Puyang Huang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Puyang Huang has a total of 123 co-authors. Here are the top co-authors by number of shared publications:
- Xufeng Kou
ShanghaiTech University
11 shared publications - Yumeng Yang
ShanghaiTech University
6 shared publications Xinqi Liu
ShanghaiTech Laboratory for Topological Physics, ShanghaiTech University, Shanghai 201210, China
5 shared publicationsAlbert Lee
Suzhou Inston Technology Company Ltd
4 shared publications- Guoqiang Yu
Beijing National Laboratory for Condensed Matter Physics Institute of Physics Chinese Academy of Sciences Beijing China
4 shared publications - Kang. L. Wang
Department of Electrical and Computer Engineering University of California Los Angeles California USA
4 shared publications
Latest Publications
- 1 September 2026
- Published by AIP Publishing in Applied Physics Reviews
- Vol. 13 (3)
Spin orbit torque magnetic tunneling junction for noise-tolerant and resource-saving probabilistic binary neural network
The deployment of large-scale 32-bit neural networks is hindered by the escalating parameter sizes and inherent susceptibility to noise. Here, we present a nonvolatile memory (NVM)-based probabilistic computing architecture, which harnesses both the stochastic nature and in-memory computing capabilities of NVM to enhance computation efficiency and robustness. Utilizing in-plane magnetized spin–orbit torque magneto-resistive random-access memory (i-SOT) cells, we demonstrate ultra-fast (400 ps) all-electrical switching, 1011 data endurance, and voltage-controllable probabilistic states with a low variation of 3.8%. These features enable the implementation of an on-chip SOT probabilistic binary neural network (SOT-PBNN) hardware, achieving near-baseline accuracy (88.2%) on CIFAR-10 classification tasks. Moreover, the stochastic training process endows the SOT-PBNN to attain a sevenfold improvement in classification accuracy over its 32-bit convolutional neural network counterpart under 25% write/read noise, while reducing the parameter size and inference energy by 1–2 orders. Our work establishes a lightweight framework suitable for realizing artificial intelligence platforms on resource-constrained edge devices.
- 1 September 2026
- Published by AIP Publishing in Applied Physics Reviews
- Vol. 13 (3)
Spin orbit torque magnetic tunneling junction for noise-tolerant and resource-saving probabilistic binary neural network
The deployment of large-scale 32-bit neural networks is hindered by the escalating parameter sizes and inherent susceptibility to noise. Here, we present a nonvolatile memory (NVM)-based probabilistic computing architecture, which harnesses both the stochastic nature and in-memory computing capabilities of NVM to enhance computation efficiency and robustness. Utilizing in-plane magnetized spin–orbit torque magneto-resistive random-access memory (i-SOT) cells, we demonstrate ultra-fast (400 ps) all-electrical switching, 1011 data endurance, and voltage-controllable probabilistic states with a low variation of 3.8%. These features enable the implementation of an on-chip SOT probabilistic binary neural network (SOT-PBNN) hardware, achieving near-baseline accuracy (88.2%) on CIFAR-10 classification tasks. Moreover, the stochastic training process endows the SOT-PBNN to attain a sevenfold improvement in classification accuracy over its 32-bit convolutional neural network counterpart under 25% write/read noise, while reducing the parameter size and inference energy by 1–2 orders. Our work establishes a lightweight framework suitable for realizing artificial intelligence platforms on resource-constrained edge devices.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Electron Device Letters
- Vol. 47 (6) , 1125-1128
