Skip to content

Nios® V/g TinyML LiteRT for Microcontroller Design

Introduction

Nios® V/g TinyML LiteRT Example Design Overview

This design demonstrates the TinyML application using LiteRT for microcontrollers software with Nios® V/g processor in the Agilex® 7 FPGA F-Series Development Kit P-Tile and E-Tile DK-DEV-AGF014EA.

Prerequisites

  • Agilex® 7 FPGA F-Series Development Kit, ordering code DK-DEV-AGF014EA. Refer to the board documentation for more information about the development kit.
  • Mini and Micro USB Cable. Included with the development kit.

Release Contents

Binaries

  • Prebuilt binaries are located here.
  • The sof and elf files required to run the design can be found in "ready_to_test" folder
  • Program the sof and download the elf file on board

Nios® V/g TinyML LiteRT Design Archiecture

This example design includes a Nios® V processor connected to the On Chip RAM-II, JTAG UART IP and System ID peripheral core. The objective of the design is to accomplish data transfer between the processor and soft IP peripherals.

Block Diagram

Nios® V/g Processor

  • General-Purpose Processor- High Performance (For interrupt driven baremetal and RTOS code)
  • Nios® V/g processor is highly customizable and can be tailored to meet specific application requirements, providing flexibility and scalability in embedded system designs.

IP Cores

The following IPs are used in this Platform Designer component of the design: - Nios® V/g soft processor core

  • On Chip RAM-II
  • JTAG UART
  • System ID
  • Clock Bridge, Reset Controller

Hardware Setup

Refer to Agilex™ 5 FPGA E-Series 065B Premium Development Kit User Guide to setup the hardware connection.

Address Map Details

Nios V Address Map

Address Offset Size (Bytes) Peripheral Description
0x0 12M On-Chip RAM To store application
0x0101_0048 8 JTAG UART Communication between a host PC and the Nios V processor system
0x0101_0040 8 System ID Hardware configuration system ID (0x00000009)

User Flow

There are two ways to test the design based on use case.

User Flow 1: Testing with Prebuild Binaries.
User Flow 2: Testing Complete Flow.
User Flow Description Required for User flow 1 Required for User flow 2
Environment Setup Tools Download and Installation Yes Yes
Compilation Hardware compilation No Yes
Software compilation No Yes
Programing Program Hardware Binary SOF Yes Yes
Program Software Image ELF Yes Yes
Testing Open JTAG UART Terminal Yes Yes

Environment Setup

Tools Download and Installation

  1. Quartus Prime Pro
  • Download the Quartus® Prime Pro Edition software version 25.1 from the FPGA Software Download Center webpage of the Intel website. Follow the on-screen instructions to complete the installation process. Choose an installation directory that is relative to the Quartus® Prime Pro Edition software installation directory.
  • Set up the Quartus tools in the PATH, so they are accessible without full path. 
    export QUARTUS_ROOTDIR=~/intelFPGA_pro/25.1/quartus/
export PATH=$QUARTUS_ROOTDIR/bin:$QUARTUS_ROOTDIR/linux64:$QUARTUS_ROOTDIR/../qsys/bin:$PATH

Compilation

Hardware Compilation

  • Invoke the quartus_py shell in the terminal
  • Run the following command in the terminal from top level project directory:
quartus_py ./scripts/build_sof.py
  • The quartus tool will compile the design and generate the output files

Software Compilation

  • To create software app, run the following commands in the terminal: 
    niosv-bsp -c --quartus-project=hw/niosv_tinyml.qpf --qsys=hw/sys.qsys --type=hal --bsp-dir=sw/tflite_bsp --script=sw/bsp_script.tcl sw/tflite_bsp/settings.bsp
niosv-app -b=sw/tflite_bsp -a=sw/tflite_app -S=sw/tflite_app/image_classification/,sw/tflite_app/image_classification/model/,sw/tflite_app/signal,sw/tflite_app/tensorflow,sw/tflite_app/tensorflow/lite/,sw/tflite_app/tensorflow/lite/c/,sw/tflite_app/tensorflow/lite/core/api/,sw/tflite_app/tensorflow/lite/kernels/,sw/tflite_app/tensorflow/lite/kernels/internal,sw/tflite_app/tensorflow/lite/kernels/internal/reference/,sw/tflite_app/tensorflow/lite/kernels/internal/reference/integer_ops,sw/tflite_app/tensorflow/lite/micro,sw/tflite_app/tensorflow/lite/micro/kernels/,sw/tflite_app/tensorflow/lite/micro/memory_planner,sw/tflite_app/tensorflow/lite/schema --incs=sw/tflite_app,sw/tflite_app/image_classification,sw/tflite_app/image_classification/model/,sw/tflite_app/tensorflow,sw/tflite_app/tensorflow/lite,sw/tflite_app/third_party/flatbuffers/include,sw/tflite_app/third_party/gemmlowp,sw/tflite_app/third_party/kissfft,sw/tflite_app/third_party/ruy
cmake -S sw/tflite_app -B sw/tflite_app/build/Release -G "Unix Makefiles" -DCMAKE_BUILD_TYPE=Release
make -C sw/tflite_app/build/Release
    Note:The software can be compiled using the Ashling Visual Studio Code Extension for Altera FPGAs

For information on the build process, please refer to the following document- Ashling VSCode Extension

Programing

Note: Reduce the JTAG clock frequency to 6MHz using the following command, before programming the sof file 

jtagconfig --setparam 1 JtagClock 6M

Program Hardware Binary SOF

  • Program the generated sof and then download the elf file on the board
quartus_pgm --cable=1 -m jtag -o 'p;ready_to_test/niosv_tinyml.sof'

Program Software Image ELF

  • Download the elf file on the board 
    niosv-download -g ready_to_test/tflite_app.elf -c 1

Testing

Open JTAG UART Terminal

  • Verify the output on the terminal by using the following command in the terminal:
juart-terminal -c 1 -i 0
Last update: June 21, 2025
Created: August 7, 2024
Ask in the Forum