Analyzing and Extending Instruction Sets for Efficient Cryptographic Computation

Embedded systems and IoT devices are designed to perform specific tasks within strict limits on energy consumption and computing power. To ensure secure communication, these devices require cryptography, nowadays including post-quantum algorithms designed to resist emerging threats. However, the complexity of such algorithms often makes their implementation inefficient on constrained platforms. Although hardware optimisations have proven effective in this domain, they typically require deep algorithm-specific knowledge and manual intervention. This work proposes an alternative approach: CIRCO (Custom Instruction RISC-V Code Optimizer). This tool automatically analyses the assembly code of an application to identify patterns of instructions that can be merged into new, custom RISC-V-compliant instructions. Unlike conventional approaches, CIRCO focuses on logic and arithmetic patterns, avoiding changes to memory or control flow instructions. The CIRCO flow allows user interaction, enabling iterative exploration and guided optimisation. The software’s potential has been tested on a real application: Kyber, a post-quantum cryptographic algorithm that has been widely studied for optimisation using the traditional approach. Due to Kyber’s register-limited nature, the performance improvement achieved is modest (around 3 %), but its flexibility allows it to be used on top of an already developed solution. Beyond Kyber, CIRCO can serve as a starting point for exploring optimisations in other applications, including non-cryptographic ones, supporting the user in the process. Its software’s versatility and modular design suggest a promising direction for future research in custom instruction generation.