Power and area-efficient design of VCMA-MRAM based full-adder using approximate computing for IoT applications

Abstract

The next generation in computing era will be within the realm of internet of things (IoT). High density, near-zero leakage, and high endurance are some of the important properties of magnetic RAM (MRAM) which makes it attractive for many IoT applications. As in many of these applications, the computational error is tolerable to some extent; therefore, it is reasonable to use approximate computing to have significant gain in terms of area and power. Full-adder is one of the main parts of CPU for basic operations. In this paper, we utilize voltage-controlled magnetic anisotropy spin transfer torque (VCMA-STT) magnetic tunnel junction (MTJ) as a memory element in accurate non-volatile full-adder and present the corresponding writing circuit that improves 7.6x power consumption compared to the state-of-the-art work. We also propose several approximate magnetic full-adders (AMFs) based on STT-assisted precessional VCMA that are very cost-effective. Some proposed AMFs in this work have improved more than 50% of area and 9.5x of power consumption.