Abstract This corrigendum addresses a fundamental mathematical error in the physics-based modeling of neuromorphic hardware presented in "Magnetic Tunnel Junction Mimics Stochastic Cortical Spiking Neurons" (Scientific Reports, 2016). Specifically, it provides a critical correction to the Landau-Lifshitz-Gilbert-Slonczewski (LLGS) equation used to model Magnetic Tunnel Junctions (MTJs) as stochastic neurons. This update ensures that spin-transfer torque terms accurately represent magnetization dynamics, which is essential for simulating brain-like computing on hardware. This corrects the article DOI: 10.1038/srep30039.
Introduction
The intersection of spintronics and neuromorphic computing offers a path toward hardware that natively mimics the stochastic spiking behavior of biological neurons. Magnetic Tunnel Junctions (MTJs), typically utilized in MRAM, can be engineered to exhibit the probabilistic firing patterns observed in the brain's cortex. However, the accuracy of such hardware-based neural networks depends on the precision of the underlying physical models. This corrigendum identifies and corrects an error in Equation 2 of the original study, which governs the temporal evolution of MTJ magnetization.
Corrected Modeling of Magnetization Dynamics
The standard mathematical framework for describing the precession and damping of a magnetic moment in a solid-state device is the Landau-Lifshitz-Gilbert-Slonczewski (LLGS) equation. The correction focuses on the Spin-Transfer Torque (STT) term, which describes how electrical current influences magnetic state.
In the original publication, the interaction between the spin current ($I_s$) and the magnetization vector ($\hat{m}$) was simplified, failing to capture the full physical dynamics. The corrected equation introduces a more rigorous term:
$$\frac{1}{q\gamma N_s}(\alpha(\mathbf{\hat{m}} \times \mathbf{I}_s) - (\mathbf{\hat{m}} \times \mathbf{\hat{m}} \times \mathbf{I}_s))$$
By correctly incorporating the gyromagnetic ratio ($\gamma$) and the damping constant ($\alpha$), this version ensures the model aligns with established spintronic theory, properly balancing field-like and damping-like torque components.
Implications for Neuromorphic Hardware
Precision in these governing equations is vital for scaling up to large-scale neural network simulations. Even minor errors in device-level modeling can lead to significant discrepancies in system-level behavior.
Key insights from this correction include:
- Physical Grounding of Stochasticity: Ensures that the observed "noise" in the MTJ neuron model is physically accurate rather than a mathematical artifact.
- Reliable Engineering Blueprints: Provides a more accurate foundation for designing MTJ-based "neurons," particularly in how electrical current translates into magnetic movement.
- Hardware-Software Co-design: Highlights that hardware neurons are dynamic physical systems governed by complex differential equations, necessitating a deep integration of physics and computer architecture.
Conclusion
This correction is essential for researchers replicating the original study or utilizing the MTJ model for stochastic spiking neural network simulations. It underscores the importance of rigorous physical modeling in the development of next-generation neuromorphic hardware.
Summary updated