Designed to bring AI computation power to edge devices for various applications, the chip – called the KL520 – combines proprietary software and hardware designs to create a highly efficient and ultra-low-power neural processing unit (NPU). Running AI computations on the end device, says the company, will help generate real-time insights without relying on the cloud.
The KL520’s reconfigurable architecture allows it to run a number of different convolutional neural networks (CNNs) based on different applications, regardless of its kernel size, architecture requirements, or input size. In turn, says the company, the chip is able to achieve a very high MAC (media access control) efficiency, a key metric for efficient computation.
This software design allows the chip to dynamically adjust its function based on applications needs in order to achieve maximum efficiency. The KL520 is also equipped with a proprietary compression technology that is able to compress a large AI neural network into a very small footprint to process on the edge device, without sacrificing large amounts of power.
The KL520’s features and targeted applications are listed as follows:
- Low-powered with a small physical footprint
- The KL250 can run alongside a main chip as a co-processor; will not need a replacement chip
- For smart door lock applications, KL520 includes two ARM Cortex M4 CPU, which can serve as the main processor.
- Balances the need of performance, power and cost to bring the best solution for edge applications
- Applicable to various 3D sensor technologies such as structured light, dual-camera and ToF, and Kneron’s exclusive 3D sensing technology
- Well-suited for applications including smart locks, security cameras, drones, smart home appliances and robotics
In addition to the chip release, the company also announced partnerships with industry leaders to bring the chip “into our everyday lives.” Partnerships include fabless IC maker Etron Technology, Himax Technologies, enterprise PC OEM AAEON Technology, 3C solution provider Alltek Technology Corp, renowned ODM players Pegatron, as well as 3D sensing provider Orbbec, and IC design house Datang Semiconductor.
The company says that it will be releasing its second-generation chip during Q4 this year, targeted toward the surveillance and security market.
AI startup announces neural processors for edge devices
Edge AI boom faces power consumption, size challenges
Edge AI startup launches new-generation NPUs
AI at the edge to significantly impact semi industry