12-in Bottom-Pinned VCMA-MRAM Devices With 0.6 ns Switching Speed, 30 fJ/bit Write Energy, and 10 14 Endurance
This letter reports high-performance VCMA-MRAM devices fabricated on 12-inch wafer. By leveraging a bottom-pinned MTJ stack and synthetic antiferromagnet engineering, the optimized structure reduces stray field to 1.5 mT, enabling highly symmetric switching with a VCMA coefficient of 71.2 fJ/V$\cdot $ m. Meanwhile, a thicker MgO barrier enhances device robustness and endurance exceeding$\mathbf {{10}}^{\mathbf {{14}}}$ cycles. Moreover, the devices sustain fast voltage-driven switching down to 0.6 ns while maintaining a low write energy of 30 fJ/bit, surpassing prior state-of-the-art MRAM counterparts. Our results establish bottom-pinned VCMA-MRAM as a promising candidate for ultra-fast, energy-efficient, and high-endurance embedded nonvolatile memory.- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Electron Device Letters
- Vol. 47 (6) , 1125-1128
12-in Bottom-Pinned VCMA-MRAM Devices With 0.6 ns Switching Speed, 30 fJ/bit Write Energy, and 10 14 Endurance
This letter reports high-performance VCMA-MRAM devices fabricated on 12-inch wafer. By leveraging a bottom-pinned MTJ stack and synthetic antiferromagnet engineering, the optimized structure reduces stray field to 1.5 mT, enabling highly symmetric switching with a VCMA coefficient of 71.2 fJ/V$\cdot $ m. Meanwhile, a thicker MgO barrier enhances device robustness and endurance exceeding$\mathbf {{10}}^{\mathbf {{14}}}$ cycles. Moreover, the devices sustain fast voltage-driven switching down to 0.6 ns while maintaining a low write energy of 30 fJ/bit, surpassing prior state-of-the-art MRAM counterparts. Our results establish bottom-pinned VCMA-MRAM as a promising candidate for ultra-fast, energy-efficient, and high-endurance embedded nonvolatile memory.- 6 December 2025
- Published by Wiley in Advanced Functional Materials
- Vol. 35 (52)
Antiferromagnetic Materials Exhibiting Unconventional Properties
In the rapidly evolving field of spintronics—which exploits the electron spin rather than its charge—antiferromagnetic (AFM) materials have recently emerged as promising candidates beyond traditional ferromagnetic (FM) materials. AFMs consist of alternating magnetic sublattices that produce nearly zero net magnetization, leading to unique characteristics such as immunity to external magnetic fields, negligible stray fields, and ultrafast magnetization dynamics. These exceptional properties make AFMs highly attractive for next‐generation ultrafast, energy‐efficient computing and data‐storage applications. In this review, three distinct classes of AFM materials—collinear AFMs, noncollinear AFMs, and altermagnets are introduced—and their unconventional physical properties and potential technological impact are highlighted.
- 6 December 2025
- Published by Wiley in Advanced Functional Materials
- Vol. 35 (52)
Antiferromagnetic Materials Exhibiting Unconventional Properties
In the rapidly evolving field of spintronics—which exploits the electron spin rather than its charge—antiferromagnetic (AFM) materials have recently emerged as promising candidates beyond traditional ferromagnetic (FM) materials. AFMs consist of alternating magnetic sublattices that produce nearly zero net magnetization, leading to unique characteristics such as immunity to external magnetic fields, negligible stray fields, and ultrafast magnetization dynamics. These exceptional properties make AFMs highly attractive for next‐generation ultrafast, energy‐efficient computing and data‐storage applications. In this review, three distinct classes of AFM materials—collinear AFMs, noncollinear AFMs, and altermagnets are introduced—and their unconventional physical properties and potential technological impact are highlighted.
