Tunable Random Telegraph Noise in Stable Perpendicular Magnetic Tunnel Junctions for Unconventional Computing

Ahmed Sidi El Valli, Michael Tsao, Dairong Chen, Andrew D. Kent · 2025

Read Tunable Random Telegraph Noise in Stable Perpendicular Magnetic Tunnel Junctions for Unconventional Computing by Ahmed Sidi El Valli, Michael Tsao, Dairong Chen, Andrew D. Kent (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-13458.txt.

Introduction

Ahmed Sidi El Valli, Michael Tsao and two co-authors actuate stable magnetic tunnel junctions in 2025 to yield tunable random telegraph noise. Perpendicular junctions anchor spin-torque memory and are thermally stable. Nanosecond pulses can wake stochastic acts without losing stability.

Fluctuation rate and bias tune over wide ranges. Responses fit a Poisson process in tests. Mean dwell times span tens of nanoseconds to microseconds, over two orders of magnitude. One chip can thus host deterministic, stochastic, and in-memory acts together. Probabilistic and neuromorphic systems gain a compact noise source with dials.

Catalog record

CREATORAhmed Sidi El Valli, Michael Tsao, Dairong Chen, Andrew D. Kent
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-13458

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Tunable Random Telegraph Noise in Stable Perpendicular Magnetic Tunnel Junctions for Unconventional Computing Ahmed Sidi El Valli, Michael Tsao, Dairong Chen, Andrew D. Kent Abstract We demonstrate that thermally stable perpendicular magnetic tunnel junctions (pMTJs), widely used in spin-transfer torque magnetic random-access memory, can be actuated with nanosecond pulses to exhibit tunable stochastic behavior. This actuated-stochastic tunnel junction (A-sMTJ) concept produces random telegraph noise, with control over fluctuation rate and probability bias. The device response is shown to be consistent with a Poisson process, with fluctuation rates tunable over more than two orders of magnitude, with average state dwell times varying from 29 29 ns to greater than 2.3 ​ μ 2.3\mu s. These results establish A-sMTJs as a versatile platform for integrating deterministic, stochastic, and in-memory functionality on a single chip—advancing the development of probabilistic, neuromorphic, and unconventional computing systems. I Introduction The search for hardware building blocks that support unconventional computing architectures is a growing challenge at the interface of physics and information processing. Driven by rapid advances in artificial intelligence, there is increasing interest in addressing the memory bottlenecks of conventional von Neumann architectures by developing specialized hardware that exploits physical dynamics and brain-inspired computing models Misra et al. 2023 ; Marković et al. 2020 ; Nikhar et al. 2024 ; Finocchio et al. 2024 . One promising candidate is the perpendicular magnetic tunnel junction (pMTJ)—a device originally designed for fast, dense nonvolatile memory Kent and Worledge 2015 ; Dieny et al. 2016 . More recently, pMTJs have been studied for in-memory computing Jung et al. 2022 ; Sebastian et al. 2020 ; Hirtzlin et al. 2019 , and their intrinsic stochastic switching behavior has enabled applications in probabilistic computing and true random number generation (TRNG) Rehm et al. 2023 ; Dubovskiy et al. 2024 ; Xu et al. 2024 . In this work, we present a complementary approach by demonstrating that both non-volatile memory and superparamagnetic-like behavior can be achieved using pMTJs with stable magnetization free layers. Our results demonstrate that standard pMTJs can exhibit multifunctionality—serving not only as building blocks for conventional spin-transfer torque magnetic random access memory (STT-MRAM) in in-memory computing systems, but also as stochastic elements for probabilistic and unconventional computing architectures. Figure 1: Stochastic magnetic tunnel junctions. (a) In superparamagnetic MTJs (sMTJs) with low energy barriers ( \Delta\lesssim 20 ), thermal fluctuations spontaneously drive switching between the parallel (P, “0”) and antiparallel (AP, “1”) states, producing random telegraph signals in the device resistance. (b) In contrast, in actuated-stochastic magnetic MTJs (a-sMTJs), we suppress spontaneous switching by using pMTJs with larger energy barriers ( \Delta\gtrsim 20 ) and actuate stochastic transitions via nanosecond-scale voltage pulses of alternating polarity. Here spin-transfer torques enable controlled, tunable switching, reproducing telegraph-like behavior through a fundamentally different, externally driven mechanism. II Device Concept Stochastic pMTJ concepts are illustrated in Fig. 1 . The key parameter governing the thermally activated