Original Link:
https://phys.org/news/2025-05-based-memory-advance-brain-closer.html

Summary:
Researchers at National Taiwan University have developed a novel spintronic memory device that closely mimics the way synapses work in the human brain, marking a significant step toward more energy-efficient and accurate artificial intelligence systems. Their device, which uses “tilted anisotropy” to achieve 11 stable memory states, demonstrates exceptional reliability and could be key to future neuromorphic computing hardware.

Detailed Contents:

• Background and Motivation:
  Traditional memory devices are limited by binary states (0 and 1) and often suffer from instability, making them less ideal for brain-like, or neuromorphic, computing. The human brain, in contrast, relies on synapses that can strengthen or weaken over time, enabling complex learning and memory functions.
• Breakthrough Device Design:
  The research team introduced three new spintronic memory device designs, all controlled solely by electric current-eliminating the need for external magnetic fields. Among them, the device based on “tilted anisotropy” stood out by providing 11 distinct, stable memory states with a very low cycle-to-cycle variation of just 2%1.
• Performance and Reliability:
  This multi-state memory device reliably simulates how biological synapses adjust their strength, a crucial feature for machine learning and AI. The device’s stability and repeatability make it suitable for neuromorphic applications, where reliability is essential.
• Application in AI:
  To test real-world applicability, the researchers mapped the device’s 11 memory states to digital weights in a convolutional neural network (ResNet-18) used for image classification. Using post-training quantization, they achieved a classification accuracy of up to 81.51%, nearly matching the original software model’s performance.
• Implications:
  This advance demonstrates that spintronic synapses could enable neuromorphic chips that are faster, more compact, and far more energy-efficient than current technology. Such chips could revolutionize AI hardware by mimicking the brain’s efficiency and adaptability.
• Expert Statement:
  “Our work paves the way for reliable, scalable neuromorphic computing using purely electrical control,” said Prof. Chi-Feng Pai, highlighting the significance of this development for future AI and computing systems.

Conclusion:
The optimized spintronic memory device from National Taiwan University represents a major leap toward brain-like, energy-efficient computing, providing a promising foundation for next-generation AI hardware and neuromorphic systems.