Optimizing a Fast Stream Cipher for VLIW, SIMD, and Superscalar Processors

The mismatch between traditional cipher designs and efficient operation on modern Very Long Instruction Word, Single Instruction Multiple Data, superscalar, and deeply pipelined processors is explored. Guidelines are developed for efficiently exploiting the instructionlevel parallelism of these processor architectures. Two stream ciphers, WAKE-ROFB and WiderWake, incorporating these ideas are proposed. WAKE-ROFB inherits the security characteristics of WAKE, from which it is derived, but runs almost three times as fast as WAKE on a commercially available VLIW CPU. Throughput in excess of 40 MByte/s on a 100 MHz processor is demonstrated. Another derivative, WiderWake, whose security characteristics are not directly transferrable from WAKE runs in excess of 50 MByte/s on the same processor.