| Popis: |
As the frontier of Artificial Intelligence (AI) expands, embedding AI models into compact devices, particularly in microcontroller units (MCUs), becomes increasingly critical. This study introduces an innovative approach leveraging a dual-core architecture, notably the STM32H747 MCU, to significantly elevate inference efficiency in MCUs beyond the traditional single-core configurations. We detail the adaptation of lightweight AI models to this architecture, emphasizing preprocessing, memory sharing techniques, and the independent operation of dual cores. Our comparative analysis reveals substantial enhancements: data processing speeds in MCUs increase by up to 6 times, and overall processing efficiency improves by up to 30% compared to conventional single-core systems. These advancements not only underscore the dual-core system’s capability to surpass previous benchmarks but also its potential to drive significant performance and speed enhancements for real-time AI applications in embedded systems. This work sets a new precedent for optimizing AI deployment on resource-constrained devices, opening avenues for more sophisticated AI applications in the realm of embedded computing. |
