Bi-Directional Dynamic RAM Power Management (BDRPM) Heap for Ultra-Low-Power IoT Devices

When developing microcontroller-based embedded systems with limited resources, such as IoT devices, using an RTOS that supports the Tickless-idle technique can readily improve energy efficiency. This is because when there are no tasks to execute, the RTOS scheduler automatically cuts power to almost all hardware components, minimizing wasted energy. However, even with this technique, power to the RAM is not cut off. This is because RAM is a device for storing data, and power must be continuously supplied to retain the data. In practice, though, not all data stored in RAM needs to be retained at all times. Therefore, if we can identify RAM regions that do not need to retain data during execution in real-time and cut off power to those regions, we can further improve energy efficiency. In this study, we propose a new dynamic memory management technique called Bi-directional Dynamic RAM Power Management (BDRPM) heap, which automatically cuts off power by identifying RAM regions that do not need to retain data in real-time. The BDRPM heap tracks RAM usage in real-time during execution, distinguishing between allocated and unallocated regions. Allocated regions are further divided into regions that must retain data continuously and regions that only need to retain data at specific times. Then, power is continuously supplied to allocated regions that must retain data continuously, while power is momentarily supplied only at specific times to regions that need to retain data temporarily. Unallocated regions have their power cut off. The BDRPM heap was implemented and evaluated in Ubinos, an RTOS developed by our research team. The evaluation results showed that idle state power consumption decreased by up to 53% compared to using only the Tickless-idle technique.