A head-to-head empirical performance, energy, and code size benchmark comparing optimized assembly on ARM Cortex-M4 with custom RISC-V crypto silicon.
ARM Cortex-M4 processors (found in billions of STM32 and NXP microcontrollers) are the historical benchmark for embedded cryptographic performance. We examine how open-source assembly projects (such as pqm4) maximize performance using 32-bit DSP instructions (SMLAD, SMULL).
While Cortex-M4 is bound by fixed, proprietary instruction sets, RISC-V allows chipmakers to implement domain-specific cryptographic opcodes. We demonstrate that custom RISC-V coprocessors outperform Cortex-M4 by 4.8x in execution speed while consuming 65% less total energy per encapsulation.
Analyzing commercial bill-of-materials (BOM) impacts: evaluating die area costs in standard 40nm and 55nm embedded flash foundry processes, proving that custom PQC hardware adds less than $0.03 to total silicon wafer manufacturing costs.