AMD Launches Program Enabling Turnkey Designs Based on its Kria SOMs

January 18, 2023 by Jeff Child

Leveraging the company’s FPGA-based Kria system-on-modules (SOMs) and its app store methodology, AMD is unveiling a new ODM partner effort.

There’s no doubt that FPGAs offer powerful capabilities that are applicable to a wide arena of applications. Problem is there’s a lot of complex engineering effort necessary getting FPGA-based designs from concept to market.

Helping to smooth the way, today AMD’s FPGA group (formerly Xilinx) announced a new program aimed at helping engineers craft turnkey systems based on AMD’s Kria system-on-modules (SOMs). The program is dubbed the AMD Kria SOM ODM Partner Ecosystem.

AMD’s Kria SOM ODM Partner Ecosystem uses its Kria FPGA modules as the basis for turnkey, pre-built hardware and software platforms.

AMD’s Kria SOM ODM Partner Ecosystem uses its Kria FPGA modules as the basis for turnkey, pre-built hardware and software platforms.


In this article, we discuss the need that AMD’s new program addresses, describe some examples of the program in action, and we share details from our interview with K.V. Thanjavur Bhaaskar, AMD’s manager, marketing and strategy for industrial, vision, healthcare and sciences.


Turnkey SOM-based Solutions 

According to AMD, the purpose of the new partner program is to enable engineer customers a way to deliver production-grade, fully functional Kria SOM-based solutions to get to market faster without the need for dedicated chip-design resources. This is a continuation of AMD’s Kria SOM effort that began in 2021.

The program is aimed at engineers at original design manufacturer (ODM) types of companies. To put the effort into context, Bhaaskar says that AMD’s new partner program consists of products built on top of AMD’s Kria K26 SOM. “This is important because that SOM offers the most accessible option in embedded electronics in the ‘make versus buy’ continuum,” he says.

The resulting ODM product is production ready off-the shelf and both the software and hardware can be customized for a fee if needed. “The result is that engineers can get to market faster with lower development cost,” Bhaaskar.

We’ve covered AMD’s Kria SOMs before, including the company’s SOM kit aimed specifically at robotics. A key part of that solution was that it expanded on AMD’s concept of an “app store” model. These apps are pre-built applications that engineers can use for evaluation with no FPGA expertise required. Each app converts the Kria SOM into a function-specific node.


Shown here is the classic make vs. buy continuum in embedded electronics.

Shown here is the classic make vs. buy continuum in embedded electronics.


We asked Bhaaskar to connect the dots as to how that app store approach fits into AMD’s new partner program. For this, he put things into the context of the classic engineering “make vs. buy” decision (see image above).

Bhaaskar explains how engineers have a choice of entry points these days (from an AMD point of view).

  • An engineer can start with an ODM product that is product ready with all the software and hardware baked in for specific use cases.
  • An engineer can start with an SOM, build their own production carrier card, enclosure and use AMD’s App Store for the application software.
  • An engineer can start with a chip-down adaptive SoC (FPGA) or develop their own ASIC.

Putting those options along the spectrum, the first option is closer to “buy” and the last option is closer to “make,” he says. “We see this new partner ecosystem as an addition and a step up in the make vs. buy continuum,” says Bhaaskar. “The Kria app store is still a significant piece for engineers starting at the ‘integrate SOM’ phase and for developers that are new to Kria and trying to get started with the platform.”


AMD App Store Keeps Growing

According to Bhaaskar, AMD’s App Store continues to grow and is constantly updated. We asked him to brief us on what changes have happened over the past several months. “We refreshed all the AMD-built apps based on PetaLinux to Ubuntu 22.04 in December for KV260,” he says. “We also had our partners at Avnet Silica create a Smart Model Select app.”

Bhaaskar says AMD also released its 10GiGE vision and TSN app as part of the KR260 launch. As an early preview, he says the company has two partner apps and one AMD app coming in the next 15 days to kickstart the new year.


AMD’s App Store lets users select function-specific pieces of functionality that can be programmed on a Kria SOM’s FPGA.


In addition, AMD has two demonstration apps to showcase how the app store’s digital rights management works and how the security capabilities of K26 SOM works for any partner planning to leverage those functions in their Kria app.

Demonstration apps are similar to AMD’s accelerated apps except they are intended for evaluation purposes only. Engineers can evaluate capabilities on a Kria starter kit before moving on to the next steps with the accelerated app provider. 

Meanwhile, Bhaaskar says AMD now has a “become a Kria app partner” form on its app store. “The form gives us a way to connect with folks interested in being part of the app store,” he says.


Kria ODM Partner Program Examples

As part of its announcement today, AMD provided details on a selection of currently available ODM products that were built using the Kria ODM Ecosystem. Among these is an industrial IoT edge gateway from Ectron. The gateway is a multipurpose industrial computer that can be deployed on the factory floor for intelligent automation and connectivity. In the system, a Kria K26 SOM powers the high-performance computing capabilities for the system's productivity algorithms.

Another example partner product is the Smilodon 10G EVO from Optomotive. A fully customizable, user-programmable, high-speed industrial smart camera, the device is based on a Kria K26 SOM. AMD says the camera consists of a rapid imaging sensor and 1- or 10-Gigabit Ethernet capability. Also included are an Arm SoC and industrial Gpixel imaging sensors.

AMD’s third example partner product is an AI edge appliance built by VVDN Technologies. This box-level system integrates a Kria K26 SOM and four Kinara Ara-1 processors (this device is shown in the top image in this article).

These technologies enable it to run vision artificial intelligence (AI) applications on eight concurrent streams. The company claims this can be done at high performance levels but at a fraction of the cost compared to GPU-based solutions.


A Shortcut to Leveraging AI

Like FPGAs, AI is a technology that requires some specialized knowledge to get up and running. With that in mind, Bhaaskar explains how the AI edge appliance mentioned above can enable engineers to implement AI-based systems quickly.

“The VVDN system can handle eight encoded input streams, decode them and perform AI inference on all of these streams,” he says. “An AI developer just has to port their algorithm—their secret sauce—to the device and they are ready to go. No further hardware development required.”


The VVDN Edge AI system can decode eight input streams and perform AI inference on them. Developers can input their own algorithms or use pre-defined AI models from the Kria app store.

The VVDN Edge AI system can decode eight input streams and perform AI inference on them. Developers can input their own algorithms or use pre-defined AI models from the Kria app store.


To make things even easier, Bhaaskar says AI developers can use AI models from the App Store. “AMD and our partner Model Zoo covers all the popular AI models in the smart city and retail spaces,” he says. “Those can be used with a lower level of expertise, again without further hardware development needed.”


A New Era of Faster Development?

In electronics technology, the pressures of time-to-market have only become more acute over time. There was a time that engineering complexity automatically meant living with long design cycles.

Programs like AMD’s Kria ODM Partner Ecosystem are perhaps a sign that those days are past, and that engineers can leverage sophisticated FPGA-based hardware and the software that runs on them, without doing every piece themselves.


All images used courtesy of AMD