- 11 October 2025
- Published by American Chemical Society (ACS) in ACS Nano
- Vol. 19 (41) , 36653-36662
Spin–Orbit Torque-Assisted Voltage-Controlled Magnetization Switching for Reliable Nonvolatile Memory
Voltage-controlled magnetic or magnetoelectric random-access memory (VC-MRAM or MeRAM), which leverages the voltage-controlled magnetic anisotropy (VCMA) effect, enables efficient ultrafast device switching on subnanosecond time scales with write energy Ew below femtojoules. However, the high write error rates (WER)─stemming from the oscillatory nature of voltage switching─hinders its development and necessitates precise time control of write pulses. Ideally, spin-orbit torque (SOT) can also support ultrafast magnetization switching, provided the current exceeds a critical threshold, determined by the magnetic anisotropy. Here, we demonstrate a back-end-of-line compatible device utilizing a Ta/Mo seed layer capable of operating in VC, SOT, or hybrid modes. In the hybrid mode, we apply subcritical SOT current to assist VCMA-induced switching. The WER of VC-MRAM can be reduced by over two orders of magnitude, which is attributed to modified magnetization dynamics and the change of micromagnetic free energy during the operation. The enhancement in WER results in an improved energy-delay product surpasses that of the conventional VC-MRAM scheme. Simulations further show that the improved performance positions the SOT-assisted VC-MRAM as a promising candidate for compute-in-memory applications, including binary neural networks.
- 11 October 2025
- Published by American Chemical Society (ACS) in ACS Nano
- Vol. 19 (41) , 36653-36662
Spin–Orbit Torque-Assisted Voltage-Controlled Magnetization Switching for Reliable Nonvolatile Memory
Voltage-controlled magnetic or magnetoelectric random-access memory (VC-MRAM or MeRAM), which leverages the voltage-controlled magnetic anisotropy (VCMA) effect, enables efficient ultrafast device switching on subnanosecond time scales with write energy Ew below femtojoules. However, the high write error rates (WER)─stemming from the oscillatory nature of voltage switching─hinders its development and necessitates precise time control of write pulses. Ideally, spin-orbit torque (SOT) can also support ultrafast magnetization switching, provided the current exceeds a critical threshold, determined by the magnetic anisotropy. Here, we demonstrate a back-end-of-line compatible device utilizing a Ta/Mo seed layer capable of operating in VC, SOT, or hybrid modes. In the hybrid mode, we apply subcritical SOT current to assist VCMA-induced switching. The WER of VC-MRAM can be reduced by over two orders of magnitude, which is attributed to modified magnetization dynamics and the change of micromagnetic free energy during the operation. The enhancement in WER results in an improved energy-delay product surpasses that of the conventional VC-MRAM scheme. Simulations further show that the improved performance positions the SOT-assisted VC-MRAM as a promising candidate for compute-in-memory applications, including binary neural networks.
- Open Access
- 16 May 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 11 (20) , eadt6084
Tunable chiral magneto-transport through band structure engineering in magnetic topological insulators Mn(Bi 1− x Sb x ) 2 Te 4
Berry curvature and spin texture are representative tuning parameters that govern spin-orbit coupling–related physics and are also the foundation for future device applications. Here, we investigate the impact of the Sb-to-Bi ratio on shaping the electronic band structure and its correlated first- and second-harmonic magneto-transport signals in the intrinsic magnetic topological insulator Mn(Bi 1− x Sb x ) 2 Te 4 . First-principles calculations reveal that the introduction of Sb not only triggers a topological phase transition but also changes the integral of the Berry curvature at the shifted Fermi level, which leads to the reversal of the anomalous Hall resistance polarity for Sb fractions x > 0.67. Moreover, it also induces the opposite spin splitting of the valence bands compared to the Sb-free host, and the resulting clockwise/counterclockwise spin chirality gives rise to a tunable unidirectional second-harmonic anomalous Hall response. Our findings pave the way for constructing chiral spin-orbitronic devices through band structure engineering.
