Lattice Blog


sensAI 3.0 Now Available!

sensAI 3.0 Now Available!
Posted 05/21/2020 by Hussein Osman

Posted in

According to ABI Research, the market for Edge devices that support some form of AI inferencing will hit U.S. $71 billion by 2024. As device OEMs work to develop new high-tech products to capitalize on the rapidly emerging Edge computing trend, they face a host of challenges around how to properly handle the data that AI inferencing applications require (challenges like data latency, privacy, security, and the bandwidth required to move data from the Edge to the Cloud for analysis).

The range of Edge applications this affects is astonishing – from drones to smart homes, automobiles to automated factories, and everything in between. And each application has different performance and power consumption requirements. While a battery-powered occupancy sensor requires relatively low performance, keeping power consumption to an absolute minimum is vital. At the other end of the scale, the ADAS system in a modern car needs to process huge amounts of information in real time, so performance and reliability are critical.

Power/Performance Requirements Vary at The Edge

Because of their low power, small form factor, and design flexibility, Lattice FPGAs are increasingly the hardware platform of choice for developers of all types of Edge AI applications. Of course, developers need ready-to-use software, too, which is why the combination of Lattice Semiconductor’s low power FPGA families and the Lattice sensAI™ solutions stack make such a compelling design platform for Edge AI applications.

But don’t just take my word for it. Take a look at sensAI 3.0 (the latest version of our award-winning solutions stack) running on Lattice’s newest FPGA, the CrossLink-NX™. The first devices created on the 28 nm, FD-SOI-based Lattice Nexus™ FPGA development platform, CrossLink-NX FPGAs were specifically designed for Edge AI and smart vision applications. For example, CrossLink-NX FPGAs support a high-speed MIPI D-PHY interface capable of up to 2.5 Gbps of bandwidth. In fact, they’re currently the only low-power FPGA on the market capable of delivering this level of MIPI performance. MIPI, of course, is the I/O standard of choice for many components (sensors, application processors, etc.) used in smart vision systems. CrossLink-NX FPGAs’ also deliver excellent smart vision performance thanks to a special version of the neural network IP included in the version 3.0 of the sensAI stack. The IP was engineered to take advantage of the parallel processing architecture of FPGAs in general, as well as the ample on-chip RAM resources and optimized DSP blocks featured on our CrossLink-NX FPGAs. Thanks to this optimized IP and the new features of the CrossLink-NX family, the sensAI stack now delivers twice the performance at half the power when compared to prior releases.

That’s a bold claim, but one that can be substantiated by measuring real world application performance. The sensAI stack includes an available object-counting demo, as object counting is a common application for smart vision systems of all types. Running on a CrossLink-NX FPGA, the demo can deliver 10 frames-per-second (FPS) while consuming only 200 mW.

sensAI 3.0 and CrossLink-NX: Object Counting Application Performance

With the enhancements in version 3.0 of the sensAI stack, Lattice is widening the range of power and performance options available to customers looking to enable Edge AI and smart vision support on their Edge devices.