Tunable multi-bit stochasticity in La 0.67 Sr 0.33 MnO 3 -based probabilistic bits

Ishitro Bhaduri, Azminul Jaman, Walter Quiñonez, Ayush Gupta, Tamalika Banerjee · 2025

Read Tunable multi-bit stochasticity in La 0.67 Sr 0.33 MnO 3 -based probabilistic bits by Ishitro Bhaduri, Azminul Jaman, Walter Quiñonez, Ayush Gupta, Tamalika Banerjee (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-2503-20970.txt.

Introduction

Ishitro Bhaduri and co-authors present a 2025 paper on a new kind of random bit for computing. Some future computers will use units that flip by chance, called p-bits, to solve hard problems with less energy. Most p-bits so far rely on tiny magnets that grow unstable with heat, so other materials are being sought.

The team made p-bits from thin films of a metal oxide that switches between states when a voltage is applied. Uneven patches in the film give two modes of chance behavior: timed two-state flips and free multi-level flips. Voltage tunes both modes. One material that offers both modes could serve varied computing tasks.

Catalog record

CREATORIshitro Bhaduri, Azminul Jaman, Walter Quiñonez, Ayush Gupta, Tamalika Banerjee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-20970

Browse the full library · About the Tower of Babel

Full text

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

Tunable multi-bit stochasticity in La 0.67 Sr 0.33 MnO 3 -based probabilistic bits Preprint: APS/123-QED Ishitro Bhaduri, Azminul Jaman, Walter Quiñonez, Ayush Gupta, Tamalika Banerjee Abstract One promising approach to combat the rapidly escalating computational demands is to use networks of naturally stochastic units called probabilistic bits (p-bits). To date, hardware implementations of p-bits have predominantly relied on thermally unstable nanomagnets. However, the search continues for alternative material platforms exhibiting easily accessible, intrinsic forms of stochasticity. In this work, we demonstrate hitherto unreported p-bit functionality in epitaxial thin films of La 0.67 Sr 0.33 MnO 3 , a material showing an electrically triggered metal-insulator transition, grown on twin-textured LaAlO 3 substrates. By leveraging the combined effects of phase and structural inhomogeneities, we show two distinct modes of voltage-tunable stochastic operation: clocked binary switching and unclocked multi-bit switching. Our findings suggest that the differences result from variations in the energy landscape near the phase transition. This tunability of the energy landscape is promising for designing diverse stochastic behavior within the same material, highlighting its potential for applications in true random number generation for cryptography and probabilistic computing. I Introduction In recent years, the computing demands and energy consumption of modern artificial intelligence algorithms have surged dramatically, driving the development of alternative, domain-specific computing paradigms beyond the conventional von-Neumann architecture Big 2018 ; Bey 2020 . One such example is probabilistic computing, often regarded as the classical counterpart of quantum computing Kaiser and Datta 2021 ; Camsari et al. 2019 ; Chowdhury et al. 2023 . This computing scheme can efficiently address tasks typically performed by quantum computers — such as inference Ostwal et al. 2018 , invertible logic Camsari et al. 2017a and combinatorial optimization Sutton et al. 2017 ; Borders et al. 2019 — while circumventing the associated practical challenges of decoherence and cryogenic operation Preskill 2018 . The fundamental building block of probabilistic computers is the probabilistic bit (p-bit), a stochastic unit which function as a tunable random number generator. The p-bit’s normalized output randomly fluctuates between 0 and 1 and can be controlled through an input parameter. Thus, a physical p-bit requires a material platform that is inherently stochastic yet controllable. Designing hardware implementations of p-bits relies on tuning the energy landscape in materials. While a range of non-CMOS-based physical systems have demonstrated p-bit functionality Studholme and Brown 2024 ; Valle et al. 2022 ; Woo et al. 2022 ; Gutiérrez-Finol et al. 2023 ; Park et al. 2022 ; Rhee et al. 2023 ; Liu et al. 2022 ; Woo et al. 2024 , the most widely adopted approach involves thermally unstable nanomagnets, integrated into magnetic tunnel junctions (MTJs) Camsari et al. 2017b ; Ostwal and Appenzeller 2019 ; Borders et al. 2019 ; Vodenicarevic et al. 2017 ; Daniels et al. 2020 ; Selcuk et al. 2024 . In these systems, a precisely tuned low-energy barrier