Showing 1–11 of 11 results for author: Braganca, P M

Measurement-driven neural-network training for integrated magnetic tunnel junction arrays

Authors: William A. Borders, Advait Madhavan, Matthew W. Daniels, Vasileia Georgiou, Martin Lueker-Boden, Tiffany S. Santos, Patrick M. Braganca, Mark D. Stiles, Jabez J. McClelland, Brian D. Hoskins

Abstract: The increasing scale of neural networks needed to support more complex applications has led to an increasing requirement for area- and energy-efficient hardware. One route to meeting the budget for these applications is to circumvent the von Neumann bottleneck by performing computation in or near memory. An inevitability of transferring neural networks onto hardware is that non-idealities such as… ▽ More The increasing scale of neural networks needed to support more complex applications has led to an increasing requirement for area- and energy-efficient hardware. One route to meeting the budget for these applications is to circumvent the von Neumann bottleneck by performing computation in or near memory. An inevitability of transferring neural networks onto hardware is that non-idealities such as device-to-device variations or poor device yield impact performance. Methods such as hardware-aware training, where substrate non-idealities are incorporated during network training, are one way to recover performance at the cost of solution generality. In this work, we demonstrate inference on hardware neural networks consisting of 20,000 magnetic tunnel junction arrays integrated on a complementary metal-oxide-semiconductor chips that closely resembles market-ready spin transfer-torque magnetoresistive random access memory technology. Using 36 dies, each containing a crossbar array with its own non-idealities, we show that even a small number of defects in physically mapped networks significantly degrades the performance of networks trained without defects and show that, at the cost of generality, hardware-aware training accounting for specific defects on each die can recover to comparable performance with ideal networks. We then demonstrate a robust training method that extends hardware-aware training to statistics-aware training, producing network weights that perform well on most defective dies regardless of their specific defect locations. When evaluated on the 36 physical dies, statistics-aware trained solutions can achieve a mean misclassification error on the MNIST dataset that differs from the software-baseline by only 2 %. This statistics-aware training method could be generalized to networks with many layers that are mapped to hardware suited for industry-ready applications. △ Less

Submitted 14 May, 2024; v1 submitted 11 December, 2023; originally announced December 2023.

Comments: 17 pages, 9 figures

Journal ref: Phys. Rev. Applied 22, 014057 (2024)

Implementation of a Binary Neural Network on a Passive Array of Magnetic Tunnel Junctions

Authors: Jonathan M. Goodwill, Nitin Prasad, Brian D. Hoskins, Matthew W. Daniels, Advait Madhavan, Lei Wan, Tiffany S. Santos, Michael Tran, Jordan A. Katine, Patrick M. Braganca, Mark D. Stiles, Jabez J. McClelland

Abstract: The increasing scale of neural networks and their growing application space have produced demand for more energy- and memory-efficient artificial-intelligence-specific hardware. Avenues to mitigate the main issue, the von Neumann bottleneck, include in-memory and near-memory architectures, as well as algorithmic approaches. Here we leverage the low-power and the inherently binary operation of magn… ▽ More The increasing scale of neural networks and their growing application space have produced demand for more energy- and memory-efficient artificial-intelligence-specific hardware. Avenues to mitigate the main issue, the von Neumann bottleneck, include in-memory and near-memory architectures, as well as algorithmic approaches. Here we leverage the low-power and the inherently binary operation of magnetic tunnel junctions (MTJs) to demonstrate neural network hardware inference based on passive arrays of MTJs. In general, transferring a trained network model to hardware for inference is confronted by degradation in performance due to device-to-device variations, write errors, parasitic resistance, and nonidealities in the substrate. To quantify the effect of these hardware realities, we benchmark 300 unique weight matrix solutions of a 2-layer perceptron to classify the Wine dataset for both classification accuracy and write fidelity. Despite device imperfections, we achieve software-equivalent accuracy of up to 95.3 % with proper tuning of network parameters in 15 x 15 MTJ arrays having a range of device sizes. The success of this tuning process shows that new metrics are needed to characterize the performance and quality of networks reproduced in mixed signal hardware. △ Less

Submitted 6 May, 2022; v1 submitted 16 December, 2021; originally announced December 2021.

