Kneron Edge AI SoC Powered by Andes RISC-V Processor Core D25F

Find the Latest SoC Solutions
for... Automotive... IoT... Security... Audio... Video

aithority.com, Nov. 09, 2021 – 

Kneron Inc., the San Diego-based Edge AI solution provider, together with Andes Technology Corporation, a leading supplier of high-performance, low-power 32/64-bit RISC-V processor cores, announced formal mass production of Kneron's next-generation Edge AI SoC KL530, powered by Andes' D25F processor in consideration of its efficient pipeline architecture, powerful Packed-SIMD DSP extension instructions, and IEEE754-compliant high-performance single/double precision floating RVFD extensions.

KL530 is the latest generation of heterogeneous AI chip from Kneron, with a brand new NPU architecture. It is the first of its kind in the industry to support INT4 precision and Transformer. Compared with other Edge AI chips, it has higher computing efficiency and lower power consumption. The use of heterogeneous AI chips embedded with RISC-V processors, powerful image processing capabilities and interfaces will further enable the application of Edge AI chips in ADAS, AIoT and other market.

The computing power of KL530 can reach 1 TOPS@INT 4, and the processing efficiency is up to 70% higher than that of INT 8 under the same hardware conditions. Its reconfigurable NPU design takes advantage of the high performance of the D25F RISC-V core, and supports multiple AI models such as CNN, Transformer, RNN Hybrid, etc. Its Smart ISP can optimize image quality based on AI, and powerful codec can achieve high-efficiency multimedia compression. In addition, its cold start time is less than 500ms, and average power consumption is less than 500mW.

The D25F CPU, one of the most popular cores from AndesCore 25-series, is equipped with RISC-V P-extension ISA draft to efficiently manipulate multiple data sets simultaneously in one instruction. Andes initiated the P-extension, chairs its Task Group in RISC-V International and leads the specification definition. D25F is accompanied with complete development tools including compiler with auto-generation of SIMD instructions based on vector data type, optimized DSP libraries, neural network libraries, and near cycle-accuracy simulator. It delivers near 9 times speedup for popular machine learning algorithms, including Tensorflow keyword spotting, CIFAR10 image classification, and P-net object detection.

Recommended AI News: Smart Engines Brings OCR to the Travel Industry in Kazakhstan

click here to read more...

 Back