AVNet RASynBoard Starter Kit

Table of contents

RoadTest: Enroll to Review the Avnet RASynBoard EVK

Author: amgalbu

Creation date: 27 Dec 2023

Evaluation Type: Evaluation Boards

Did you receive all parts the manufacturer stated would be included in the package?: True

What other parts do you consider comparable to this product?: Arduino Nicla Voice

What were the biggest problems encountered?: No relevant problems

Detailed Review:

RASynBoard is a low-cost evaluation kit that includes the RASynBoard core module with an I/O board for prototyping and development.

RASynBoard is a tiny (25mm x 30mm), ultra-low power, edge AI/ML board, based on a Syntiant NDP120 Neural Decision Processor, a Renesas RA6M4 host MCU plus a power-efficient DA16600 Wi-Fi/BT combo module. The NDP120 subsystem with onboard digital microphone, IMU motion sensor and SPI Flash memory, achieves highly efficient processing of acoustic- and motion events. Battery and USB-C device connectors facilitate stand-alone use, while a compact under-board connector enables integration with custom OEM boards and additional sensors.

In this roadtest, I will go through al the steps required to build a simple gesture-recognition application from start to finish

1. Unboxing

The evaluation kit is shipped in a simple cardboard box, inside an antistatic bag. The two boards that make up the evaluation kit (aka the RASynBoard core board and the RASynBoard Starter kit) are already assembled

2. Hardware description

The evaluation kit is made of two parts: the RASynBoard core board and the RASynBoard starter kit.

The RASyn core board RASynBoard Core Board is a tiny (25mm x 30mm), ultra-low power, edge AI/ML board, based on Syntiant NDP120 Neural Decision Processor, a Renesas RA6M4 host MCU plus a power-efficient DA16600 Wi-Fi/BT combo module. The NDP120 subsystem with on-board digital microphone, IMU motion sensor and SPI Flash memory, achieves highly efficient processing of acoustic- and motion events. Battery and USB-C device connectors facilitate stand-alone use, while a compact under-board connector enables integration with custom OEM boards and additional sensors. 

Here are some pictures of the board

{gallery}RASynBoard Starter kit

And this is the block diagram of the RASyn core board

The relevant feature is this block diagram is the NDP120 manufactured by Syntiant. 

A neural decision processor (NDP) is a type of artificial intelligence (AI) hardware that is specifically designed to run deep learning models. NDPs are typically very energy-efficient and can be used to run AI applications on devices with limited power budgets, such as smartphones, wearables, and IoT devices. NDPs are typically based on a custom architecture that is optimized for running deep learning models. This architecture may include specialized hardware accelerators, such as tensor processing units (TPUs), that can perform the computations required for deep learning models very efficiently. Additionally, NDPs may also include software optimizations that help to further improve the energy efficiency of the processor.

One of the key benefits of NDPs is their ability to run AI applications on-device. This means that the AI processing is done directly on the device, rather than being offloaded to a cloud server. This can provide several benefits, such as improved privacy, reduced latency, and reduced reliance on cloud connectivity. 

NDPs are a relatively new technology, but they have already been used in a number of commercial products. For example, Syntiant's NDP100 processor is used in Amazon's Echo Dot smart speaker, and Google's Pixel Buds wireless earbuds use a custom NDP from Google AI.

Some of the key features of NDPs are:

• Ultra-low power consumption: NDPs are designed to consume very little power, making them ideal for use in battery-powered devices.
• High performance: NDPs can provide high performance for running deep learning models, even on devices with limited processing power.
• Small size: NDPs are typically very small and can be easily integrated into existing devices
• Low cost: NDPs are becoming increasingly cost-effective as the technology matures.

RASynBoard is equipped with the NDP120, which is one of the most powerful processor in the Syntiant portfolio

The RASynBoard Starter Kit adds an IO Board (50mm x 30mm), for a versatile, compact, two-board evaluation kit assembly. This pins-out a subset of the NDP120 and RA6M4 I/Os to popular Pmod, Click header and expansion header footprints, enabling connection with additional external microphones and sensor options. An onboard debugger MCU (SWD and UART interfaces), button switches, RGB LED and removable MicroSD storage, further maximize the prototyping versatility and utility.

