news

On August 8, Beijing AppleCore Technology Co., Ltd. (hereinafter referred to as AppleCore Technology) released the PIMCHIP-N300 integrated storage and computing NPU and PIMCHIP-S300 multi-modal intelligent perception chip in Beijing.

Also targeting artificial intelligence, what is the difference between integrated storage and computing chips and general AI chips? In an interview with the "Daily Economic News" and other media, Yang Yue, CEO of Apple Core Technology, said that before the emergence of integrated storage and computing, the problem faced by the industry was that the computing unit and the storage unit of the computing architecture were physically separated. Therefore, when computing occurs, data needs to be continuously transmitted through the bus between these two units. "More than 90% of the power consumption of AI computing chips is consumed in transmission. After AI comes up, the amount of computing is even more intensive than the previous general computing. In terms of efficiency, the original architecture is not very well supported."

Yang Yue Photo source: Provided by the interviewee

Yang Yue also mentioned that the industry has been seeking a solution, namely upgrading the process technology. "But this solution has actually encountered a bottleneck in recent years. The manufacturing cost is getting higher and higher, and the efficiency improvement brought by (upgrading the process technology) to the chip is increasingly unable to match the design cost and manufacturing cost." Yang Yue said that until around 2020, a group of companies began to no longer rely on advanced processes and tried to change the chip architecture to reduce data movement. Storage computing has become the most feasible option.

In Yang Yue's view, the recent trend of miniaturization, lightweight, and intelligent localization has indeed increased the difficulty of manufacturing for suppliers, but for storage computing companies, the opportunities outweigh the challenges. "We see that more and more customers need to carry some AI functions with them, including some high-end medical industries, cars, mobile phones, etc." Taking batteries as an example, Yang Yue said that the characteristic batteries that can drive AI computing are difficult to achieve in traditional architectures, which gives storage computing a great opportunity. In addition, to achieve miniaturization, storage computing also has the advantage of strong hardware IP reuse capabilities.

From the development history of storage and computing integration, since 2017, international giants such as NVIDIA, Microsoft, and Samsung have begun to propose storage and computing integration prototypes, and storage and computing integration chip companies have begun to emerge. Domestically, in addition to large companies, most start-ups focus on the field of in-memory computing, targeting different application scenarios. For example, Apple Core Technology, which released new products this time, is more focused on scenarios such as the Internet of Things, wearable devices, and smart homes.

Daily Economic News