This brain chip is so efficient it could bring advanced AI to your phone
So goes the theory. The majority of CIM chips that run AI algorithms are primarily focused on chip design. They demonstrate their capabilities by using simulations rather than full-blown hardware. These chips are also unable to adapt to multiple AI tasks, such as voice perception and image recognition. This limits their ability to be integrated into smartphones and other everyday devices.
A study published in Nature this month upgraded CIM completely. Instead of focusing on the design of the chip, the international team led by neuromorphic hardware specialists Dr. H.S. Philip Wong, Stanford University and Dr. Gert Cawenberghs, UC San Diego optimized the entire system from the technology to the architecture to the algorithms that calibrate hardware.
NeuRRAM is a powerful neuromorphic computer with 48 cores in parallel and 3,000,000 memory cells. The chip is extremely versatile and tackled AI standard tasks with over 84 per cent accuracy. These included reading handwritten numbers, identifying objects and cars in images, as well as decoding voice recordings.