they perform different operations depending on the bits in the secret values. Repeated iterations of the algorithm use the same secret values, and so their power traces are statistically correlated to the secret values.

In order to mask a private key or other secret value to improve resistance to DPA-like attacks, a random value is introduced into the algorithm as shown in FIG. 2. This random value avoids repeated use of a secret value in order to eliminate correlation among the power traces. There will be no signal to differentiate from the background noise since no operation is repeated on subsequent iterations of the algorithm.

In the case of a long-term private key, the private key d is split into two parts d₁ and d₂ such that d=d₁+d₂. As seen in FIG. 2, the card generates its private key d (110), then computes the public key dG (112). The public key is sent to the server (114), which keeps it in a directory for future use. A smart card is initialized with a private key d being split into the values d₁=d (118) and d₂=0 (116) as is illustrated in FIG. 2. The initialization is performed either by embedding the private key at manufacture or by instructing the smart card to generate its own private key. These initial values d₁ and d₂ are stored in the device instead of storing the value for d. Each time a digital signature is generated, a random value Δ is generated using the hardware random number generator 11 and d₁ and d₂ are updated as follows:d₁=d_1(old)+Δ (mod n), and d₂=d_2(old)−Δ (mod n).

The formula for s, one component of the digital signature, then becomes:s=k⁻¹(e+(d₁r+d₂r)) mod n.

When computing the above formula, the quantities d₁, and d₂ are essentially random values because of the random quantity Δ that is introduced after each signature. When comparing subsequent signatures, there is no correlation in the side channels to either the calculation of d,r or d₂r corresponding to the secret key d since the quantities d₁ and d₂ are randomized in each successive signature but only together does the correlation to d emerge and this changes every time. As a result, leakage of the private key d is minimized when computing the component s of the digital signature. However, the component r of the digital signature is also calculated using the private key k and the calculation of r has still in the past been vulnerable to power analysis type attacks. In order to compute r, the signer must compute kG and so information about the value of the secret key k may leak during the repeated group operations.

In order to protect the per-message secret key k during computation of r, the signer modifies the group generator used. In order to mask the value of k, a random value β is introduced and stored for each smart card such that G′=βG where β is a random number generated for each smart card. The point G′ can be used as a secret generating point for each user, thus using the random value β to hide some information about k.

It is recognized that the signer's effective per-message secret key is kβ, corresponding to the public key kβG. The security is thus based on the secrecy of the derived value kβ, which could be computed from k and β, both of which are secret. It is also recognized that the per-message secret key may be regarded as k and the per-message public key as kG′. However, unless the point G′ were shared publicly, knowledge of k alone would not permit the computation of shared keys based on kG′.

During smart card personalization, when the private/public key pair is generated on the smart card, the point G′ is computed. The introduction of β in the calculation of a digital signature means the formula still contains a constant value, making it vulnerable to power analysis type attacks. In order to overcome these attacks, β is split into two parts β₁ and β₂, and those parts are updated by a random value π every time a signature is generated. This process is detailed in FIG. 3.β₁=β_1(old)+π.β₂=β_2(old)−π.

In order to verify signatures produced in this manner, the verifier uses standard ECDSA verification from ANSI X9.62 since the signer's secret key remains unchanged when using this technique.

Thus the formulae for the ECDSA signature scheme in the preferred embodiment are:K=kG′;

r=K_x mod n, where K_x, is the x coordinate of K and n is the order of the point G′; ands=(kβ₁+kβ₂)⁻¹ (e+(d₁r+d₂r)) mod n.

Using these formulae to compute ECDSA signatures reduces the vulnerability of the algorithm to power analysis attacks. It is recognized that similar techniques may be applied to other signatures. For example, ECNR or any other signature form could be used. These techniques may also be used individually, not necessarily in combination. Also, the ECDSA signature equation is not a necessary component of these techniques.

FIG. 3 shows the generation of a digital signature in accordance with the above protocol. First, the signer generates a random private session key k (200), and stores k (210) for future use in the algorithm. The signer updates the values β₁ (224) and β₂ (226) as described above by generating a random π (222) and then computes the public session key r (220). The signer then obtains the input message e or hash thereof (250). The signer then computes the signature s (260). The signer updates the private key parts d₁ (264) and d₂ (266) as described earlier by generating a random Δ (262).

The inverse algorithm used in the generation of the digital signature to compute k⁻¹ is also potentially vulnerable to power analysis attacks since it performs repeated operations on the secret key every time a signature is generated. This vulnerability is reduced in a further embodiment by introducing a random w and computing (kw)⁻¹ instead of W⁻¹. The signing formula works since k⁻¹=w(kw)⁻¹.

Thus the formulae for the ECDSA signature scheme in this embodiment are:K=kG′;

r=K_x mod n, where K_x is the x coordinate of K and n is the order of the point G′; ands=w(kwβ₁+kwβ₂)⁻¹ (e+(d₁r+d₂r)) mod n.

