Security analysis of neural cryptography implementation

Neural cryptography is a recent approach that aims to solve the key exchange problem with non classical computing through neural networks training on the same input patterns. Recently, there is a great interest in the cryptographic community to study the security of implementations of cryptographic protocols. Timing analysis and power analysis are the most known and successful mechanisms to obtain information about the protocols secret parameters without the need to solve a hard problem. Some parts of the circuits needed for the protocol are implemented and synthesized in VHDL. HSPICE simulator is used to measure the power consumption with different inputs. In this paper, the information leakage through the learning process is investigated. It is also shown how this information can be used to reduce the complexity of the genetic attack, a neural cryptography known attack strategy. To overcome these vulnerabilities some solutions are proposed to make the neural key exchange protocol immune against this simple power analysis attack. In addition, Trojan insertion attacks are introduced.