AI-on-modules bring opportunities for edge AI

ADLINK has developed AIoM products in collaboration with chipmakers NVIDIA and Intel and has formulated a hardware optimisation strategy to help solution architects, system integrators and OEM customers address performance, size, weight and power (SWaP) requirements for Edge AI applications.

“AI at the edge has tremendous value to mission-critical applications. AI can enhance security, productivity and safety to prevent financial and non-financial loss by enabling data-driven decision making. Applications include access control of facility security, quality assurance in manufacturing as well as railway and runway inspection in transport,” said Jim Liu, Chairman and CEO of ADLINK Technology. “To accelerate and facilitate AI deployment, ADLINK’s GPU-, VPU- and NPU-accelerated AIoM products offer building blocks that allow customers to quickly turn ideas into products and open up new business opportunities.”

ADLINK support heterogeneous compute by integrating one or more types of processing cores, and ADLINK’s AIoM offering includes:

• Mobile PCI Express Module (MXM) GPU modules: ADLINK developed a series of MXM GPU modules featuring NVIDIA Quadro Embedded GPUs based on the latest Turing architecture and popular Pascal architectures. ADLINK also provides a full line-up of MXM-compatible platforms meeting SWaP requirements of Edge AI.
• VPU-accelerated SMARC modules: ADLINK offers a variety of development kits, Vizi-AI and Neuron Pi equipped with Intel Movidius Myriad X VPU. These enable system architects and solution architects to speed up the prototyping process, thereby quickly turning their concepts into reality. Commercially available options with tight version control and longevity support are also available for deployment.
• VPU-accelerated COM Express modules: ADLINK supplies module users with high-performance modules and enables them to quickly integrate AI into their applications.

Other form factors include PC/104, VPX, CompactPCI and XMC, and standards such as USB3 Vision and GigE Vision are supported for defence, railway and manufacturing applications.

ADLINK will demonstrate its Edge AI applications based on AIoM products at Embedded World 2020, including: 

• Access control powered by ADLINK’s MXM GPU module and DLAP-3000-CFL platform will showcase how an AI-enabled network video recorder (AINVR) enables users to search hours of video footage for objects of interest to enhance security measures for factories, public transport, smart building and more.
• An inspection robot based on NVIDIA Jetson TX2 will use AI to detect broken and disconnected pipes, identify meter readings, and find anomalies at power stations, factories, etc.
• A Vizi-AI development kit allows users to simply scale to other AI products in the ADLINK catalogue. This integrated hardware/software approach gives customers the ultimate flexibility, knowing they can start development on ADLINK’s low-cost Vizi-AI and choose their processor (e.g. CPU, GPU, VPU, TPU, NPU) at the time of deployment. AI-accelerated frame grabbers PCIe-GIEIMX and PCIe-GIENVQ will detect cracked candies and read deformed labels.
• ADLINK's Neuron Pi solution based on the compact SMARC form factor enables customers to rapidly prototype for robotic applications. Supporting ROS open source robotics middleware, including ROS libraries and packages, ADLINK’s Neuron Pi provides an integrated solution for education and research to shorten robotic application development time and efforts. The Neuron Pi enables autonomous mobility of robots to enhance the human-robot collaboration required to improve industrial processes for both workers and their work environment.

ADLINK’s Edge AI solutions enable system architects to construct and optimise system architecture for both AI inferencing and training applications. ADLINK offers a diverse Edge AI portfolio including building block technology, AI-on-Modules, development kits and Edge AI platforms and servers to help customers bring AI to reality.