Lattice Blog

Our system is going under maintenance starting May 25, 2024 at 7:00 AM Pacific and ending May 25, 2024 at 1:00 PM Pacific. During this window, the website may not be reachable. For immediate assistance, please contact


Lattice CrossLink-NX FPGAs: Enabling Embedded Vision and AI to the Edge

CrossLink-NX Blog
Posted 12/10/2019 by PJ Chiang

Posted in

There is growing interest among embedded vision developers in adding AI/ML technology to their designs to give them the intelligence needed to run applications like object counting or presence detection. However, supporting AI/ML in embedded vision applications can be challenging, particularly as embedded vision applications trend towards increasing complexity. For example, to boost the accuracy of their AI/ML results, more sensors and higher resolution/faster frame rate cameras are being added to embedded vision systems. At the same time, embedded vision designers are looking to use components compliant with the MIPI standard. Originally developed for the mobile market, developers across a growing array of applications are looking for ways to capitalize on the high performance and economies of scale afforded by MIPI components.

CrossLink-NX Blog Trends

Figure 1: As embedded vision systems grow in complexity, the need for hardware to support a specific set of performance characteristics is increasing.

Industry analyst Pat Moorhead of Moor Insights & Strategy has been following the AI/ML trend, and sums up the problems thusly:

“Technology trends like 5G connectivity, cloud-based analytics, factory automation and the smart home are driving demand for embedded vision solutions that support machine learning. However, the data latency, cost and privacy issues associated with cloud-based ML analytics have sparked interest among developers in moving more data processing from the cloud to the Edge. But doing so requires OEMs have access to Edge AI/ML inferencing solutions that offer high performance data processing, low power operation, and a small form factor.”

To help developers add AI/ML support to new and existing Edge device designs, Lattice is excited to announce CrossLink-NX™, the first family of FPGAs implemented on our new Lattice Nexus™ FPGA platform. CrossLink-NX FPGAs provide the energy efficiency, small form factor, reliability, and high performance necessary to create innovative embedded vision solutions for the Edge in many of Lattice’s target markets: industrial, automotive, computing, mobile, and surveillance and security. Not only does CrossLink-NX enable Edge computing, it also redefines expectations for the kind of performance a small FPGA can provide, outperforming similar class competitive devices in the process.

CrossLink-NX Blog Block Diagram

Figure 2: CrossLink-NX FPGAs combine a low power, high performance FPGA fabric with fast I/Os to provide AI/ML capabilities for embedded vision applications.

While there are many features of CrossLink-NX that accelerate AI/ML performance in embedded vision, for the purposes of this blog let’s focus on two: a new FPGA fabric and CrossLink-NX’s I/O support.

Redesigned by Lattice and manufactured on the Lattice Nexus FPGA platform, CrossLink-NX’s FPGA fabric provides up to 40K logic cells, lots of memory (170 bits of memory per cell, the highest memory-to-logic ratio in comparison to similar FPGAs) and DSP blocks for data processing. This gives the CrossLink-NX the processing horsepower needed to handle AI/ML inferencing at the Edge device without having to upload data to the cloud for analysis. By handling AI/ML processing needs on chip, developers can address the data latency, cost and privacy issues described above.

CrossLink-NX also offers more robust I/O support than similar competitive devices currently on the market. To help developers connect CrossLink-NX to other MIPI components in an embedded vision system, CrossLink-NX supports up to eight MIPI D-PHY lanes at 2.5 Gbps in hardware. Supplement that with up to an additional 12 MIPI D-PHY ports in programmable I/Os, and CrossLink-NX should have no issues with data bottlenecks as data moves on and off the chip. CrossLink-NX’s I/Os also configure themselves in less than 3 ms and the entire device is configured in 15 ms; this is crucial for mission critical embedded vision applications (industrial and automotive, for example) where devices must be ready in an instant to take autonomous action and keep users and property safe.

The performance claims above and below are based on extensive in-house competitive benchmark testing between CrossLink-NX and two other mainstream FPGAs of comparable logic density.

  • CrossLink-NX provides up to a 75 percent reduction in power consumption compared to competing devices of a similar class.. With power this low, even battery-powered Edge devices can process AI/ML data locally, rather than upload it to the cloud for analysis.
  • Depending on the specific configuration, CrossLink-NX FPGAs are up to 90 percent smaller than similar competitive devices. This allows developers to add AI/ML processing capabilities to a new or existing design without having to increase the size of overall design footprint.
  • Thanks to physical characteristics of its circuits and onboard ECC and SEC support, CrossLink-NX FPGAs provide extremely high reliability. Specifically, CrossLink-NX FPGAs are 100 times more resistant to soft errors (a phenomenon caused by high energy particles striking transistors in an FPGA and disrupting performance) than similar competitive devices.
  • Available in package sizes as small as 3.7 mm x 4.1 mm, CrossLink-NX is up to 90 percent smaller than similar competitive devices, making it easy to integrate the device into new or existing embedded vision systems.

CrossLink-NX Blog Family Table

Figure 3: The Lattice CrossLink-NX family is available in two logic densities (7K or 40K logic cells) and a variety of packages.

Learn more