Abstract

As the technology node shrinks to the nanoscale, several challenges such as high power density have become more critical. One of the most effective methods for addressing these problems is utilizing spintronic devices based on magnetic tunnel junction (MTJ). Furthermore, approximate computing is an emerging paradigm for reducing the power consumption and design complexity in some applications, where the computational error is tolerable. In this paper, a novel non-volatile hybrid MTJ/FinFET-based approximate full adder is presented. The proposed design uses the spin Hall effect (SHE) assisted method for writing data on MTJs, which significantly reduces the power consumption and write energy of the MTJ switching as compared to the conventional STT writing method. The proposed design leads to an effective trade-off between efficiency and quality, as it has a simple and energy-efficient structure, while it provides an acceptable quality in applications like approximate image processing. The circuits are simulated using HSPICE with 14 nm FinFET and SHE perpendicular-anisotropy MTJ models. According to the extensive HSPICE and MATLAB simulations, the proposed approach improves power consumption, delay, power-delay product, and energy-delay product on average by 47%, 53%, 75%, and 86%, while achieving relatively better quality metrics as compared to its state-of-the-art counterparts.

中文翻译：

---

具有自旋霍尔辅助功能的节能磁近似全加器，适用于信号处理应用

摘要

随着技术节点缩小到纳米级，诸如高功率密度之类的若干挑战变得越来越关键。解决这些问题的最有效方法之一是利用基于磁隧道结（MTJ）的自旋电子器件。此外，近似计算是一种新兴的范式，用于在某些可以容忍计算误差的应用中降低功耗和设计复杂性。本文提出了一种基于非易失性混合MTJ / FinFET的近似全加器。所提出的设计使用自旋霍尔效应（SHE）辅助方法在MTJ上写入数据，与传统的STT写入方法相比，它大大降低了MTJ开关的功耗和写入能量。拟议的设计导致了效率和质量之间的有效折衷，因为它具有简单且节能的结构，同时在近似图像处理等应用中提供了可接受的质量。使用具有14 nm FinFET的HSPICE和SHE垂直各向异性MTJ模型对电路进行仿真。根据广泛的HSPICE和MATLAB仿真，所提出的方法将功率消耗，延迟，功率延迟乘积和能量延迟乘积平均提高了47％，53％，75％和86％，同时实现了相对较好的质量指标与最先进的同类产品相比