Mike Hamburg : 2009 Security Meeting


Thursday, April 16, 2009
Location: Fisher Conference Center, Arrillaga Alumni Center

"Accelerating AES with Vector Permute Instructions"


The block cipher AES has rapidly become the standard symmetric cipher in most applications. Secure implementation of AES is difficult because its table lookups leak information through the CPU cache. We present a technique for implementing AES using the vector permute instructions in processors such as the Motorola PowerPC G4 (our benchmark machine), IBM Cell, IBM Xenon and AMD "Bulldozer". This technique vastly increases the performance of the cipher --- more than doubling it in some of our benchmarks --- while protecting against timing attacks. Unlike implementations based on bit-slicing, our technique achieves good performance even with minimal parallelism or none at all.


Mike Hamburg is a 3rd-year PhD candidate at Stanford University. He studies cryptography with Dan Boneh.