Comments: 22 pages plus 8 pages supplemental material; 7 figures plus 7 supplemental figures

Journal ref: Physical Review Applied, 18(1) 014039 (2022)

A quasi-linear spin-torque nano-oscillator via enhanced negative feedback of power fluctuations

Authors: O. J. Lee, P. M. Braganca, V. S. Pribiag, D. C. Ralph, R. A. Buhrman

Abstract: We report an approach to improving the performance of spin torque nano-oscillators (STNOs) that utilizes power-dependent negative feedback to achieve a significantly enhanced dynamic damping. In combination with a sufficiently slow variation of frequency with power this can result in a quasi-linear STNO, with very weak non-linear coupling of power and phase fluctuations over a range of bias curren… ▽ More We report an approach to improving the performance of spin torque nano-oscillators (STNOs) that utilizes power-dependent negative feedback to achieve a significantly enhanced dynamic damping. In combination with a sufficiently slow variation of frequency with power this can result in a quasi-linear STNO, with very weak non-linear coupling of power and phase fluctuations over a range of bias current and field. An implementation of this approach that utilizes a non-uniform spin-torque demonstrates that highly coherent room temperature STNOs can be achieved while retaining a significant tunability. △ Less

Submitted 19 June, 2013; originally announced June 2013.

Ultrafast switching of a nanomagnet by a combined out-of-plane and in-plane polarized spin-current pulse

Authors: O. J. Lee, V. S. Pribiag, P. M. Braganca, P. G. Gowtham, D. C. Ralph, R. A. Buhrman

Abstract: We report on spin valve devices that incorporate both an out-of-plane polarizer (OPP) to quickly excite spin torque (ST) switching and an in-plane polarizer/analyzer (IPP). For pulses < 200 ps we observe reliable precessional switching due largely to ST from the OPP. Compared to a conventional spin valve, for a given current in the short pulse regime the addition of the OPP can decrease the puls… ▽ More We report on spin valve devices that incorporate both an out-of-plane polarizer (OPP) to quickly excite spin torque (ST) switching and an in-plane polarizer/analyzer (IPP). For pulses < 200 ps we observe reliable precessional switching due largely to ST from the OPP. Compared to a conventional spin valve, for a given current in the short pulse regime the addition of the OPP can decrease the pulse width necessary for switching by a factor of 10 or more. The influence of the IPP is most obvious at longer, smaller pulses, but also has beneficial ST consequences for short pulse switching. △ Less

Submitted 14 June, 2009; originally announced June 2009.

Comments: 14 pages, 2 figures

Enhancement in spin-torque efficiency by nonuniform spin current generated within a tapered nanopillar spin valve

Authors: P. M. Braganca, O. Ozatay, A. G. F. Garcia, O. J. Lee, D. C. Ralph, R. A. Buhrman

Abstract: We examine the effect a spatially non-uniform spin current with a component polarized partially out of the plane has on a low saturation magnetization nanomagnet free layer. Micromagnetic simulations indicate that the spin torque efficiency acting upon the reversing nanomagnet can be enhanced through this process, resulting in faster switching with smaller currents. In doing so, we determine tha… ▽ More We examine the effect a spatially non-uniform spin current with a component polarized partially out of the plane has on a low saturation magnetization nanomagnet free layer. Micromagnetic simulations indicate that the spin torque efficiency acting upon the reversing nanomagnet can be enhanced through this process, resulting in faster switching with smaller currents. In doing so, we determine that micromagnetic structure within the nanomagnets can be beneficial for reversal processes. We verify this enhancement experimentally in devices with a tapered nanopillar geometry that generates a spin current polarized partly out of plane. Finally, to take even better advantage of these effects, we examine micromagnetically the benefits of a tapered three-magnetic-layer structure that further reduces reversal times while maintaining the thermal stability of the free layer. △ Less

Submitted 15 November, 2007; originally announced November 2007.

