Lattice Semiconductor

Lattice Semiconductor sensAI™ 3.0 Solution Stack

Doubles performance and halves power consumption in edge AI applications

The latest version of Lattice Semiconductor’s complete solutions stack for on-device Artificial Intelligence (AI) processing at the edge, Lattice sensAI™ 3.0, has added support for the CrossLink-NX™ family of FPGAs for low-power smart vision applications. The Lattice AI stack also now includes Convolutional Neural Network (CNN) accelerator IP embedded in the CrossLink-NX FPGA which simplifies implementation of common CNN networks, and which takes advantage of the parallel-processing capabilities of FPGAs.

 

By enhancing the sensAI stack, Lattice has widened the range of power and performance options available to OEMs to implement AI at the edge. For applications such as smart vision and object recognition, which require high edge AI performance, CrossLink-NX FPGAs running sensAI software provide twice the performance at half the power when compared to prior releases of the stack.

 

Most edge devices are battery-powered or otherwise sensitive to power consumption. The sensAI 3.0 stack enables developers to specify the processing capabilities needed for AI applications, while keeping power consumption as low as possible. In addition, CrossLink-NX FPGAs are fabricated in a 28nm process which offers a 75% reduction in power consumption compared to that of competing bulk CMOS FPGAs in the same class.

 

The sensAI software stack is backed by new, ready-to-use demonstrations and reference designs for popular AI applications. These include a CrossLink-NX-based object-counting demonstration using the VGG neural networking algorithm. Running on a CrossLink-NX FPGA, it scans at a rate of 10 frames/s while consuming only 200mW.

 

Other new and updated features of the sensAI solutions stack include:

  • Optimized FPGA architecture for CrossLink-NX – when running on a CrossLink-NX FPGA, the sensAI solutions stack offers up to 2.5Mbits of distributed memory and block RAM and additional DSP resources, providing for efficient on-chip implementation of AI workloads and reducing the need for cloud-based analytics.
  • MIPI I/O speeds of up to 2.5Gbits/s on CrossLink-NX FPGAs – many smart vision systems require support for the MIPI I/O standard, and CrossLink-NX devices are currently the only low-power FPGAs to deliver MIPI I/O speeds of up to 2.5Gbits/s. This makes sensAI IP and software running on a CrossLink-NX FPGA an ideal platform for smart vision applications requiring fast MIPI performance at low power.
  • Support for the MobileNet v2, SSD, and ResNet models on the Lattice ECP5™ family of general-purpose FPGAs. Earlier versions of sensAI running on ECP5 already supported the VGG and MobileNet v1 neural network models.

Features

  • Reference designs available for:
    • Hand gesture detection
    • Key phrase detection
    • Human face identification
    • Presence detection
    • Object counting
  • Model training can be performed in standard machine learning frameworks such as Tensorflow
  • Sample dataset with instruction on how to tag the dataset is provided

Applications

  • Surveillance and security equipment
  • Robotics
  • Automotive
  • Computing

Evaluation Kit

Board part number: LIFCL-40-EVN

 

The CrossLink-NX evaluation board includes a USB-B connection for device programming and an I2C utility, 128Mbits of SPI boot Flash, eight input DIP switches, four push buttons, three status LEDs, 14 LEDs for demonstration purposes, and multiple reference clock sources.

Share This

Share on facebook
Share on twitter
Share on linkedin

Subscribe to our newsletters

Subscribe to Future Electronics

Get access to the latest product information, technical analysis, design notes and more

Choose your region

FTM Board Club

Sign up for access to exclusive development boards, an essential tool for many innovative design projects.

*Available to pre-qualified EMEA customers only.