On November 4, 2021, Kneron, a leading edge computing (Edge AI) solution manufacturer, and Andes Technology, a leading supplier of 32/64-bit RISC-V embedded processor cores (TWSE: 6533; SIN: US03420C2089; ISIN: US03420C1099), today jointly announced that Kneron Intelligence’s next-generation AI intelligent edge computing chip KL530 has been officially mass-produced. The KL530 adopts Andes’ D25F processor, which includes efficient pipeline, powerful Packed-SIMD DSP extension instruction and high-performance single/double-precision floating-point RVFD extension instruction set conforming to IEEE754.
KL530 is Kneron Intelligence’s latest heterogeneous AI chip. It adopts a new NPU architecture. It is the first product in the industry to support INT4 precision and Transformer operations. Compared with other chips, KL530 has higher computing efficiency and lower power consumption. This AI chip is embedded with a RISC-V CPU and has powerful image processing capabilities and rich interfaces, which can further promote the application of edge smart chips in ADAS, AIoT, etc. The computing power of KL530 reaches 1 TOPS INT 4. Under the same hardware configuration, the processing efficiency of INT 8 is improved by up to 70%. Its reconfigurable NPU design is matched with the high-performance computing of RISC-V D25F core, which can support CNN, Transformer, RNN Hybrid, etc. A variety of AI models, as well as intelligent ISP, can optimize image quality based on AI, powerful Codec to achieve high-efficiency multimedia compression, and the cold start time is less than 500ms, and the average power consumption is less than 500mW.
The D25F CPU is one of the most widely used cores in the AndesCore® 25 series, which supports the RISC-V P-extension ISA draft standard, which can efficiently process multiple data at the same time in one instruction . Andes is the original architect of the P extension instruction set and leads its specification in the RISC-V International Association task group. The D25F provides complete development tools, including a compiler that automatically generates SIMD instructions from the vector data format, an optimized DSP function library, a neural network function library, and a near-accurate cycle simulator. D25F can provide nearly 9x speedup on commonly used machine learning algorithms, including Tensorflow keyword recognition, CIFAR10 graph classification and P-net object detection.
“Kneron has a unique reconfigurable architecture that can be easily incorporated into different convolutional neural networks (CNNs) without compromising design requirements, allowing seamless and precise application to various AI models.” Founder of Kneron Intelligence Liu Juncheng, chief executive officer, said. “The Andes D25F CPU core and its powerful DSP instructions and its software development framework allow Kneron to maximize the performance of its peer-leading AI algorithms without sacrificing optimal power performance. Crucially. We are delighted to partner with Andes Technology, a leading computing expert focused on RISC-V. With Andes RISC-V core and DSP solutions, Kneron was able to develop smoothly in a very short period of time With this high-end solution, we are very proud that the KL530 is now in mass production and serving our customers.”
“We are very pleased that Kneron Intelligence has selected D25F as the CPU core of KL530 after a series of comprehensive evaluation,” said Lin Zhiming, CEO of Andes Technology and director of RISC-V International Association. , power consumption and other key indicators are excellent.Keron leads the industry, provides edge AI SoC solutions with embedded RISC-V core, and quickly launched KL530 into mass production, showing the team’s ultra-high efficiency. The strong competitiveness of the company is astounding. Thanks to the close cooperation between Kneron and Andes, we have jointly completed a highly competitive solution and will accelerate the application of artificial intelligence into various products.”