This board adds multiple connectors to add external peripherals, namely

• up to two external microphones
• access to main MCU's GPIOs
• access to NDP120's serial port and SPI interface
• SD card to store NDP120 configuration files and records from IMU and microphones for later processing on the Edge Impluse platform
• multicolor LED
• USB debugger
• main MCU's serial terminal

3. Software installation

RASynBoard is programmed from the E^2Studio, an Eclipse-based IDE that is available for download from this link

https://www.renesas.com/us/en/software-tool/e-studio

The installation is straightforward. I added an image gallery with the steps just for reference

{gallery}E^2Studio setup

4. Build the project

A fully-functional project for the RASynBoard is available on github

https://github.com/Avnet/RASynBoard-Out-of-Box-Demo/tree/rasynboard_v2_tiny

4.1 Clone the project

The first step, is to clone the project

git clone https://github.com/Avnet/RASynBoard-Out-of-Box-Demo.git

This command will clone the repository in a folder named RASynBoard-Out-of-Box-Demo.

4.2 Configure the project

We can now open and configure the project in E^2Studio. Steps to go trough this task are explained in gallery below

4.3 Build the project

Click "Project" -> "Build all" to build the project

The project builds and generates output files in the Debug Directory. Two files we're interested in are:

• rasynboard_ew_demo.elf This is the file we'll load with the debugger
• reaynboard_ew_demo.srec This is the file we can distribute and side-load onto RASynBoard hardware using the Renesas Flash Programmer application

5. Setup the hardware

6. Copy NDP files to SD card

The microSD card on the IO Board is used to store the NDP120 images and a config.ini file that we can use to change the behavior of the OOB Demo application

• Copy all the files from the project directory <Project dir>\ndp120\synpkg_files directory to the root directory of your microSD card

• Insert the microSD card into the RASynBoard IO Board microSD cage on the bottom of the board

7. Debug the project

8. Tinkering with the board

The OOB demo application is quite complete and shows many (if not all) the relevant features of this board. Now I want to go deeper into how things works. In particular, I want to train NDP to recognize specific gestures. I supposed the steps to nail are

• collect IMU data
• transfer collected data to Edge Impulse cloud platform and train the model
• install the model on the RASynBoard
• test the model

According to the documentation, there two ways to collect data either from IMU or from microphone. Data can be stored on the SD card or can be transferred via serial terminal to a PC. Let's see both solutions

8.1 Create a Edge Impulse account

First things first: before proceeding, we need to create an account on Edge Impulse. The procedure is extremely easy, so I will not include details here.

8.2 Collect IMU data on SD card

To collect IMU data on SD card, we need first to change config.ini stored on SD card. The "Write_to_file" option needs to be set to 1

This will instruct the application to store IMU data for the number of seconds configured in the "Recording_Period" option after the user button is pressed.

Insert the SD card, connect the USB-C cable, download the application from E2Studio and, when ready, press the user button. Now the gesture can be simulated. After the configured amount of time the collected data is streamed to the SD card. The procedure can be repeat as many time as one want to collect a certain number of samples.

Now disconnect the USB-C cable to power off the board, remove the SD card and insert the SD card itself in the PC. A certain number of files are present 

8.3 Collect IMU data through Edge Impulse Forwarder

There is actually another way to collect data: by means of a tool called "Edge Impulse Forwarder". This is a NodeJS application that reads IMU data from RASyn board through the serial port and immediately forward to the Edge Impulse portal. I tried this approach but, despite it looks a convenient way to train a model, actually it does not allow you to "pre-filter" your data and eliminate, for example, wrong gestures or pauses between samples. This spurious data may interfere with the model training, so I preferred to follow the approach based on the SD card.

8.4 Train the model

To train the model, we need to upload data collected on the Edge Impulse portal. The following gallery shows the relevant steps to accomplish this task

8.5 Deploy the model

The synpkg file can now be copied to the SD card. I simply changed the config.ini file to use the new synpkg file and everything worked smoothly: LED0 was switched on when "left" was detected and LED1 was switched on when "right" was detected. Very nice!