- 16 May 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 11 (20) , eadt6084
Open AccessTunable chiral magneto-transport through band structure engineering in magnetic topological insulators Mn(Bi 1− x Sb x ) 2 Te 4
Berry curvature and spin texture are representative tuning parameters that govern spin-orbit coupling–related physics and are also the foundation for future device applications. Here, we investigate the impact of the Sb-to-Bi ratio on shaping the electronic band structure and its correlated first- and second-harmonic magneto-transport signals in the intrinsic magnetic topological insulator Mn(Bi 1− x Sb x ) 2 Te 4 . First-principles calculations reveal that the introduction of Sb not only triggers a topological phase transition but also changes the integral of the Berry curvature at the shifted Fermi level, which leads to the reversal of the anomalous Hall resistance polarity for Sb fractions x > 0.67. Moreover, it also induces the opposite spin splitting of the valence bands compared to the Sb-free host, and the resulting clockwise/counterclockwise spin chirality gives rise to a tunable unidirectional second-harmonic anomalous Hall response. Our findings pave the way for constructing chiral spin-orbitronic devices through band structure engineering.
- Open Access
- 27 March 2025
- Published by Springer Nature in Research Square
SOT-MRAM-enabled noise-tolerant and resource-saving probabilistic binary neural network
The deployment of large-scale full-precision neural networks is hindered by the escalating parameter sizes and inherent susceptibility to noise. Here, we present a non-volatile memory (NVM)-based probabilistic computing architecture, which harnesses both the stochastic nature and in-memory computing capabilities of NVM to enhance computation efficiency and robustness. Utilizing in-plane magnetized spin-orbit torque magneto-resistive random-access memory (i-SOT MRAM) cells, we demonstrate ultra-fast (400 ps), field-free write operations and voltage-controllable probabilistic states with a low variation of 3.8%. These features enable the implementation of an on-chip SOT probabilistic binary neural network (SOT-PBNN) hardware, achieving near-baseline accuracy (88.2%) on CIFAR-10 classification tasks. Moreover, the stochastic training process endows the SOT-PBNN to attain a 7-fold improvement in classification accuracy over full-precision networks under a 25% write/read noise, while reducing the parameter size by 1–2 orders of magnitude. Our work establishes a lightweight framework suitable for realizing artificial intelligence platforms on resource-constrained edge devices.
- 27 March 2025
- Published by Springer Nature in Research Square
Open AccessSOT-MRAM-enabled noise-tolerant and resource-saving probabilistic binary neural network
The deployment of large-scale full-precision neural networks is hindered by the escalating parameter sizes and inherent susceptibility to noise. Here, we present a non-volatile memory (NVM)-based probabilistic computing architecture, which harnesses both the stochastic nature and in-memory computing capabilities of NVM to enhance computation efficiency and robustness. Utilizing in-plane magnetized spin-orbit torque magneto-resistive random-access memory (i-SOT MRAM) cells, we demonstrate ultra-fast (400 ps), field-free write operations and voltage-controllable probabilistic states with a low variation of 3.8%. These features enable the implementation of an on-chip SOT probabilistic binary neural network (SOT-PBNN) hardware, achieving near-baseline accuracy (88.2%) on CIFAR-10 classification tasks. Moreover, the stochastic training process endows the SOT-PBNN to attain a 7-fold improvement in classification accuracy over full-precision networks under a 25% write/read noise, while reducing the parameter size by 1–2 orders of magnitude. Our work establishes a lightweight framework suitable for realizing artificial intelligence platforms on resource-constrained edge devices.
- 18 February 2025
- Published by Royal Society of Chemistry (RSC) in Nanoscale
- Vol. 17 (11) , 6562-6569
Controllable magnetism and an anomalous Hall effect in (Bi 1− x Sb x ) 2 Te 3 -intercalated MnBi 2 Te 4 multilayers
In (Bi 1− x Sb x ) 2 Te 3 intercalated MnBi 2 Te 4 multilayers, structural engineering is employed to modulate both the anomalous Hall response and magnetism.
- 18 February 2025
- Published by Royal Society of Chemistry (RSC) in Nanoscale
- Vol. 17 (11) , 6562-6569
Controllable magnetism and an anomalous Hall effect in (Bi 1− x Sb x ) 2 Te 3 -intercalated MnBi 2 Te 4 multilayers
In (Bi 1− x Sb x ) 2 Te 3 intercalated MnBi 2 Te 4 multilayers, structural engineering is employed to modulate both the anomalous Hall response and magnetism.
