Optimized RISC architecture for multiple-precision modular arithmetic

Public-key cryptosystems normally spend most of their execution time in a small fraction of the program code, typically in an inner loop. The performance of these critical code sections can be significantly improved by customizing the processor's instruction set and microarchitecture, respectively. This paper shows the advantages of instruction set extensions to accelerate the processing of cryptographic workloads such as long integer modular arithmetic. We define two custom instructions for performing multiply-and-add operations on unsigned integers (single-precision words). Both instructions can be efficiently executed by a (32 x 32 + 32 + 32)-bit multiply/accumulate (MAC) unit. Thus, the proposed extensions are simple to integrate into standard 32-bit RISC cores like the MIPS32 4Km. We present an optimized Assembly routine for fast multiple-precision multiplication with "finely" integrated Montgomery reduction (FIGS method). Simulation results demonstrate that the custom instructions double the processor's arithmetic performance compared to a standard MIPS32 core.