Embedded World 2020 was unique on many fronts, to say the least. The COVID-19 virus was in its early days, and some companies had pulled out of Embedded World. But many, upwards of 80% of the total, were in attendance. That left plenty of people to have meaningful discussions with, which included the most dominant topic, artificial intelligence (AI).
The exhibit halls were teeming with AI demos. But because AI has been a staple at Embedded World for several years, that came as no surprise. However, the breadth of demos and technologies was far more interesting this year than in the past. They ranged from high-end servers running facial-recognition algorithms to low-power Arm-based processors identifying objects on a conveyor belt. Essentially, it included technologies at various layers in the IIoT from Edge to Fog to Cloud, each with its own unique challenges and requirements for processing and security.
Server-class performance typically brings to mind climate-controlled server rooms and the nebulous Cloud computing layer. However, some of the Embedded World AI demos included high-performance ruggedized servers at the Edge. These demos were running at various compute levels and could handle functions like facial recognition, age and gender detection, and even the mood of people walking by.
Functions like these take a tremendous amount of processing power and include high-performance CPUs and GPGPUs. These are embedded systems, as they are part of another system and ruggedized for deployments in potentially harsh environments. They have the processing power to handle machine learning as well as processing the inference engine, once the processing rules are developed. It’s this kind of high-performance computing that helps drive new specifications, such as COM-HPC from PICMG.
As we have seen for the last several years, numerous companies have been developing carrier boards for the Nvidia Jetson modules. These modules combine Arm CPUs with Nvidia GPUs for power-efficient AI computing platforms. Demos based on such a platform were abundant and the list of carrier module providers continues to grow.
While the Jetson carriers are primarily used as AI inference engines, they can also be used for AI training in some applications. Numerous demos throughout Embedded World displayed the Jetson carriers in industrial processing applications ranging from machine vision to object sorting.
Not to be outdone, several manufacturers were offering Intel Movidius modules to provide AI processing capabilities in embedded systems. The commercially available USB sticks are cool, but they’re not a perfect fit for industrial applications. From the WINSYSTEMS’ perspective, the most interesting Movidius-based platforms were MiniPCIe cards with industrial heat sinks. These modules can add AI processing to industrial computing platforms such as WINSYSTEMS’ SBC35-427 single-board computer (SBC) family that features Intel’s Apollo Lake E3950 CPU. These SBCs provide considerable AI functionality, especially as Intel continues to expand its software offerings to further enable these technologies.
SBCs based on Arm processors were also on display, running various AI inference engine demos. For example, the WINSYSTEMS NXP i.MX8M-based ITX-P-C444 was running the Qt Embedded software platform and included an edge detection algorithm, scanning people as they walked by.
The power of the i.MX8M makes the ITX-P-C444 a perfect candidate for industrial AI applications. Other features include dual Ethernet, industrial I/O, and a few expansion options. The main processor is joined by an M4-based microcontroller to form a real-time subsystem that’s ideal for industrial IoT applications requiring high performance in harsh conditions, including digital signage, industrial automation, energy, building automation, and others, as well as those based on AI.
The bottom line is that AI for industrial applications was on full display at Embedded World. Hence, it’s real, it’s here, and WINSYSTEMS has the technology you need.