I did not know that there are now also German articles about us. It's already a month old but we here like Akida!
"These are the top articles of the month on all-electronics
Which articles on all-electronics.de have the readers been most interested in over the past month? We have put together the top 10 for you.
The top articles from March 2023
10th place: Why the EU Machinery Regulation now belongs on the agenda
9th place: How Mathworks intends to systematize development
8th place: ZVEI guest comment: Bidirectional charging – AC or DC?
7th place: Pi-Day: Everything there is to know about the circle number Pi
6th place: Tesla's plans with rare earths and SiC - against the trend
5th place: Highlights and impressions from Embedded World 2023
4th place: "For me, sustainability is THE economic factor of the future"
3rd place: Why the Extremely Large Telescope is also a triumph of automation
2nd place: Fraunhofer technology makes quantum computers suitable for industry
and finally
1st place: This is achieved by the 2nd generation Akida platform from Brainchip
The manufacturer of digital neuromorphic AI IP announces the second generation of its Akida platform. This should enable efficient and intelligent edge devices for AIoT (Artificial Intelligence of Things). Which technologies ensure this.
This brings the Vision Transformer acceleration to the second Akida generation (Image: Brainchip)"
See the full article below
https://www.all-electronics.de/markt/das-sind-die-top-artikel-des-monats-auf-all-electronics-812.html
________________________
Because it is so beautiful here again:
from German
"AIoT: Vision Transformer and Spatial-Temporal Convolution
This is what Brainchip's 2nd generation Akida platform does
The manufacturer of digital neuromorphic AI IP announces the second generation of its Akida platform. This should enable efficient and intelligent edge devices for AIoT (Artificial Intelligence of Things). Which technologies ensure this.
The second generation of Akida now includes Temporal Event Based Neural Nets (TENN) Spatial-Temporal Convolutions for processing continuous-time raw data streams. (Image: Brainchip)
The very efficient yet powerful neural processing system developed by Brainchip for embedded edge AI applications now offers 8-bit processing along with features such as time domain convolutionos and vision transformer acceleration. This enables high levels of performance in sub-watt devices, taking them from perception to cognition.
The second generation of Akida now includes Temporal Event Based Neural Nets (TENN) spatial-temporal convolutions (Spatial-Temporal Convolutions) that allow the processing of continuous-time raw data streams such as video analysis, target tracking, audio classification, analysis of MRI and CT scans for prediction of vital signs and time series analysis for forecasting and predictive maintenance. These skills are badly needed in industry, automotive, digital healthcare, smart home and smart city. The TENNs allow for significantly simpler implementations by retrieving the raw data directly from the sensors, drastically reducing the model size and operations performed while maintaining very high accuracy.
The Akida platform at a glance: From the Akida-E for maximum efficiency, through the Akida-S (sensor balanced) to the Akida-P with maximum performance. (Image: Brainchip)
market Spiking Neural Networks
With the early access program for the neuronal SoC Akida, Brainchip is responding to the high demand for development systems after the announcement of the start of production for autumn 2020. You can find out more about the early access program here.
This brings the Vision Transformer acceleration to the second Akida generation
Another addition to the second Akida generation is Vision Transformers Acceleration (ViT), a neural network that has proven extremely powerful in various computer vision tasks such as image classification, object recognition, and semantic segmentation. This powerful acceleration, combined with Akida's ability to process multiple layers simultaneously and hardware support for skip connections, allows the execution of complex networks like RESNET-50 to be managed entirely in the neural processor itself without CPU intervention, minimizing system load .
The Vision Transformer technology is particularly powerful for tasks such as image classification and object recognition. (Image: Brainchip)
The Akida IP platform has the ability to learn on the device, enabling continuous improvement and dataless customization that enhances security and privacy. This, combined with the available efficiency and performance, allows for highly differentiated solutions that were previously not possible. This includes secure small form factor devices such as B. Hearing aids and wearable devices that process raw audio data, as well as medical devices that monitor heart and breathing rates and other vital data that consume only 1 µW of power. This scales up to HD resolution imaging solutions delivered via high quality battery powered or fanless devices,
TENNs allow for significantly easier implementation by retrieving raw data directly from the sensors, reducing model size and operations performed. (Image: Brainchip)
electronics development Spiking Neural Network will initially be launched as an FPGA solution
Brainchip Holdings Ltd. specializes in neuromorphic computers. is the first company to mass market a spiking neural network (SNN) architecture: the Akida neuromorphic system-on-chip (NSoC) device. Read the background to the presentation of the first Akida generation here.
Brainchip CEO Sean Hehir on the second generation Akida
"Our customers wanted us to enable advanced predictive intelligence, target tracking, object detection, scene segmentation, and advanced image processing capabilities. This new generation of Akida enables designers and developers to do things previously not possible in a low-power edge device "By inferring and learning from raw sensor data without the need for pre-processing of digital signals, we're taking a big step towards a cloudless edge AI experience."
Software and tooling by Akida
Akida's software and tooling further simplify the development and delivery of solutions and services with these features:
- An efficient run-time engine that manages model accelerations autonomously and completely transparent to the developer.
- MetaTF software that developers can use with their preferred framework, like TensorFlow/Keras, or their development platform, like Edge Impulse, to easily build, fine-tune, and deploy AI solutions.
- Supports all types of Convolutional Neural Networks (CNN), Deep Learning Networks (DNN), Vision Transformer Networks (ViT), as well as Spiking Neural Networks (SNNs), making it future-proof as models become more advanced.
Akida has a model zoo and a growing ecosystem of software, tool and model providers as well as IP, SoC, foundry and system integrator partners. BrainChip works with early adopters on the second generation IP platform. General availability will occur in Q3 2023.
The author: Dr.-Ing. Nicole Ahner
(Image: Hüthig)
Her enthusiasm for physics and material development ensured that she found her true calling during her electrical engineering studies, which she then also placed at the center of her professional work: microelectronics and semiconductor production. After years in semiconductor research, she now researches and writes with deep specialist knowledge about electronic components. Her special interests are in wide-bandgap semiconductors, batteries, the technologies behind electromobility, materials research topics and electronics in space.
More articles from Nicole Ahner"
https://www.all-electronics.de/elektronik-entwicklung/das-leistet-die-akida-plattform-der-2-generation-von-brainchip-526.html