Xilinx Kria

Xilinx Kria: A New Way to Accelerate Robotics and Vision AI Applications

Robotics and vision AI are two of the most demanding and rapidly evolving fields in the industrial sector. They require high-performance, low-latency, and adaptable computing platforms that can handle complex tasks such as machine learning, image processing, and multi-axis control. However, traditional solutions such as CPUs, GPUs, and ASICs often fall short in meeting these requirements, either due to power consumption, cost, or inflexibility.

That's why Xilinx, the leader in adaptive computing, has introduced Kria, a new family of adaptive system-on-modules (SOMs) that offer unmatched AI performance, power efficiency, and flexibility for robotics and vision AI applications. Kria SOMs are based on the Xilinx Zynq UltraScale+ MPSoC devices, which combine programmable logic, ARM processors, and AI engines on a single chip. Kria SOMs also come with pre-built hardware acceleration for common industrial applications, such as smart camera, machine vision, security, retail analytics, and more.

Xilinx Kria

DOWNLOAD

In this article, we will explore the features and benefits of Kria SOMs, how they simplify development and deployment of robotics and vision AI solutions, and how you can get started with Kria today.

What are Kria SOMs?

Kria SOMs are small form factor cards that integrate a custom-built Zynq UltraScale+ MPSoC device with memory, storage, power management, and connectivity interfaces. They are designed to be easily integrated into production-ready systems for edge deployment in various industrial scenarios.

Kria SOMs come in two variants: commercial grade and industrial grade. The commercial grade Kria SOMs are suitable for applications that operate in normal temperature ranges and do not require high reliability or ruggedization. The industrial grade Kria SOMs are designed for applications that operate in harsh environments and require extended temperature ranges, shock and vibration resistance, and long life cycles.

The first product in the Kria family is the K26 SOM, which is optimized for edge vision applications. The K26 SOM features a quad-core ARM Cortex-A53 processor running at 1.5 GHz, a dual-core ARM Cortex-R5 processor running at 600 MHz, a Mali-400 MP2 GPU running at 667 MHz, a 256-bit wide programmable logic fabric with 1.4 million logic cells, and an AI engine array with 400 AI cores delivering up to 4 TOPS of AI performance. The K26 SOM also supports up to 4 GB of LPDDR4 memory, 32 GB of eMMC storage, Gigabit Ethernet, USB 3.0, PCIe Gen2 x4, HDMI 2.0b, DisplayPort 1.2a, MIPI CSI-2 and DSI interfaces.

How do Kria SOMs accelerate robotics and vision AI applications?

Kria SOMs offer several advantages over traditional computing platforms for robotics and vision AI applications:

AI at half the power with low latency. Kria SOMs leverage the AI engines on the Zynq UltraScale+ MPSoC devices to deliver up to 4 TOPS of AI performance at less than 15 W of power consumption. This enables Kria SOMs to run complex neural networks and computer vision algorithms with high accuracy and low latency at the edge.

Accelerates the whole application from AI to control. Kria SOMs not only accelerate the AI part of the application but also the control part that involves sensor fusion, motion planning, feedback control, and communication. By using the programmable logic fabric on the Zynq UltraScale+ MPSoC devices,

c8f7815bcf