Adaptive spatial-temporal information processing based on in-memory attention-inspired devices | Nature Communications
Pan et al. report an attention-inspired architecture for adaptive spatial-temporal information processing based on 0D-2D hetero-dimensional interface between MoS2 and Ag filament. Wafer-scale device array is prepared for in-memory analog computing and applied to autonomous driving edge intelligence scenarios.
Download PDF Subjects Electrical and electronic engineering Electronic devices Two-dimensional materials Abstract Spatial-temporal information perception is widely used for motion processing in dynamic scenes, but present technology requires relatively huge hardware resource consumption. The attention mechanism helps the human brain extract required information from tremendous data at a low cost. Here, we propose an attention-inspired artificial intelligence architecture based on hetero-dimensional modulations between zero-dimensional contact and two-dimensional electrostatic interfaces. An ad
saved by
related reading
- transformer_attention.pdfarxiv.org
- A Million Spiking-Neuron Integrated Circuit with a Scalable Communication Network and Interfacepaulmerolla.com
- Towards a “universal translator” for neural dynamics at single-cell, single-spike resolutionarxiv.org
- 2502.11089arxiv.org
- pdfopenreview.net
- [2205.14135] FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenessarxiv.org
- [2501.00663] Titans: Learning to Memorize at Test Timearxiv.org
- Sparse Attention Post-Training for Mechanistic Interpretabilityarxiv.org
- AI Overcomes Stumbling Block on Brain-Inspired Hardware | Quanta Magazinequantamagazine.org
- [2109.13392] The Tensor Brain: A Unified Theory of Perception, Memory and Semantic Decodingarxiv.org
- Arousal as a universal embedding for spatiotemporal brain dynamicsnature.com
- Learned Spectrumtheodore.net