Kneron’s auto-grade KL730 NPU chip revolutionises edge AI
Qualcomm-backed Kneron has unveiled its latest breakthrough neural processing unit (NPU) chip, which promises to be a game-changer for edge AI.
The KL730 auto-grade NPU chip packs an integrated Image Signal Processor (ISP) and promises to bring secure and energy-efficient AI capabilities to an extensive range of applications, spanning from enterprise-edge servers to smart home appliances and advanced driving assistance systems.
The KL730 sets itself apart as a groundbreaking chip specifically designed to accommodate artificial intelligence, boasting Kneron’s renowned energy-efficient and secure technology innovation. Featuring a cutting-edge peripheral interface that seamlessly connects various digital signals like images, videos, audio, and millimetre waves, the chip unlocks the potential for diverse AI applications across multiple industries.
Notably, the KL730 addresses a key barrier to the widespread adoption of AI technology: the high costs associated with energy-inefficient hardware.
The KL730 achieves an impressive 3-4x leap in energy efficiency compared to its predecessors and claims to be up to 2x more energy efficient than major competitors in the industry.
Albert Liu, Founder and CEO of Kneron, said:
“Running AI requires AI-dedicated chips with an architecture that is completely different from anything we’ve seen before. A simple re-appropriation of adjacent technologies, such as graphics-dedicated GPU chips, simply isn’t going to do the job.
The KL730 is a game-changer for edge AI. With its unprecedented efficiency and support for transformer neural networks, we are empowering users across industries to unlock the full potential of AI without compromising on data privacy and security.”
Kneron has long championed edge AI without the need for cloud connectivity and has continually advanced secure capabilities through a series of lightweight yet scalable chips.
In 2021, Kneron introduced the KL530—a pioneering edge AI chip that supports transformer neural networks, forming the backbone of GPT (Generative Pre-trained Transformer) models.
The introduction of the KL730 to the lineup provides a base-level compute power ranging from 0.35-4 effective tera operations per second, broadening its capacity to support cutting-edge lightweight GPT large language models such as nanoGPT.
The KL730 stands out as a powerful catalyst for transforming security in the AIoT landscape, enabling users to run GPT models partially or fully offline.
By leveraging Kneo, Kneron’s proprietary and secure edge AI network, the KL730 allows AI to reside on users’ edge devices and affords them greater control over data privacy. The implications span across industries – from enterprise server solutions to vehicles to AI-powered medical devices.
Bolstered security fosters increased collaboration between devices while preserving privacy. For instance, engineers can design new semiconductor chips without exposing confidential data to major cloud companies running data centres.
Since its establishment in 2015, Kneron has consistently earned accolades for its reconfigurable NPU architecture and has garnered recognition, including the prestigious IEEE Cas Society’s Darlington Award for breakthrough technologies.
Kneron serves a diverse clientele spanning AIoT, security, automotive, and edge server applications. Renowned companies such as Toyota, Quanta, Hanwha, and Dessmann have entrusted Kneron’s expertise to fuel their technological advancements.
Companies eager to explore the possibilities enabled by the KL730 shouldn’t have to wait long, with Kneron saying that samples will be available “soon”.
(Image Credit: Kneron)
See also: IBM Research unveils breakthrough analog AI chip for efficient deep learning
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with Edge Computing Expo.
Explore other upcoming enterprise technology events and webinars powered by TechForge here.
