A tiny new hardware chip mimics the human brain to retrieve memories with 25 times the energy efficiency of a modern GPU.
Memristor-based networks perform associative memory retrieval without the massive power drain of digital processors. This hardware overcomes the linear storage limit of traditional neural networks, allowing it to store far more information in a smaller space. By moving AI from digital logic to analog synapses, this technology enables complex intelligence on tiny, battery-powered devices. It marks a shift away from massive data centers toward smart hardware that can live inside a sensor or a medical implant. This leap in efficiency could bring high-end AI to the most remote parts of the world.
Distance-aware attention-inspired memristive networks for energy-efficient analog retrieval
research_square · rs-9465022
Abstract Associative memories, especially those based on low-energy analog hardware, offer a promising primitive to handle the memory dominated era of large artificial intelligence (AI) models and in-sensor/edge intelligence. However, a critical but unsolved limitation of such systems is their iterative updates, which accumulate the native hardware noise, in addition to consuming high energy and latency. While emerging attention-equivalent softmax-based dense associative memories (SDAMs) reduce