The perfect combination of high-performance self-developed NPU and DSP/CPU flexible programmability of Artosyn supports more neural network operators and completes more tasks in a shorter time.
Developed a computing array technology with completely independent intellectual property rights, and the computing structure of the chip has crossed from 2D to 3D era, providing abundant computing power resources for deep learning applications.
Using compression, multi-level buffering, computing rearrangement and other technologies, an efficient interconnection structure is created for the chip and the data transmission load is greatly reduced.
The NPU software compiler has two quantization techniques:
PTQ (Post-Training Quantization) and QAT (Quantization Aware Training) help customers easily port AI algorithms to chips.
The high-performance NPU brings a new change to ISP, equips cameras with a new brain and greatly improves the image quality of traditional ISP.
Using the integrated storage and computing technology, the perfect combination of large computing power and low power consumption is suitable for all kinds of extreme use environments.
Sweep and add WeChat
© Copyrights 2011-2023 Hefei Kuxin Microelectronics Limited 皖ICP备2022016767号
