The energy cost of asymmetric cryptography, a vital component of modern secure communications, inhibits its wide spread adoption within the ultra-low energy regimes such as Implantable Medical Devices (IMDs), Wireless Sensor Networks (WSNs), and Radio Frequency Identification tags (RFIDs). In literature, a plethora of hardware and software acceleration techniques exists for improving the performance of asymmetric cryptography. However, very little attention has been focused on the energy

more »

... ncy. Therefore, in this dissertation, I explore the design space thoroughly, evaluating proposed hardware acceleration techniques in terms of energy cost and showing how effective they are at reducing the energy per cryptographic operation. To do so, I estimate the energy consumption for six different hardware/software configurations across five levels of security, including both GF (p) and GF (2 m ) computation. First, we design and evaluate an efficient baseline architecture for pure software-based cryptography, which is centered around a pipelined RISC processor with 256KB of program ROM and 16KB of RAM. Then, we augment our processor design with simple, yet beneficial instruction set extensions for GF (p) computation and evaluate the improvement in terms of energy per cryptographic operation compared to the baseline microarchitecture. While examining the energy breakdown of the system, it became clear that fetching instructions from program memory was contributing significantly to the overall energy consumption. Thus, we implement a parameterizable instruction cache and simulate various configurations. We determine that for our working set, the energy-optimal instruction cache is 4KB, providing a 25% energy improvement over the baseline architecture for a 192-bit key-size. Next, we introduce a reconfigurable GF (p) accelerator to our microarchitecture and meaii