Nothing new but I see our Gen 2 got an article recently in a French publication.
BrainChip passe la deuxième avec sa puce neuromorphique taillée pour l’embarqué à ultrabasse consommation
www-lembarque-com.translate.goog
BrainChip goes second with its neuromorphic chip tailored for ultra-low-power embedded devices
Posted on 04-19-2023 by Pierrick Arlot
The Australian-based company
BrainChip, which developed in the form of IP a 100% digital neuromorphic processor intended to instill artificial intelligence (AI) in SoC system chips at the edge of the network (edge), and this at difficult levels of performance and consumption to be achieved by other approaches, announces the availability of the second generation of its Akida platform for the third quarter of 2023. The key: efficient 8-bit processing which, according to the company, opens the way to advanced functionalities such as as time-domain convolutions and acceleration Vision Transformers that aim to deliver an unprecedented level of performance in edge products consuming less than 1W. Features that will be able to give these products cognitive abilities,therefore beyond “simple” perception.
In detail, the second generation of the Akida platform now integrates spatiotemporal convolutions of the TENN (Temporal Event-based Neural Networks) type, a new class of neural networks which are distinguished by their lightness and efficiency in use. of resources and which are particularly well suited to the needs of applications where it is necessary to continuously process raw data containing important temporal information. Like video analysis, target tracking, audio classification, analysis of MRIs and CT scans for the prediction of vital signs, or time series analysis used in predictive maintenance.
According to BrainChip, such capabilities are needed in areas like industrial, automotive, digital health, smart home and smart city.
TENNs would thus allow simpler implementations on silicon because they can consume raw data directly from the signals recovered by the sensors. Which, adds BrainChip, significantly reduces the size of the models and the number of operations to be performed, while maintaining very high accuracy.
The other important addition to the second generation of the Akida platform is the acceleration of Vision Transformers (ViT), edge-of-network neural networks that have proven to perform extremely well on various machine vision tasks, such as image classification, object detection and semantic segmentation. This type of acceleration, combined with the ability of the Akida platform to process multiple layers simultaneously and to support residual or skip connections, allows it to manage the execution of networks itself. complexes like ResNet-50 entirely within the neural processor without host CPU intervention, minimizing overall system overhead, the company claims.
Among the products particularly targeted by the 2nd generation of the Akida platform, BrainChip cites compact and secure devices such as hearing aids or worn on the person, which accept an input for raw audio data, or health devices capable of monitor heart and respiratory rates and other vital signs and consume only a few microwatts. The company also has its sights set on battery-operated or fanless HD resolution vision solutions for monitoring and industrial production management applications.
"
We are seeing a growing demand for real-time intelligence embedded in AI applications powered by our microcontrollers and the need to make sensors smarter in industrial and IoT equipment," notes Roger Wendelken, VP of IoT and Infrastructure . by Renesas.
We have licensed Akida Neural Processors because of their unique neuromorphic approach that brings ultra-efficient acceleration to the currently most prevalent AI models at the edge of the network. With the addition of advanced temporal convolutions and vision transformer accelerators, we anticipate that low-power microcontrollers will revolutionize vision, perception and predictive applications in a wide variety of markets such as industrial IoT and consumer and personalized healthcare, to name a few . »
"
The next generation of the Akida platform enables designers and developers to do things that weren't possible before in a low-power edge device ," said Sean Hehir, CEO of BrainChip.
By providing learning and inference capabilities from raw sensor data, which eliminates the need for digital signal pre-processing, we are taking an important step towards delivering a cloud-agnostic Edge AI experience. . »
In order to simplify the development, optimization and deployment of AI solutions and services, BrainChip additionally offers an efficient execution engine that autonomously manages model accelerations in a transparent way for the developer, as well as the MetaTF software which developers can use with their favorite AI framework, like TensorFlow/Keras, or with an AI development platform like the one offered by
Edge Impulse .
Among its technological partners, BrainChip counts companies such as Arm, Intel,
Prophesee and
SiFive , to which are added in the field of AI/ML development Ai Labs, Edge Impulse, Emotion3D and
nViso .
Note that the 2nd generation Akida IP will be offered in three product classes. The IP Akida-E, extremely energy efficient, will target an operation always very close to the sensors or can be integrated within sensors. The Akida-S IP will be able to be embedded in a microcontroller or other general-purpose platforms used in a wide variety of sensor-related applications. Finally the Akida-P IP will target mid- to high-end configurations with optional vision transformers for very high performance while remaining eco-efficient.
You can also follow our news on the LinkedIN showcase of L'Embarqué dedicated to artificial intelligence in the embedded: Embedded-IA
I will then step into a plane and come over to TALK to you all in person!!! The management DOES NOT DESERVE THEIR SALARY AND THEREFORE ALSO NOT A BONUS!!!!
..
)