- 3 January 2025
- Published by AIP Publishing in Applied Physics Letters
- Vol. 126 (1)
Tunable interfacial Rashba spin–orbit coupling in asymmetric AlxIn1−xSb/InSb/CdTe quantum well heterostructures
We report the manipulation of the Rashba-type spin–orbit coupling (SOC) in molecular beam epitaxy-grown AlxIn1−xSb/InSb/CdTe quantum well heterostructures. The effective band bending warrants a robust two-dimensional quantum confinement effect, and the unidirectional built-in electric field arisen from the asymmetric hetero-interfaces leads to a pronounced Rashba SOC strength. By tuning the Al concentration in the top AlxIn1−xSb barrier layer, the optimal structure of x = 0.15 exhibits the largest Rashba coefficient of 0.23 eV Å as well as the highest low-temperature electron mobility of 4400 cm2 · V−1 · s−1. Moreover, quantitative investigations of the weak anti-localization effect further justify the prevailing D'yakonov–Perel spin relaxation mechanism during the charge-to-spin conversion process. Our results underscore the importance of quantum well engineering in shaping the magneto-resistance responses, and the narrow bandgap semiconductor-based heterostructures may serve as a reliable framework for designing energy-efficient spintronic applications.
- 3 January 2025
- Published by AIP Publishing in Applied Physics Letters
- Vol. 126 (1)
Tunable interfacial Rashba spin–orbit coupling in asymmetric AlxIn1−xSb/InSb/CdTe quantum well heterostructures
We report the manipulation of the Rashba-type spin–orbit coupling (SOC) in molecular beam epitaxy-grown AlxIn1−xSb/InSb/CdTe quantum well heterostructures. The effective band bending warrants a robust two-dimensional quantum confinement effect, and the unidirectional built-in electric field arisen from the asymmetric hetero-interfaces leads to a pronounced Rashba SOC strength. By tuning the Al concentration in the top AlxIn1−xSb barrier layer, the optimal structure of x = 0.15 exhibits the largest Rashba coefficient of 0.23 eV Å as well as the highest low-temperature electron mobility of 4400 cm2 · V−1 · s−1. Moreover, quantitative investigations of the weak anti-localization effect further justify the prevailing D'yakonov–Perel spin relaxation mechanism during the charge-to-spin conversion process. Our results underscore the importance of quantum well engineering in shaping the magneto-resistance responses, and the narrow bandgap semiconductor-based heterostructures may serve as a reliable framework for designing energy-efficient spintronic applications.
- 15 October 2024
- Published by American Chemical Society (ACS) in ACS Nano
- Vol. 18 (43) , 29469-29478
Integrated Artificial Neural Network with Trainable Activation Function Enabled by Topological Insulator-Based Spin–Orbit Torque Devices
Nonvolatile memristors offer a salient platform for artificial neural network (ANN), yet the integration of different function and algorithm blocks into one hardware system remains challenging. Here we demonstrate the brain-like synaptic (SOT-S) and neuronal (SOT-N) functions in the Bi2Te3/CrTe2 heterostructure-based spin-orbit torque (SOT) device. The SOT-S unit exhibits highly linear and symmetrical long-term potentiation/depression process, resulting in a fast-training of the MNIST data set with the classification accuracy above 90%. Meanwhile, the Sigmoid-shape transition curve inherited in the SOT-N cell replaces the software-based activation function block, hence reducing the system complexity. On this basis, we employ a serial-connected, voltage-mode sensing ANN architecture to enhance the vector-matrix multiplication signal strength with low reading error of 0.61% while simplifying the peripheral circuitry. Furthermore, the trainable activation function of SOT-N enables the implementation of the Batch Normalization algorithm and activation operation within one clock cycle, which bring about improved on/off-chip training performance close to the ideal baseline.
