Nov, 2021 - By SMI
Kneron has unveiled the first-ever automotive-grade chip, backed by investment from Alibaba, Qualcomm, Sequoia, Foxconn, and Horizons Ventures which it claims will pave the way for vehicle automation.
Legacy and technology companies such as Qualcomm, Intel, and AMD have long controlled the semiconductor industry. Kneron, which has received more than US$ 100 million in funding, has swiftly established itself as a startup worthy of attention. Otus, which provides additional ADAS to Mitsubishi, Toyota, and Nissan, is already using its chips. The company hopes to gain a larger portion of the automotive market with the new chip, the Kneron KL530.
The Kneron KL530 features two significant innovations: it enables Vision Transformers rather than standard Convolutional Neural Networks, and it includes 4-bit data processors. As a result, image detection is much more precise and computing time is reduced. Both ViTs and CNNs are deep learning models, but transformers have been hard to implement in hardware until now. The transformer is a totally new type of architecture design that can greatly outperform the performance and accuracy [of CNN processors]. The transformer can do significantly better in anything connected to time-series judgment, such as voice recognition or behavior recognition.
In contrast to CNN’s, ViTs draw assumptions comprehensively, based on an aggregation of all features in an area rather than on particular features. As an example, consider a large white truck against an overcast sky. A CNN may infer that the large white item is also part of the sky, placing the vehicle at the chance of colliding with it, whereas a ViT would analyze not just the visible white area but also the features that distinguish it as a truck, such as the wheels and the border surrounding it. This can handle more fps, cutting operating time by up to 66% without increasing power usage. It can identify more apertures in a given amount of time, allowing tasks like image recognition to be accelerated by up to a half-second. The chipset has up to twice the TOPS/watt of the KL520 custom chip. The processor also has an image system processor, allowing for 1080P resolution, blind spot identification, categorization, range measuring, and threat recognition.
The microchip will be available in two versions: one for aftermarket use and one that will be incorporated right into the vehicle. The chips will be utilized in upcoming automobiles manufactured by Foxconn under the Foxtron brand. Kneron is currently distributing prototype orders to other clients and expects to begin mass production in 2022.
533 Airport Boulevard, Suite 400, Burlingame, CA 94010, United States
403, 4th Floor, Bremen Business Center
Aundh, Pune, Maharashtra 411007
In search of customized market research solution? We are here to help you. Contact us.