Memristive Magnetic Memories for spintronics synapses
Published : 24 April 2017
Conventional electronic circuits consume much more energy than human brain at comparable performances (~50MW for supercomputer vs 20W for human brain). Consequently, there is a strong interest in developing electronic circuits which mimic the working principle of the brain. These circuits are particularly suited for learning functions, associative functions, pattern recognition etc. For that purpose, it is necessary to develop new electronic components which realize neurons and/or synaptic functions. Synapses are interconnection elements between neurons able to keep the memory of the history of the current pulses to which they have been submitted.
In this thesis, we propose to develop such electronic synapses based on spintronics phenomena and particularly the tunnel magnetoresistance of magnetic tunnel junctions and spin transfer torque. The component resistance will vary continuously between a minimum and a maximum value depending on the succession of current pulses flowing through it (memristor). The work will capitalize on the strong know how existing at SPINTEC on spin-transfer torque oscillators and MRAM magnetoresistive memories.
The thesis will start by micromagnetic simulations aiming at dimensioning the device and properly choosing the most appropriate materials. These materials will then be experimentally optimized and memristive devices will be fabricated in our clean room. They will then be electrically tested in the lab to demonstrate the memristive behavior. Simple neuromorphic circuits will then be built to demonstrate the capability of learning and pattern recognition.