Lattice sensAI Stack

Accelerate Integration of Flexible, Low Power Inferencing at the Edge

The full-featured Lattice sensAI stack includes everything you need to evaluate, develop and deploy FPGA-based Machine Learning / Artificial Intelligence solutions - modular hardware platforms, example demonstrations, reference designs, neural network IP cores, software tools for development, and custom design services.

The Lattice’s FPGA-based machine learning solutions are highly flexible, easy to implement, low power (from under 1 mW-1 W), small (package sizes starting at 5.5 mm2) and priced for high volume production.

Lattice can help you accelerate deployment of always-on, on-device AI into a wide range of Edge applications including mobile, smart home, smart city, smart factory, and smart car products.

Lattice sensAI Solution Stack

Jump To

Video

Lattice FAD sensAIExpand Image

Lattice sensAI: accelerating low power AI at the edge

The full-featured Lattice sensAI stack includes everything developers need to evaluate, develop and deploy FPGA-based Machine Learning / Artificial Intelligence solutions - modular hardware platforms, example demonstrations, reference designs, neural network IP cores, software tools for development, and custom design services.

Design Methodology Overview

sensAI Block Diagram

  • End to end solution stack is provided enabling fast time to market
  • NN Models supported are pruned to fit the resources available
  • Training can be done using standard Machine Learning frameworks such as Tensorflow on a GPU machine
  • Sample dataset with instruction on how to tag the dataset is provided

System Architecture Examples

The following diagrams are examples of how FPGAs are commonly used to power AI/ML solutions. Lattice FPGAs give you a powerful and scalable AI/ML solution that’s simple, low-power and small-footprint. Scroll down to see hardware demonstrations you can download and evaluate today.

Preprocessor: CrossLink-NX AI Accelerator

  • Optimized FPGA Architecture built on low power process (FDSOI)
  • Flexible interface to connect to high quality image sensors through MIPI CSI-2
  • Ability to pass video data through CSI-2 or PCIe with AI acceleration meta data

Stand-alone: ECP5/iCE40 UltraPlus FPGA AI enabled System

  • Always-on, integrated solutions on ECP5 or iCE40 UltraPlus FPGA
  • Low latency and secure implementation
  • FPGA resources can be used for system control

Preprocessor: iCE40 UltraPlus Event Trigger

  • iCE40 UltraPlus FPGA for always-on detection of key-phrases or objects
  • Wakes-up a high performance ASIC/ASSP for further analytics only when required
  • Reduces overall system power consumption

Preprocessor: ECP5 FPGA AI processor

  • Scalable performance/power with ECP5 based neural network acceleration
  • ECP5 based I/O flexibility to seamlessly interface to on-board legacy devices including sensors
  • Low-end MCU for flexible system control

Postprocessor: ECP/iCE40 UltraPlus AI Accelerator

  • Add AI acceleration without significant modification of the system
  • Scalable performance/power with ECP or iCE40 UltraPlus
  • ASIC/ASSP can handle data preprocessing and pass on inference data only

Reference Designs

  • Key Phrase Detection

    Reference Design

    Key Phrase Detection

    Continuous searches for a key phrase utterance via a digital MEMS microphone. Can be re-configured to work with any trained word or phrase.
    Key Phrase Detection
  • Human Face Identification

    Reference Design

    Human Face Identification

    Uses a Convolutional Neural Network in the ECP5 FPGA to detect a human face, and match to known registered faces. Can be adapted to work with any other object.
    Human Face Identification
  • Human Presence Detection

    Reference Design

    Human Presence Detection

    Uses Lattice sensAI IP to continuously search for the presence of a human and reports results. Can be adapted to detect any other object.
    Human Presence Detection
  • Object Counting

    Reference Design

    Object Counting

    An example object counting application based on the Lattice sensAI stack. Includes SPI, DDR IP blocks, ISP engine, 8 CNN engines and a counting / overlay engine
    Object Counting
  • IP Cores

  • CNN Accelerator IP

    IP Core

    CNN Accelerator IP

    Implement AI solutions with CNNs from common or custom networks. Configure up to 16-bit widths. Works with Lattice Neural Network Compiler software tool.
    CNN Accelerator IP
  • CNN Compact Accelerator IP

    IP Core

    CNN Compact Accelerator IP

    Implement AI solutions with CNNs or BNNs that have power consumption in the mW range. Works with Lattice Neural Network Compiler software tool.
    CNN Compact Accelerator IP
  • CNN Plus Accelerator IP

    IP Core

    CNN Plus Accelerator IP

    Implement Ultra-Low Power AI solutions with CNNs. Configure up to 16-bit widths. Works with Lattice Neural Network Compiler software tool.
    CNN Plus Accelerator IP
  • Development Kits & Boards

