Variational inference (VI) is an approximation of statistically valid Bayesian inference which is well-suited for analog accelerators, and stochastic nanomagnetic devices in particular are a strong candidate to implement this feature by exploiting tunable randomness in magnetic thin-films that can be run quickly and with a low power-draw. In this work, we a) discuss how VI can be reliably implemented with a combination of low-noise nanomagnetic synapses and tunable noise generating sources (magnetic tunnel junctions (MTJs) in a single analog design and b) summarize efforts to characterize the state-dependent noise profiles of various MTJ designs for various applications.
We propose a four-terminal domain wall-magnetic tunnel junction (DW-MTJ) neuron that enables the first-ever purely spintronic multilayer perceptron with unsupervised learning. The leaky integrate-and-fire neuron has a ferromagnetic DW track coupled to a binary MTJ by an electrically insulated layer. Current through the DW track performs integration by moving the DW. Leaking occurs by moving the DW in the opposite direction of integration due to either dipolar magnetic field, anisotropy gradient, or shape variation. When the DW passes underneath the MTJ, it fires by switching between the resistive and conductive states.
In a crossbar perceptron, the DW track of each neuron is connected to the analog three-terminal DW-MTJ synapses and the MTJ terminals cascade multiple layers. Finally, an unsupervised learning algorithm results from the feedback between the neuron MTJ and the analog synapses, providing best results of 98.11% accuracy on the Wisconsin breast cancer clustering task.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.