Lattice Semiconductor announced several enhancements to its award-winning Lattice sensAI solution stack to accelerate the development of AI/ML applications on Lattice's low-power FPGAs: support for the Lattice Propel design environment for embedded processor-based development tasks and the framework using TensorFlow deep learning techniques for on-device inference.
The new version includes the Lattice sensAI Studio design environment for complete training, validation and compilation of ML models.
Through sensAI 4.0, developers can use a simple interface using drag-and-drop operations to create FPGA-based systems using a RISC-V processor and a CNN acceleration engine to enable fast and easy deployment of ML applications on devices at the 'edge' of the network subject to power consumption constraints.
There is a growing demand in various end markets to add support for low-power IA/ML inference for applications such as object detection and classification.
IA/ML models can be trained to support applications for a range of devices that require low-power operation at the edge: control and security cameras, industrial robots, toys and consumer robotics.
The sensAI Lattice solution stack helps developers to rapidly create AI/ML applications that run on flexible, low-power Lattice FPGAs.
The following are refinements made to Lattice's 4.0 sensAI solution stack:
- TensorFlow Lite - support for the framework reduces power consumption and improves co-processing performance in AI/ML inference applications. TensorFlow Lite runs 2 to 10 times faster on a Lattice FPGA than on a Cortex-M4-based ARM microcontroller.
- Lattice Propel - the stack supports the graphical user interface of the Propel environment and command line tools to create, analyze, compile and debug both hardware and software designs of an FPGA-based processor system. Even developers unfamiliar with FPGA-based design can use the tool's intuitive drag-and-drop interface to create IA/ML applications on low-power Lattice FPGAs with support for RISC-V-based co-processing.
- Lattice sensAI Studio - a tool that features a graphical user interface for training, validating and compiling ML models optimised for Lattice FPGAs, and allows transfer learning techniques to be easily used to implement ML models.
- Improved performance - capitalising on advances in ML model pruning and compression, sensAI 4.0 can support image processing at 60 or 30 fps in QVGA or VGA resolution, respectively.
Comments
"Lattice's low-power FPGAs designed for the sensAI solution stack and embedded machine vision systems for AI/ML applications at the edge play an essential role in helping us bring cutting-edge intelligent IoT devices to market quickly and efficiently," explains Hideto Kotani, Unit Executive at Canon Inc.
"With support for TensorFlow Lite and Lattice's new sensAI Studio, it is now easier than ever for developers to use our sensAI stack to create AI/ML applications that run on battery-powered edge devices," adds Hussein Osman, Marketing Director at Lattice.
For further information click here
