Revisiting Magnetic Tunnel Junctions : From Deterministic to Stochastic Computing

Chun-Yen Chen¹, Yu-Hui Tang¹, Jhen-Yong Hong^2*

¹Department of Physics, National Central University, Taoyuan City 320317, Taiwan
²Department of Physics, Tamkang University, New Taipei City 251301, Taiwan

* Presenter:Jhen-Yong Hong, email:jyhong@mail.tku.edu.tw

Magnetic tunnel junctions (MTJs) have been crucial components in the processing and memory blocks of traditional computing systems for the past two decades. These devices rely on stabilized magnetization and resistance. Recently, innovative approaches such as probabilistic bits and neuromorphic computing have emerged as promising solutions for next-generation memory and logic devices aimed at unconventional computing. Modern computing techniques often require large circuit areas and consume significant energy for applications. In contrast, stochastic computing presents a compelling alternative to conventional binary computing, as it offers lower area costs, reduced power consumption, and greater robustness against noise.

One of the most promising approaches is spintronics, which takes advantage of the intrinsic stochasticity of magnetization dynamics in MTJs. This presentation will provide a brief overview of the development of spintronics, tracing its evolution from giant magnetoresistance (GMR) to tunneling magnetoresistance (TMR) in MTJs. The tunable, intrinsic stochasticity of MTJs and memristors will also be examined.

In the second part, I will focus on the unconventional characteristics, particularly the charge stochasticity in MTJs [1]. This includes an investigation of resistive switching, 1/f noise and random telegraph noise (RTN) associated with charge fluctuations in these junctions [2-4]. The methods to analyze this include bias-dependent noise results using the partition function and theoretical calculations based on the tight-binding model. Meanwhile, the mechanisms that govern the relaxation time of stochastic charge fluctuations in MTJs will also be investigated. Our findings reveal that random telegraph noise is highly sensitive to external voltage perturbations, offering a valuable opportunity to study the stochastic nature of charge transport dynamics. These results can be interpreted through the concept of modulating the energy landscape with external stimuli and highlight the potential of charge stochasticity as a crucial alternative for probabilistic and neuromorphic computing, as well as for in-memory computing, from a practical perspective.

References:
1. Nano Lett. 25, 11776 (2025); Featured on the Journal Cover. “Stochastic Nature of Voltage-Controlled Charge Dynamics in AlO_x Magnetic Tunnel Junctions”
2. Sci. Rep. 14, 13664 (2024); “Bias polarity dependent low‑frequency noise in ultra‑thin AlO_x‑based magnetic tunnel junctions”
3. Electronics 15, 2525 (2021); “Low-Frequency 1-f Noise Characteristics of Ultra-Thin AlO_x-Based Resistive Switching Memory Devices with Magneto-Resistive Responses”
4. Sci. Rep. 11, 6027 (2021); “Electrically programmable magnetoresistance in AlO_x-based magnetic tunnel junctions”

Keywords: Stochastic Computing, Random telegraph noise (RTN), Magnetic tunnel junctions (MTJs), Spintronics