Deep learning has proven to be a successful solution in many applications when environmental parameters become variable or algorithms become complex. For use in production, there is now a solution that can handle bandwidths in excess of 200 MB/s. In keeping with the requirements for high computing power, the microEnable 5 marathon deepVCL was specially developed, on which larger neural networks can be calculated without delay using a more powerful FPGA.

microEnable 5 marathon deep VCL with CNN runtime licence for VisualApplets

The deepVCL is a Camera Link frame grabber of the programmable marathon V-series. With an input bandwidth of up to 850 MB/s, support for all Camera Link formats and the PCI Express x4 bus interface, it can be used for all industrial applications. The integrated FPGA is characterized by high parallelism of processing, low heat output, deterministic latencies and long market availability.
The deep VCL board offers high speeds and bandwidths coupled with a high prediction accuracy of results above 98%. It already comes with a CNN runtime license for VisualApplets and offers a cheaper, more energy efficient and faster overall solution than a comparable with an industrial GPU.

Full Service and Implementation Service

Deep Learning solutions on the deepVCL frame grabber are now offered exclusively as a service by us. Due to the high number of requests we already offer services as full development or implementation support. An extension of VisualApplets for self-programming is planned and will appear at a later date.
Using graphical FPGA programming with VisualApplets, suitable network architectures of different sizes and complexity can be integrated and pre-trained configuration parameters for the weights of the networks can be loaded for a variety of image processing applications.

With VisualApplets, suitable network architectures can be integrated and pre-trained configuration parameters for the weights of the networks can be loaded

Two service packages for the implementation of customer-specific deep learning applications

Two service packages are available for implementing a customer-specific deep learning application. With the full service package, all tasks are taken over by us: Setup and training of the network based on the images supplied and classified by the customer as well as the FPGA implementation. For experienced users we offer the implementation service: If users have already set up a network and trained it, we transfer and implement it together with the weights (parameters) on the frame grabber FPGA. In both cases, the delivery is tested and documented by us for the desired bandwidth and accuracy.
For more information on the service packages, please contact our sales department at Tel: +49 621 789507-39 and sales@silicon-software.de.

Since FPGAs are up to ten times more energy efficient than GPUs, CNN-based applications can be implemented particularly well on embedded and mobile systems with the necessary low thermal output. CNNs run on frame grabber FPGAs, but also on VisualApplets compatible cameras and vision sensors. Especially in the decentralized computing approach of Industry 4.0 there is a need for embedded vision with Deep Learning.

High speeds and bandwidths paired with a high prediction accuracy of the results

Learn more about Deep Learning in our Webinar on 16 October.
We will be happy to inform you about the registration procedures. If you are interested, please send us an e-mail to sales@silicon-software.de

Meet us at the upcoming trade fairs VISION (6-8 November) on booth 1C72, CRAV (24-25 October) in U.S.A., SPS IPC Drives (27-29 November) on booth 10.1-318 and ITE (5-7 December) in Japan. We look forward to hearing from you.