9. Comparison with Arduino Nicla Voice

There is actually a board in the Arduino family that features the same neural processor installed on this board: the Arduino Nicla Voice. Let's compare the two alternatives

	Arduino Nicla Voice	RasynBoard
Neural processor	Syntiant® NDP120 Neural Decision Processor (NDP): 1x Syntiant Core 2 ultra-low-power deep neural network inference engine 1x HiFi 3 Audio DSP 1x Arm® Cortex® M0 core up to 48 MHz	Syntiant® NDP120 Neural Decision Processor (NDP): 1x Syntiant Core 2 ultra-low-power deep neural network inference engine 1x HiFi 3 Audio DSP 1x Arm® Cortex® M0 core up to 48 MHz
Main processor	Nordic Semiconductor nRF52832: 64 MHz Arm® Cortex M4	Renesas RA6M4 processor: 200 MHz Arm® Cortex M4
Sensors	High performance microphone (IM69D130) 6-Axis IMU (BMI270) 3-axis magnetometer (BMM150)	IMU 6-axis motion sensor (ICM42670) PDM digital microphone (MMICT5838)
I/O	Castellated pins with the following features: 1x I2C bus (with ESLOV connector) 1x serial port 1x SPI 2x ADC Programmable I/O voltage from 1.8-3.3V	On the core board 2x28 pin board-to-board connector , On the Starter kit 2x28 pin board-to-board connector 2x8 pin MikroE Click shuttle box header 2x6 pin Pmod type-6A (I2C) socket 2x7 pin MCU expansion header 2x3 pin DMIC expansion headers (two) 3.3V level-translation of expansion interfaces Micro SD card cage for removable storage
Interface	External microphone connector (ZIF) USB interface with debug functionality	On the core board USB 2.0 device interface On the Starter kit Renesas E2 OB debugger MCU (USB-C to SWD and VCOM serial interface) 3.3V buck regulator for debugger circuits
Memory	512KB Flash / 64KB SRAM  16MB SPI Flash for storage 48KB SRAM dedicated for NDP120	1 MB flash memory 16 MBit SPI NOR Flash 256 KB SRAM
Power	High speed USB (500mbps) Pin Header 3.7V Li-po battery with Integrated battery charger and fuel gauge (BQ25120AYFPR)	USB port LiPo battery management and connector
Connectivity	Bluetooth® Low Energy (ANNA-B112)	802.11bgn 1x1 2.4 GHz Wi-Fi BT 5.1
Price	69 Euros	Starter kit 125.57 euros Core board 70 euros?

The RASynBoard is better than Nicla board in almost all aspects (the Arduino Nicla Voice features a magnetometer, but all other hardware components are better on the RASynBoard -memory, CPU frequency, etcetera). In my opinion, where the RASynBoard really shine is connectivity: BT and Wifi on such a small board is definitely a plus.

9. Conclusions

To summarize this roadtest, I am going to make a list with three things I like and three things I didn't like

9.1 What I liked

• The integration in the Edge Impulse platform: once you have created your model, create a synpkg file ready to be deployed is just a matter of a few clicks
• The out-of-the-box experience. According to my experience, it's not so common to get an evaluation board and be provided with a example application that shows all the relevant features of the platform. The demo project available on github lets you easily and quickly test the capabilities of the NDP processor. You can tinker with the board by simply editing files on SD card, and this is another thing I appreciated
• The NDP library: interacting with the NDP120 hardware is as easy as calling two functions: all the boilerplate is there and ready to be reused in your next project

9.2 What I didn't like

• Why do  need two cables? To develop on this board, you need to connect two USB cables: one to power and debug, the other one to get console output from the application. Today, as far as I can see, a single USB connection can provide both the endpoints (debugger and serial monitor) 
• The 2x28 connector. The RASynBoard core board has a 2x28 header connector to bring signals to an auxiliary board. As a maker, I think this is bad (or at least, not so good) for two reasons
  • The 2x28 connector as a very small step, so it is a little bit difficult to solder
  • If you need, for your project, just a single GPIO, you need to create an external board. This, in some ways, nullifies the advantages of the small form factor of the RASynBoard core board
    
• github repository with the demo project and a lot of other useful documentation was not so easy to find. I would appreciate if a direct link to the github page were present on the product page on the Avnet site  

The overall experience is absolute positive: from the hardware to the software (both E^2 Studio and Edge Impulse integration), I didn't encounter any major problem and I was enable to go through all my tests with no hassle. 

Useful links

• Demo project: https://github.com/Avnet/RASynBoard-Out-of-Box-Demo
• Documentation: https://github.com/Avnet/RASynBoard-Out-of-Box-Demo/tree/rasynboard_v2_tiny/docs