Spin-transfer torque magnetic memory as a stochastic memristive synapse for neuromorphic systems

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Spin-transfer torque magnetic memory (STT-MRAM) offers non-volatility, high speed, and endurance.
• STT-MRAM cells, known as spin-transfer torque magnetic tunnel junctions (STT-MTJs), are under extensive research.
• Emerging applications explore STT-MTJs beyond conventional memory functions.

Purpose of the Study:

• To investigate the potential of STT-MTJs as stochastic memristive devices for synaptic functions.
• To identify and compare different current regimes for programming STT-MTJs.
• To evaluate the suitability of STT-MTJs for implementing learning-capable synapses in cognitive systems.

Main Methods:

• Analysis of STT-MTJ behavior in low, intermediate, and high current programming regimes.
• System-level simulations for a vehicle counting task to assess learning system performance.
• Monte Carlo simulations to evaluate device variation robustness and compare programming regimes.

Main Results:

• STT-MTJs can operate as stochastic memristive devices for synaptic applications.
• The intermediate current regime minimizes energy consumption while maintaining high robustness to device variations.
• System-level simulations demonstrate the technology's potential for learning systems.

Conclusions:

• STT-MTJs offer a promising pathway for developing robust, low-power cognitive systems.
• The intermediate current regime is optimal for energy-efficient and reliable synaptic implementations.
• This research opens new avenues for STT-MTJ applications in neuromorphic computing.