Updating the parts of the private key may occur before or after the generation of the random w.

In a further embodiment, since G′=β₁G+β₂G, the value of kG′ can be computed as (kβ₁)G+(kβ₂)G. In this way, the value of k is masked when computing kG′, even if the value of β is determined. The formula for K then becomes: K=(kβ₁)G+(kβ₂ )G.

Although the invention has been described with reference to certain specific embodiments, various modifications thereof will be apparent to those skilled in the art without departing from the spirit and scope of the invention as outlined in the claims appended hereto. For example, it is not necessary that there be two components combining to make the private key.

Claims

1. A computer-implemented method of masking a cryptographic operation comprising generation of a signature r, s in a public key elliptic curve cryptographic system, said operation utilizing both a long term private key d and a short term private key k, wherein r=Kx mod n with K being a public key derived from the short term private key k and a generating point G, Kx being the representation of the x coordinate of the key K and n the order of the generating point G, and wherein s=1/k (e+dr) with e being a hash of the message to be signed and d being the long term private key, said method comprising the steps of: applying masking values to each of said long term private key and short term private key during at least a portion of said cryptographic operation to produce a result, said result corresponding to that obtained without application of said masking values whereby observation of said private keys is inhibited by said masking values during said cryptographic operation.

2. The computer-implemented method according to claim 1 wherein a first masking value is applied to said long term private key and a second masking value different from said first masking value is applied to said short term private key.

3. The computer-implemented method according to claim 2 wherein one of said masking values is applied to a respective one of said keys to obtain a scalar multiple of said key.

4. The computer-implemented method according to claim 1 wherein one of said masking values is applied to a respective one of said keys to obtain a scalar multiple of said key.

5. The computer-implemented method according to claim 1 wherein said cryptographic operation requires inversion of one of said keys and a respective one of said masking values is applied to said one key to mask said key during inversion.

6. The computer-implemented method according to claim 1 wherein said masking values change after each cryptographic operation.

7. The computer-implemented method according to claim 1 wherein one of said keys is split into component parts and a masking value applied to each of said parts.

8. The computer-implemented method according to claim 1 wherein masking values are applied to each of the private keys K and d in the signature component s.

9. The computer-implemented method according to claim 1 wherein a masking value is applied to the component s to produce a scalar multiple of the short term key k.

10. The computer-implemented method according to claim 8 wherein a first masking value is applied to said short term private key k and a second value different from said first masking value is applied to said long term private key d.

11. The computer-implemented method according to claim 10 wherein said long term private key d is divided into multiple parts and said second masking value applied in a complementary manner to each of said parts, such that, when said parts are combined, the value of said long term private key d is obtained.

12. A cryptographic co-processor for masking a cryptographic operation comprising generation of a signature r, s in a public key elliptic curve cryptographic system, said operation utilizing both a long term private key d and a short term private key k, wherein r=Kx mod n with K being a public key derived from the short term private key k and a generating point G, Kx being the representation of the x coordinate of the key K and n being the order of the generating point G, and wherein s=1/k (e+dr) with e being a hash of the message to be signed and d being the long term private key, said cryptographic module being configured for: applying masking values to each of said long term private key and short term private key during at least a portion of said cryptographic operation to produce a result, said result corresponding to that obtained without application of said masking values whereby observation of said private keys is inhibited by said masking values during said cryptographic operation.

13. The cryptographic co-processor according to claim 12 wherein a first masking value is applied to said long term private key and a second masking value different from said first masking value is applied to said short term private key.

14. The cryptographic co-processor according to claim 13 wherein one of said masking values is applied to a respective one of said keys to obtain a scalar multiple of said key.

15. The cryptographic co-processor according to claim 12 wherein one of said masking values is applied to a respective one of said keys to obtain a scalar multiple of said key.

16. The cryptographic co-processor according to claim 12 wherein said cryptographic operation requires inversion of one of said keys and a respective one of said masking values is applied to said one key to mask said key during inversion.

17. The cryptographic co-processor according to claim 12 wherein said masking values change after each cryptographic operation.

18. The cryptographic co-processor according to claim 12 wherein one of said keys is split into component parts and a masking value applied to each of said parts.

19. The cryptographic co-processor according to claim 12 wherein masking values are applied to each of the private keys K and d in the signature component s.

20. The cryptographic co-processor according to claim 12 wherein a masking value is applied to the component s to produce a scalar multiple of the short term key k.

21. The cryptographic co-processor according to claim 20 wherein a first masking value is applied to said short term private key k and a second value different from said first masking value is applied to said long term private key d.

22. The cryptographic co-processor according to claim 21 wherein said long term private key d is divided into multiple parts and said second masking value applied in a complementary manner to each of said parts, such that, when said parts are combined, the value of said long term private key d is obtained.