switching behavior is the dimensionless ratio \Delta=E_{B}/(kT) , where E_{B} is the energy barrier to magnetization reversal, k is the Boltzmann constant, and T is the device temperature. As shown in Fig. 1 (a), when \Delta\lesssim 20 , the free layer magnetization undergoes spontaneous fluctuations between the parallel (P, labeled “0”) and antiparallel (AP, labeled “1”) states. The switching rate is governed at least approximately by an Arrhenius law, \Gamma\simeq\Gamma_{0}\exp(-\Delta) , where \Gamma_{0} is the attempt frequency, typically on the order of tens of GHz, but much higher rates have also been found in experiment Bedau et al. 2010a ; Soumah et al. 2025 . These thermal fluctuations generate random telegraph signals in the electrical resistance due to the tunnel magnetoresistance effect and are known as superparamagnetic magnetic tunnel junctions (sMTJs). In contrast, here we demonstrate an actuated MTJ device concept in which stochastic behavior is externally controlled. We use pMTJs with larger energy barriers ( \Delta\gtrsim 20 ) where spontaneous thermally activated switching is suppressed. In this regime, inspired by the device operation model presented in Ref. Sun 2024 , we induce stochastic switching by applying a periodic stream of nanosecond-duration voltage pulses of alternating polarity (hereafter referred to as “AC pulses” for brevity), as illustrated in Fig. 1 (b). These pulses generate spin-transfer torques that modulate the magnetization dynamics, enabling fully controlled and tunable stochastic fluctuations. The switching probability depends on the amplitude and duration of the pulses and, under appropriate conditions, produces a random telegraph signal—functionally analogous to that of a superparamagnetic MTJ (sMTJ). We designate this as an actuated-stochastic magnetic tunnel junction (A-sMTJ), to distinguish it from the thermally driven fluctuations of the sMTJ shown in Fig. 1 (a). III Device Operation The device operating principle is based on the fact that spin-transfer torque switching is probabilistic, with the switching probability depending on the pulse polarity, amplitude and duration. The pulse polarity sets the switching sense, e.g. , in the case of our experiment, positive pulse polarity favors P \rightarrow AP (“0” to “1”) switching and negative pulse polarity AP \rightarrow P (“1” to “0”) switching. For a given switching sense and collinear magnetization reference and free layers the switching probability is generally a monotonically increasing function of the pulse amplitude and duration (see, for example, Refs. Liu et al. 2014 ; Rehm et al. 2019 ). For STT-MRAM applications the pulse parameters are chosen to lead to highly reliable switching with very small write-error-rates (WER) ( <10^{-9} or lower depending on the application), termed a “deterministic” write pulse (see, for example, Ref. Nowak et al. 2011 ). In our application the WER is chosen to be much larger to enable stochastic behavior. We denote these stochastic write pulses, to distinguish them from the write pulses used in STT-MRAM. Figure 2: Voltage pulse sequence. Each cycle consists of a positive stochastic write pulse (S1) that promotes 0\rightarrow 1 transitions, followed by a negative stochastic write pulse (S2) that promotes 1\rightarrow 0 transitions. A subsequent low-amplitude read pulse is applied to detect the MTJ state. Our device operates using two stochastic write pulses of opposite polarity, labeled S1 and S2, followed by a low-amplitude read pulse that does not perturb the MTJ state, as illustrated in Fig. 2 . This approach contrasts with our previous work, which employed a combination of deterministic and stochastic pulses—generating random bits from a well-defined initial state Rehm et al. 2023 ; Rehm et al. 2024 ; Dubovskiy et al. 2024 ; Sidi El Valli et al. 2025 . The write pulses are nanosecond in duration, placing the device operation near the ballistic spin-transfer regime, where the stochasticity arises primarily from thermal variations in the initial magnetization state Bedau et al. 2010b ; Liu et al. 2014 . In this regime, the switching probability depends on the product of pulse amplitude and duration— i.e. , the total charge transmitted through they junction, a consequence of angular momentum conservation. We also note that operation at current amplitudes approximately twice the threshold current has been shown to minimize energy dissipation within the MTJ due to Joule heating Bedau et al. 2010b . To analyze the device operation we consider the switching probabilities associated with pulses S1 and S2. The first pulse favors the “1” state with a switching probability \Gamma_{01} . Whereas, the

Read the complete text