  • Embedded Vision Development Kit

    Board

    Embedded Vision Development Kit

    Three-board Video Interface Platform (VIP) modular kit. Includes 2 MIPI/D-PHY camera input, FPGA processing, HDMI output. More Input/Output boards available.
    Embedded Vision Development Kit
  • CrossLink-NX VIP Sensor Input Board

    Board

    CrossLink-NX VIP Sensor Input Board

    Intended for use with the Lattice Embedded Vision Development Kit. Includes 4 camera inputs and HyperRam for embedded video applications with CrossLink-NX
    CrossLink-NX VIP Sensor Input Board
  • HM01B0 UPduino Shield

    Board

    HM01B0 UPduino Shield

    A complete development kit for implementing Artificial Intelligence (AI) using vision and sound as sensory inputs to a low-cost, low-power iCE40 UltraPlus FPGA.
    HM01B0 UPduino Shield
  • DPControl iCEVision Board

    Board

    DPControl iCEVision Board

    A highly flexible and expandable prototyping platform for vision based AI and many other applictions
    DPControl iCEVision Board
  • Machine Learning Adapter Card

    Board

    Machine Learning Adapter Card

    Mass Storage Adapter and Microphone for VIP ecosystem
    Machine Learning Adapter Card
  • Demos

  • Key Phrase Detection

    Demo

    Key Phrase Detection

    Uses artificial intelligence (AI) to detect a specific key-phrase using a tiny, low-power iCE40 UltraPlus FPGA
    Key Phrase Detection
  • Hand Gesture Detection

    Demo

    Hand Gesture Detection

    Uses artificial intelligence (AI) to implement hand gesture detection algorithm using a tiny, low-power iCE40 UltraPlus FPGA
    Hand Gesture Detection
  • Human Counting

    Demo

    Human Counting

    Human upper-body detection and counting demonstration utilizes Lattice’s ECP5 FPGA and a Convolutional Neural Network (CNN) acceleration engine
    Human Counting
  • Human Face Detection

    Demo

    Human Face Detection

    Uses Lattice sensAI IP to detect human faces on a tiny, low-power iCE40 UltraPlus FPGA implementing AI at the edge. Adaptable to detect other objects.
    Human Face Detection
  • Human Face Identification

    Demo

    Human Face Identification

    Register and identify faces without retraining, eliminating the need for uploading images and lengthy retraining using a GPU.
    Human Face Identification
  • Software Tools

    The Lattice sensAI solution stack includes the Neural Network Compiler for easy integration of networks developed in TensorFlow,Caffe or Keras into Lattice FPGAs - no prior RTL experience required.

    The output of the Neural Network Compiler can then be integrated into the standard Lattice FPGA development tool flows - Lattice Diamond for ECP5, or Radiant for iCE40 UltraPlus, and finally deployed to the FPGA in your system.

    • Neural Network Compiler – Rapidly analyze, simulate, and compile various networks for implementation onto Lattice CNN/CNN Compact Accelerator IP cores.
    • Lattice Radiant Software – FPGA design tool for design creation, importing IP, implementation, bitstream generation, downloading the bitstream onto an FPGA. Supports iCE40 UltraPlus device.
    • Lattice Diamond Software – FPGA design and exploration tool targeted for ECP5 family of FPGAs.

    Custom Design Services

    The Lattice sensAI stack includes an ecosystem of select, worldwide design service partners that can deliver custom solutions for a range of end-applications, including mobile, smart home, smart city, smart factory, and smart cars. Click here for more information on Lattice sensAI certified partners.

    Documentation

    Information Resources
    TITLE NUMBER VERSION DATE FORMAT SIZE
    Lattice sensAI Stack Product Brochure
    I0265 7.0 5/20/2020 PDF 1.1 MB
    TITLE NUMBER VERSION DATE FORMAT SIZE
    Accelerating Implementation of Low Power Artificial Intelligence at the Edge
    WP-0014 1.1 11/28/2018 PDF 1.2 MB
    Rising Edge AI Requirements Demand Higher Performance Solution
    WP-0019 1.0 8/17/2019 PDF 1.1 MB
    Harnessing the Power of AI: An Easy Start with Lattice’s sensAI
    WP-0017 1.0 1/31/2019 PDF 2.3 MB


    Awards

    The Electronics Industry Awards 2019

    Internet of Things Product of the Year

    EDN Hot 100 Product Award

    Tools & Development

    AI Breakthrough Award

    Best AI-Based Solution for Engineering

    Assodel Award 2019

    Best Industrial and Automotive Related Hardware Product

    China Electronic Market 2018 Editor's Choice Awards

    Most Competitive FPGA Product

    Like most websites, we use cookies and similar technologies to enhance your user experience. We also allow third parties to place cookies on our website. By continuing to use this website you consent to the use of cookies as described in our Cookie Policy.