柔性边缘智能AI芯片突破“存储墙”性能限制

Core Insights - Tsinghua University's research team has developed the "FLEXI flexible digital computing chip," which fills a gap in high-performance flexible AI computing chip technology [1] - The chip is designed for extreme edge devices like wearable sensors and brain implants, addressing the need for local basic computation on flexible substrates with low-cost components [1][2] - The new flexible AI chip features low power consumption, low cost, and high integration, overcoming previous limitations of flexible processors that were slow and energy-intensive [1] Technical Advancements - The research team innovated by increasing the number of metal layers, enabling complex chip interconnections that traditional flexible electronics could not support [1] - The chip employs a digital "in-memory computing" architecture, processing data within the memory to eliminate time and energy costs associated with data transfer, thus surpassing the performance limitations of traditional analog in-memory computing solutions [1] Performance and Applications - The chip has been tested to work stably in folded and rolled states, maintaining its computational capabilities after 40,000 cycles of folding, and exhibits good resistance to temperature, humidity, and light aging [2] - It can be integrated into wearable devices to recognize daily human activities using physiological signals such as heart rate, respiratory rate, and body temperature [2] - A commentary in the journal Nature highlights that the chip's design and manufacturing optimization allows stable operation of neural network tasks under low-cost and low-power conditions, providing new directions for innovation in smart hardware within the Internet of Things (IoT) sector [2]