With transistors reaching nanometer dimensions, dissipated energy has become of great importance in recent years. A practical approach to reducing energy consumption is to use logic-in-memory (LIM) structures based on magnetic tunnel junction (MTJ) devices combined with approximate computing. In this paper, we propose energy-efficient MTJ/FinFET-based approximate 5:2 compressors for error-resilient in-memory computing, providing an accuracy close to the exact design (1.54% error rate) while reducing the energy consumption by more than 50%. The innovative Boolean equations and the structure of the proposed approximate circuits based on spin-Hall effect assisted MTJs lead to a significantly more effective compromise between energy and accuracy than the previous exact and approximate counterparts. The simulation results provided using HSPICE with 7nm FinFETs and SHE-assisted MTJ models demonstrate the superior hardware parameters of the proposed designs. Furthermore, the MATLAB simulations show an average peak signal-to-noise ratio (PSNR) of more than 43 and an average structural similarity index metric (SSIM) of more than 0.99 across image multiplication, sharpening, and smoothing operations.

中文翻译：

---

用于容错内存计算的高精度自旋电子近似压缩器

近年来，随着晶体管达到纳米尺寸，耗散能量变得非常重要。降低能耗的一种实用方法是使用基于磁隧道结 (MTJ) 器件并结合近似计算的内存中逻辑 (LIM) 结构。在本文中，我们提出了基于 MTJ/FinFET 的节能型近似 5:2 压缩器，用于内存中的容错计算，提供接近精确设计的精度（1.54% 错误率），同时将能耗降低超过50%。创新的布尔方程和基于自旋霍尔效应辅助的 MTJ 所提出的近似电路的结构导致能量和精度之间的折衷比以前的精确和近似对应物显着更有效。使用 HSPICE 提供的仿真结果与 7nm FinFET 和 SHE 辅助的 MTJ 模型展示了所提出设计的优越硬件参数。此外，MATLAB 仿真显示，在图像乘法、锐化和平滑操作中，平均峰值信噪比 (PSNR) 超过 43，平均结构相似性指数度量 (SSIM) 超过 0.99。