enables the magnetization to stochastically switch between two states, assisted by thermal fluctuations Camsari et al. 2019 ; Debashis et al. 2018 . MTJs with thermally stable magnets have also been implemented as p-bits, where the stochasticity can be induced by voltage-controlled magnetic anisotropy Shao et al. 2023 , spin-orbit Li et al. 2023 ; Kim et al. 2015 ; Puliafito et al. 2020 and spin-transfer torque Fukushima et al. 2014 ; Choi et al. 2014 and dual-biasing Lv et al. 2019 ; Lv et al. 2022 (a detailed review can be found in Zink et al. 2022 ). Although such MTJ-based solutions are fast, robust and scalable, they are constrained by the complexity of their stack structures and the limited tunability of the constituent materials. These limitations underscore the importance of exploring alternative materials exhibiting natural unpredictability that can be harnessed and controlled in a simple manner. Strongly correlated oxides like La 0.67 Sr 0.33 MnO 3 (LSMO), with several closely coupled degrees of freedom, are a promising class of materials for this purpose. LSMO exhibits a temperature-dependent metal-insulator transition (MIT) which can be electrically triggered, resulting in volatile resistive switching Salev et al. 2021 ; Jaman et al. 2023 ; Jaman et al. 2025 . In the critical region near the phase transition, the coexistence of multiple competing phases gives rise to rich electrical nonlinearities, which can inherently produce a stochastic response Imada et al. 1998 ; Dagotto et al. 2001 . In this work, we demonstrate stochastic electrical switching in epitaxially grown thin films of LSMO on twin-textured substrates of LaAlO 3 (LAO). We have exploited the natural phase inhomogeneity in LSMO, coupled with the unique topography of the substrate, to design a cascade of competing and coexisting energy states that the system can probabilistically switch between. The innate structural variability in the substrate allows us to demonstrate two distinct modes of voltage-controlled stochastic operation: clocked binary switching between two current states in one sample, and unclocked multi-bit switching between multiple current states in another. This substrate-induced tunability of the energy landscape provides an avenue for engineering diverse forms of stochasticity within the same material platform. Autocorrelation analyses further confirm that the switching sequences generated by our samples are indeed purely stochastic, underscoring their potential to be used as true random number generators for cryptography and probabilistic computing. II Thin film growth and structural characterization In this work, we investigate LSMO films grown on two different LAO substrates. LAO substrates are naturally endowed with twin domains and the equilibrium domain configuration varies from one substrate to another. This is brought to light in Fig. S1, which shows optical microscopy images of the two samples (named L1 and L2), with one of them having a more sparse distribution of twin domains than the other. The LAO substrates of 5 \crossproduct 5 \crossproduct 0.5 mm 3 are obtained from CrysTec, GmbH, Germany. An atomic force microscopy (AFM) scan of a pristine substrate is shown in Fig. S2(c), displaying the structural twin domains characteristic of LAO. On each twin domain, the surface roughness has a root-mean-square value of 135 pm, indicating a smooth, atomically flat surface. Thin films of LSMO, 10 nm thick, are grown on LAO substrates by pulsed laser deposition (PLD) at a substrate temperature of 750 ∘ C and an oxygen pressure of 0.35 mbar, using a KrF laser with a laser fluence of approximately 2 J cm -2 . Following growth, the films are annealed at an oxygen pressure of approximately 100 mbar for an hour, after which, they are cooled down to room temperature at a rate of 15 ∘ C/min. The surface structure of the films is monitored in situ during growth using reflective high-energy electron diffraction (RHEED). The RHEED intensity oscillations, shown in Fig. S2(b), indicate layer-by-layer growth. The grown films are epitaxial, atomically flat and highly crystalline, as confirmed by post-growth AFM and x-ray diffraction (XRD) scans, shown in Figs. S2(a) and (c). III Results Electrical characterization is done in a van der Pauw geometry using four Ti/Au electrodes, as depicted in Fig. 1 (a). In this configuration, two electrodes source a DC voltage (or current) and the other two sense the resultant current (or voltage drop). Fig. 1 (b) shows the temperature-dependence of the electrical resistance. Both LSMO films exhibit a

Read the complete text