An international research team has developed an artificial synapse based on magnetic skyrmions that operates at room temperature. The design utilizes the collective transformation of a skyrmion lattice into stripe magnetic domains within the 2D van der Waals ferromagnet Fe3GaTe2, rather than manipulating individual skyrmions.
This collective approach ensures a deterministic and linear electrical response in the material's anomalous Hall resistance. By adjusting the duration of electrical pulses, researchers can tune the synaptic weight, enabling multiply-accumulate operations essential for neural networks.
The energy consumption of the synapse is estimated at approximately 0.66 picojoules per operation, which is comparable to existing RRAM and PCM memory technologies. The team tested the device in a quantized neural network, achieving 96.1% accuracy in handwritten digit recognition.
Unlike many quantum and magnetic phenomena that require cryogenic cooling, this technology functions under ambient conditions, removing a significant barrier to practical implementation in neuromorphic chips.
