Unconventional computing with stochastic magnetic tunnel junctions

Time: Thursday, October 11th, 15:30
Speaker: Alice MIZRAHI, NIST

Magnetic tunnel junctions are bi-stable nanodevices which magnetic state can be both read and written electrically. Their high endurance, reliability and CMOS-compatibility have made them flagship devices for novel forms of computing. While they are mostly used as non-volatile binary memories, they can be made unstable and thus behave as stochastic oscillators. Here, we show how stochastic magnetic tunnel junctions are promising elements for low energy implementations of unconventional computing. In this goal, we present several uses of these devices.
First, an analogy can be drawn between stochastic magnetic tunnel junctions and stochastic spiking neurons. We apply neuroscience computing paradigm to these devices and demonstrate that they can be the building blocks of low energy artificial neural networks capable of on-chip learning.
Then, we demonstrate that these stochastic oscillators are capable of harnessing noise to synchronize on a source at much lower energy consumption than deterministic oscillators. This opens the way to low energy implementations of synchronization based computing.
Finally, we show that these devices are low energy true random number generators, which can be used for various applications such as cryptography and stochastic computing.

Go back