Recent article from a French publication talking about our new partnership with EDGX but also our new capabilities.
Particularly liked the comments by Laurent Hill, an engineer at ESA.
BrainChip se rapproche du belge EDGX pour placer sa puce neuromorphique Akida dans l’espace
www.lembarque.com
Translated.
BrainChip moves closer to the Belgian EDGX to place its neuromorphic chip Akida in the’space
Published 09-10-2023 by Pierrick Arlot
ArchitectureBrainChip
The American company of Australian origin’
BrainChip, which has developed under the name of’Akida a neuromorphic processor for network edge (edge), has signed a technological agreement with
the Belgian start-up EDGX, created in 2023 to develop intelligent space systems of rupture, in order to develop data processing units for very demanding environments. According to the young company based in Leuven, EDGX is currently developing a new generation of space computers that combine the classic acceleration of (IA) artificial intelligence’ with neuromorphic processing to reduce the power dissipation of embedded AI, increase payload adaptability and pave the way for autonomous learning capabilities embedded within next-generation satellites themselves.
BrainChip, for its part, has developed with Akida an AI processor that is 100% event-driven% digital that uses neuromorphic principles mimicking the human brain and that analyzes only the essential data detected at the point of’ acquisition. The’ idea is to process the data with efficiency and precision, while limiting the consumption. Moreover, the Akida chip, which can also be implemented in the form of d’’a’IP block in a system chip or an Asic circuit, allows the’learning on-premises, independently of the cloud, this, BrainChip says, ensures reduced latency and improves the privacy and security of data.
«
We have built the Akida technology roadmap with the AIM of providing high-performance autonomous processing with very low energy consumption that does not require a permanent connection with the cloud, this is essential for constrained environments such as satellites », says Peter van der Made, founder and technical director of BrainChip.
EDGX and BrainChip signed a non-binding memorandum of understanding at the European Space Agency (ESA) conference EDHPC held from 2 to 6 October 2023 in Juan-les-Pins (France).
« What fascinates us most about Akida technology is its ability to operate in two different modes, says Laurent Hili, engineer in microelectronics and data processing at’ESA. On one side the Akida processor can run conventional (CNN) convolutional neural networks, making it compatible with many proven CNN networks. But on the other HAND, in event-driven mode of operation, it is able to reduce the demand for’energy by several orders of magnitude. Knowing that consumption is a primordial concept in the’space environment, the, this type of operation can open the door to a new area of processing in the’space to achieve things that were not previously possible. »
At the beginning of October, BrainChip announced the availability of the second-generation Akida platform, which can be used in applications as diverse as the connected home, the smart city, the industrial’ and the automotive’, and designed for the extremely eco-efficient processing of complex neural network models on network edge devices (edge).
According to BrainChip, the support for 8-bit’activation levels and weight values, and, as well as long-range connection jumps (long-range skip connections) here extends the range of models fully accelerated by the hardware resources of the Akida platform. The second generation of the Akida platform now integrates space-time convolutions of the TENN (Temporal Event-based Neural Networks) type, and, a new class of neural networks that are distinguished by their lightness and efficiency in the’use of resources and are particularly well adapted to the needs of applications where it is necessary to treat in continuous raw data containing important time information. Like video analysis, target tracking, audio classification, MRI and computed tomography analysis for the prediction of vital signs, or the analysis of time series used in predictive maintenance.
This technology, combined with the hardware acceleration of Vision Transformers (ViT) models, paves the way for edge devices capable of handling advanced vision and video applications that consume no power, according to the company, only a few milliwatts at the level of’a sensor (a few microwatts for audio or similar applications).
«
The treatment of generative’IA and large LLM language models at the edge level is essential to intelligent situational knowledge in verticals ranging from manufacturing industry to healthcare in the field of going through the defense, Jean-Luc Chatelain, managing director of Verax Capital Advisors, said.
Breakthrough innovations such as BrainChip's TENN technology supporting Vision Tranformers based on neuromorphic principles can provide compelling solutions in ultra-low power devices and small format, without compromising accuracy. »