Abstract
Since the advent of complementary metal oxide semiconductors (CMOS), the number of transistors per die has continued to increase, reaching today several billion transistors. As a result, it has been possible to design and fabricate smart devices able to run at high speed. However, the power consumption of systems-on-chip has significantly increased due to the high density integration and the high leakage power of current CMOS transistors. As a result, the limits of heat dissipation make further improvement in performance difficult. A high level of autonomy for battery-powered devices is a real challenge. To deal with these issues, spin-transfer-torque magnetic random-access memory (STT-MRAM) technology is seen as a promising solution. In addition to its attractive performance features, STT-MRAM can bring nonvolatility to a system to allow full data retention after a complete shutdown while maintaining a fast wake-up time. Considering two 32-bit embedded processors, this letter shows how STT-MRAM can improve energy efficiency and reliability of future embedded systems thanks to normally-off computing and checkpointing/rollback techniques. A detailed analysis is performed to evaluate the cost related to the backup/recovery of the system. Index Terms—Spintronic memory and logic, embedded processor, spin-transfer-torque, magnetic random-access memory.