Comments: 22 pages, 5 figures, submitted to Phys. Rev. B

Spin-Torque Ferromagnetic Resonance Measurements of Damping in Nanomagnets

Authors: G. D. Fuchs, J. C. Sankey, V. S. Pribiag, L. Qian, P. M. Braganca, A. G. F. Garcia, E. M. Ryan, Zhi-Pan Li, O. Ozatay, D. C. Ralph, R. A. Buhrman

Abstract: We measure the magnetic damping parameter a in thin film CoFeB and permalloy (Py) nanomagnets at room temperature using ferromagnetic resonance driven by microwave frequency spin-transfer torque. We obtain $α_{CoFeB} = 0.014 \pm 0.003$ and $α_{Py}=0.010 \pm 0.002$, values comparable to measurements for extended thin films, but significantly less than the effective damping determined previously f… ▽ More We measure the magnetic damping parameter a in thin film CoFeB and permalloy (Py) nanomagnets at room temperature using ferromagnetic resonance driven by microwave frequency spin-transfer torque. We obtain $α_{CoFeB} = 0.014 \pm 0.003$ and $α_{Py}=0.010 \pm 0.002$, values comparable to measurements for extended thin films, but significantly less than the effective damping determined previously for similar nanomagnets by fits to time-domain studies of large-angle magnetic excitations and magnetic reversal. The greater damping found for the large amplitude nanomagnet dynamics is attributed to the nonlinear excitation of non-uniform magnetic modes. △ Less

Submitted 22 March, 2007; originally announced March 2007.

Comments: 13 pages, 2 figures

Magnetic vortex oscillator driven by dc spin-polarized current

Authors: V. S. Pribiag, I. N. Krivorotov, G. D. Fuchs, P. M. Braganca, O. Ozatay, J. C. Sankey, D. C. Ralph, R. A. Buhrman

Abstract: Transfer of angular momentum from a spin-polarized current to a ferromagnet provides an efficient means to control the dynamics of nanomagnets. A peculiar consequence of this spin-torque, the ability to induce persistent oscillations of a nanomagnet by applying a dc current, has previously been reported only for spatially uniform nanomagnets. Here we demonstrate that a quintessentially nonunifor… ▽ More Transfer of angular momentum from a spin-polarized current to a ferromagnet provides an efficient means to control the dynamics of nanomagnets. A peculiar consequence of this spin-torque, the ability to induce persistent oscillations of a nanomagnet by applying a dc current, has previously been reported only for spatially uniform nanomagnets. Here we demonstrate that a quintessentially nonuniform magnetic structure, a magnetic vortex, isolated within a nanoscale spin valve structure, can be excited into persistent microwave-frequency oscillations by a spin-polarized dc current. Comparison to micromagnetic simulations leads to identification of the oscillations with a precession of the vortex core. The oscillations, which can be obtained in essentially zero magnetic field, exhibit linewidths that can be narrower than 300 kHz, making these highly compact spin-torque vortex oscillator devices potential candidates for microwave signal-processing applications, and a powerful new tool for fundamental studies of vortex dynamics in magnetic nanostructures. △ Less

Submitted 9 February, 2007; originally announced February 2007.

Comments: 14 pages, 4 figures

Journal ref: Nature Physics 3, 498 - 503 (2007)

Spin-transfer-driven ferromagnetic resonance of individual nanomagnets

Authors: J. C. Sankey, P. M. Braganca, A. G. F. Garcia, I. N. Krivorotov, R. A. Buhrman, D. C. Ralph

Abstract: We demonstrate a technique that enables ferromagnetic resonance (FMR) measurements of the normal modes for magnetic excitations in individual nanoscale ferromagnets, smaller in volume by a factor of 1000 than can be probed by other methods. The measured peak shapes indicate two regimes of response: simple FMR and phase locking. Studies of the resonance frequencies, amplitudes, and linewidths as… ▽ More We demonstrate a technique that enables ferromagnetic resonance (FMR) measurements of the normal modes for magnetic excitations in individual nanoscale ferromagnets, smaller in volume by a factor of 1000 than can be probed by other methods. The measured peak shapes indicate two regimes of response: simple FMR and phase locking. Studies of the resonance frequencies, amplitudes, and linewidths as a function of microwave power, DC current, and magnetic field provide detailed new information about the exchange, damping, and spin-transfer torques that govern the dynamics in magnetic nanostructures. △ Less

Submitted 4 February, 2006; originally announced February 2006.