- 15 October 2024
- Published by American Chemical Society (ACS) in ACS Nano
- Vol. 18 (43) , 29469-29478
Integrated Artificial Neural Network with Trainable Activation Function Enabled by Topological Insulator-Based Spin–Orbit Torque Devices
Nonvolatile memristors offer a salient platform for artificial neural network (ANN), yet the integration of different function and algorithm blocks into one hardware system remains challenging. Here we demonstrate the brain-like synaptic (SOT-S) and neuronal (SOT-N) functions in the Bi2Te3/CrTe2 heterostructure-based spin-orbit torque (SOT) device. The SOT-S unit exhibits highly linear and symmetrical long-term potentiation/depression process, resulting in a fast-training of the MNIST data set with the classification accuracy above 90%. Meanwhile, the Sigmoid-shape transition curve inherited in the SOT-N cell replaces the software-based activation function block, hence reducing the system complexity. On this basis, we employ a serial-connected, voltage-mode sensing ANN architecture to enhance the vector-matrix multiplication signal strength with low reading error of 0.61% while simplifying the peripheral circuitry. Furthermore, the trainable activation function of SOT-N enables the implementation of the Batch Normalization algorithm and activation operation within one clock cycle, which bring about improved on/off-chip training performance close to the ideal baseline.
- 20 August 2024
- Published by American Chemical Society (ACS) in ACS Applied Materials & Interfaces
- Vol. 16 (34) , 45687-45694
Observation of Moment-Dependent and Field-Driven Unidirectional Magnetoresistance in CoFeB/InSb/CdTe Heterostructures
Magnetoresistance effects are crucial for understanding the charge-spin transport as well as propelling the advancement of spintronic applications. Here, we report the coexistence of magnetic-moment-dependent (MD) and magnetic-field-driven (FD) unidirectional magnetoresistance (UMR) effects in CoFeB/InSb/CdTe heterostructures. The strong spin-orbital coupling of InSb and the matched impedance at the CoFeB/InSb interface warrant a distinct MD-UMR effect at room temperature, while the interaction between the in-plane magnetic field and the Rashba effect at the InSb/CdTe interface induces the marked FD-UMR signal that dominates the high-field region. Moreover, owning to different spin scattering mechanisms, these two types of non-reciprocal charge transports show opposite polarities with respect to the magnetic field direction, which further enables an effective phase modulation of the angular-dependent magnetoresistance. The demonstration of the tunable UMR response validates our CoFeB/InSb/CdTe system as a suitable integrated building block for multifunctional spintronic memory and sensor designs.
- 20 August 2024
- Published by American Chemical Society (ACS) in ACS Applied Materials & Interfaces
- Vol. 16 (34) , 45687-45694
Observation of Moment-Dependent and Field-Driven Unidirectional Magnetoresistance in CoFeB/InSb/CdTe Heterostructures
Magnetoresistance effects are crucial for understanding the charge-spin transport as well as propelling the advancement of spintronic applications. Here, we report the coexistence of magnetic-moment-dependent (MD) and magnetic-field-driven (FD) unidirectional magnetoresistance (UMR) effects in CoFeB/InSb/CdTe heterostructures. The strong spin-orbital coupling of InSb and the matched impedance at the CoFeB/InSb interface warrant a distinct MD-UMR effect at room temperature, while the interaction between the in-plane magnetic field and the Rashba effect at the InSb/CdTe interface induces the marked FD-UMR signal that dominates the high-field region. Moreover, owning to different spin scattering mechanisms, these two types of non-reciprocal charge transports show opposite polarities with respect to the magnetic field direction, which further enables an effective phase modulation of the angular-dependent magnetoresistance. The demonstration of the tunable UMR response validates our CoFeB/InSb/CdTe system as a suitable integrated building block for multifunctional spintronic memory and sensor designs.