Courtesy of WEKA-Fachmedien GmbH, Tobias Schlichtmeier
Artificial Intelligence (AI) will be a very natural part of more and more business processes in just a few years. The ever-increasing hunger for data from applications around AI methods such as machine learning (ML) and deep learning (DL), which the upcoming 5G wireless standard will continue to drive, is also changing the IoT embedded market: previous standards will no longer be sufficient to meet the growing demands in the future. Similar to what was already the case with computationally intensive graphics applications with graphics processing units (GPUs), AI is also creating a need for specific hardware for high-performance computing. Only by linking hardware and software can the advantages of AI be fully exploited.
Wide range of hardware and software applications
Many new data-driven processes, services and business models, as well as the topic of Industry 4.0, require AI integration. The selection of embedded hardware based on Intel, AMD and various ARM processors is large. Depending on the application, Kontron offers Computer-on-Modules (CoMs), Single Board Computers (SBCs), motherboards, Box PCs, Panel PCs and servers for "intelligent" edge computing. However, it is important to match the right combination of performance, memory capacity, graphics power, connectivity and AI function: Only in this way can AI tasks such as ML and DL be used quickly and cost-efficiently. Embedded manufacturers need to help bridge the gap, especially with advice and pre-configured applications. For this reason, Kontron has established so-called Data Science Teams and facilitates software integration with the SUSiEtec toolset for holistic digitalisation. The new AI platforms also make it easy to get started with AI and ML. In edge-based AI applications such as predictive maintenance or visual quality inspection, developers have to adapt and train DL algorithms. Here, high-speed processing of image and video data is often required.
Scalable AI platforms
For example, a compact and cost-effective AI platform consists of a 2.5-inch PicoITX SBC with NXP's i.MX8M processor and integrated M.2 module with Google's Coral Edge Tensor Processing Unit (TPU), which performs up to 4 TOPS. Designed for the temperature range of -40 °C to +85 °C, 30 or more frames per second (fps) are possible with the kit; depending on neural network structure, host CPU and connected camera systems. Free access to pre-trained models in TensorFlow Lite helps to get started with AI quickly and reduces the time-to-market for own applications. In addition to the SBC with ARM processor, Kontron also has an AI system platform for x86 applications in its portfolio. It is based on the "KBox A-203" with Intel's Atom-E39xx processor and integrated Google Coral module.
Raspberry Pi picks up speed
In addition to ARM processors from NXP and STMicroelectronics, Kontron supports the ARM processor "R3399K" from Rockchip on a SMARC module. It is predestined for cost-effective high-end applications in the areas of kiosk, retail, POS/POI and video walls. The number of industrial applications with the Raspberry Pi is also increasing. For this, the user has a choice of Kontron's "PiTron" SBCs as well as a family of industrial controllers with "PiXtend". An active software community and many easily adaptable applications ensure fast project implementation and a short time-to-market in this environment as well.
In the next part of our blog post, you will find out which trends in the field of AI, high-performance computing and edge computing will be decisive for Industry 4.0 in the coming years and which solutions Kontron also offers for small and medium-sized companies here.
You can find the 2nd part of our blog series at:
{{comment.comment}}