Comments: 19 pages, 4 figures

Mechanisms limiting the coherence time of spontaneous magnetic oscillations driven by DC spin-polarized currents

Authors: J. C. Sankey, I. N. Krivorotov, S. I. Kiselev, P. M. Braganca, N. C. Emley, R. A. Buhrman, D. C. Ralph

Abstract: The spin-transfer torque from a DC spin-polarized current can generate highly-coherent magnetic precession in nanoscale magnetic-multilayer devices. By measuring linewidths of spectra from the resulting resistance oscillations, we argue that the coherence time can be limited at low temperature by thermal deflections about the equilibrium magnetic trajectory, and at high temperature by thermally-… ▽ More The spin-transfer torque from a DC spin-polarized current can generate highly-coherent magnetic precession in nanoscale magnetic-multilayer devices. By measuring linewidths of spectra from the resulting resistance oscillations, we argue that the coherence time can be limited at low temperature by thermal deflections about the equilibrium magnetic trajectory, and at high temperature by thermally-activated transitions between dynamical modes. Surprisingly, the coherence time can be longer than predicted by simple macrospin simulations. △ Less

Submitted 14 February, 2006; v1 submitted 30 May, 2005; originally announced May 2005.

Comments: 12 pages, 4 figures

Journal ref: Phys. Rev. B 72, 224427 (2005)

Adjustable spin torque in magnetic tunnel junctions with two fixed layers

Authors: G. D. Fuchs, I. N. Krivorotov, P. M. Braganca, N. C. Emley, A. G. F. Garcia, D. C. Ralph, R. A. Buhrman

Abstract: We have fabricated nanoscale magnetic tunnel junctions (MTJs) with an additional fixed magnetic layer added above the magnetic free layer of a standard MTJ structure. This acts as a second source of spin-polarized electrons that, depending on the relative alignment of the two fixed layers, either augments or diminishes the net spin-torque exerted on the free layer. The compound structure allows… ▽ More We have fabricated nanoscale magnetic tunnel junctions (MTJs) with an additional fixed magnetic layer added above the magnetic free layer of a standard MTJ structure. This acts as a second source of spin-polarized electrons that, depending on the relative alignment of the two fixed layers, either augments or diminishes the net spin-torque exerted on the free layer. The compound structure allows a quantitative comparison of spin-torque from tunneling electrons and from electrons passing through metallic spacer layers, as well as analysis of Joule selfheating effects. This has significance for current-switched magnetic random access memory (MRAM), where spin torque is exploited, and for magnetic sensing, where spin torque is detrimental. △ Less

Submitted 15 March, 2005; originally announced March 2005.

Comments: 3 pages, 2 figures, to appear in Appl. Phys. Lett

Spin-Transfer Effects in Nanoscale Magnetic Tunnel Junctions

Authors: G. D. Fuchs, N. C. Emley, I. N. Krivorotov, P. M. Braganca, E. M. Ryan, S. I. Kiselev, J. C. Sankey, J. A. Katine, D. C. Ralph, R. A. Buhrman

Abstract: We report measurements of magnetic switching and steady-state magnetic precession driven by spin-polarized currents in nanoscale magnetic tunnel junctions with low-resistance, < 5 Ohm-micron-squared, barriers. The current densities required for magnetic switching are similar to values for all-metallic spin-valve devices. In the tunnel junctions, spin-transfer-driven switching can occur at voltag… ▽ More We report measurements of magnetic switching and steady-state magnetic precession driven by spin-polarized currents in nanoscale magnetic tunnel junctions with low-resistance, < 5 Ohm-micron-squared, barriers. The current densities required for magnetic switching are similar to values for all-metallic spin-valve devices. In the tunnel junctions, spin-transfer-driven switching can occur at voltages that are high enough to quench the tunnel magnetoresistance, demonstrating that the current remains spin-polarized at these voltages. △ Less

Submitted 2 April, 2004; v1 submitted 31 March, 2004; originally announced April 2004.