Deep Learning Expertise
Nowadays there is a large number of high-quality neural networks available for free use. However, in order to use the maximum of data, deep learning framework and hardware it is essential to develop specialized neural networks for the use case. We at EYYES have the expertise and the tools to increase quality many times over through intelligent design, deep learning expertise and optimization tools. Without increasing the complexity or the computational effort.
Our proprietary Deep Learning Tools and optimized Deep Learning Frameworks are able to train a network optimized for the hardware platform as well as to visualize weaknesses in neural networks through Explainable AI in order to eliminate them systematically.
EMBEDDED DEEP LEARNING
Embedded Deep Learning von EYYES
Neural networks require a correspondingly high level of computing power in order to deliver the desired results promptly. Particularly in time-critical applications, where real-time data must be evaluated "in real time", for example to supply driver assistance systems with environmental information, both the software and the hardware are faced with major tasks. In order to effectively meet these requirements in "real-time", performance-optimized software and hardware is necessary.
For this purpose, EYYES offers with its Smart Neurochip Platform an updateable and at any time functionally expandable platform for use in prototype as well as production vehicles, consisting of
ADA software for the acceleration of neural networks
ADA tool for the synthesis into the hardware block
Smart Neurochip as high performance and energy saving HW for AI
ADA Deep Learning Hardware-Tuning-Tools
The software tool set - also called AVI DEEP-LEARNING-ACCELERATOR (ADA) - is able to optimize existing neural networks in such a way that an up to forty-fold increase in speed is achieved on the same hardware platform.
The hardware in the form of the Smart Neurochip (SNC), which has been specially developed for the high-performance operation of neural networks, not only increases the processing speed even further, but also reduces energy consumption to a minimum.
One of the most important advantages of the Smart Neurochip (SNC) is that no memory operations between the individual modules or function blocks and thus no copying processes are required.
The deep learning network generated by the ADA tools is configured in the software and transferred to the hardware module (= architecture) with the ADA SYNTHETISER.In contrast to conventional deep learning acceleration components, ADA and Smart Neurochip offer a universal solution for artificial intelligence that can be expanded at any time during operation.
The Generic Neurochip from EYYES
The Generic Neurochip® is a highly efficient and powerful AI processor for the entire EYYES portfolio. This neurochip provides dynamic training, dynamic calibration and deep learning in real time and is the flexible platform for the further development of the automotive industry towards "smart applications".
Maximum performance - minimum power consumption
Based on this solution, one has unimagined performance even on FPGA or ASIC basis and can achieve maximum performance with minimum power consumption. The benefit of this technology is explained using the example of a camera sensor: The power consumption for the data (objects) of a camera sensor to be calculated in real time is currently about 60 watts. With the Deep Learning Tuning Kit, the performance of the calculation is increased to such an extent that only a fraction of the power consumption is required. With the HW- and SW architecture developed by EYYES, the power consumption can be reduced to 2 Watt per camera sensor for the high-quality and intelligent evaluation of object data.
This Generic Neurochip® is used in practice in a wide variety of areas. You can find details about our products here