Industrials Sep 28, 2021 09:53 AM (GMT+8)
Recently, based on years of research, learning and practical experience in the world's leading sense, memory and computing integration in the field of vision, jiutianrui core has designed an ultra-high energy efficiency ratio (20tops / W) SRAM based sense, memory and computing integration architecture chip ada20x that can be widely used in the field of vision. This is an analog-to-digital hybrid AI vision chip that can be applied to a variety of visual scenes, with a memory computing integrated architecture like the human brain.
It breaks the system limitation of "memory wall" caused by von Neumann's computing architecture of storage and computing separation. The chip memory core unit has been returned to the chip for the first time in May 2021, and the performance verification has been completed recently. Ada20x is highly customizable. It can customize special chips with different computing power and interfaces according to different market needs. The computing power covers from 0.3tops to 200tops, but the minimum power consumption can be as low as μ W level, which can meet a variety of different visual application scenarios such as tablet computer, wearable, smart home, AR / VR, battery powered IPC, ADAS and so on. Compared with traditional digital AI vision chips, pure digital chips have large memory computing area (large computing unit) and high data handling power consumption, while ada20x has ultra-low power consumption with order of magnitude advantages, which is about 1 / 10 of the power consumption of traditional digital chips, and can achieve higher energy efficiency ratio. For a series of ultra-low power consumption such as battery powered equipment, The application field of energy efficiency ratio has unparalleled advantages of architecture and first mover technology. This is the integration technology of sensing, storage and computing, which is the most important technical breakthrough direction in the post Moore's law era.
This text is a result of machine translation.