University of Texas and TSMC test SOT-MRAM to reduce AI energy use

Engineers from the University of Texas at Austin, in collaboration with Taiwan Semiconductor Manufacturing Company (TSMC), have tested an emerging memory technology called SOT-MRAM (Spin-Orbit Torque Magnetic Random-Access Memory) designed to address the rising energy demands of artificial intelligence.

Unlike conventional volatile memory, SOT-MRAM is non-volatile, meaning it can retain information without a continuous power supply. This capability allows devices to avoid reloading data when waking from sleep modes or restarting tasks, significantly reducing energy consumption. By utilizing magnetic properties, the technology offers high speed and durability.

In testing involving neural network inference and training, the researchers achieved write operations in just 2 nanoseconds, consuming only 2 picojoules of energy per write. This performance is significantly more efficient than existing technologies, which can be hundreds of times slower and consume much higher levels of energy.