“Inference with Raspberry Pi Pico and RP2040”

Eben Upton - Raspberry Pi Foundation

March 4, 2021
 tinyML Talks Sponsors

tinyML Strategic Partner

Additional Sponsorships available – contact sponsorships@tinyML.org for info

Arm: The Software and Hardware Foundation for tinyML
1 1
Connect to Application
high-level
frameworks
2
Profiling and Optimized models for embedded
2
debugging AI Ecosystem
Supported by tooling such as Partners
end-to-end tooling Arm Keil MDK 3 Runtime
(e.g. TensorFlow Lite Micro)

3 Optimized low-level NN libraries

Connect to (i.e. CMSIS-NN)
Runtime

RTOS such as Mbed OS

Stay Connected
Arm Cortex-M CPUs and microNPUs
@ArmSoftwareDevelopers

@ArmSoftwareDev
Resources: developer.arm.com/solutions/machine-learning-on-arm
33 © 2020 Arm Limited (or its affiliates)
 WE USE AI TO MAKE OTHER AI FASTER, SMALLER
AND MORE POWER EFFICIENT

Automatically compress SOTA models like MobileNet to <200KB with

little to no drop in accuracy for inference on resource-limited MCUs

Reduce model optimization trial & error from weeks to days using
Deeplite's design space exploration

Deploy more models to your device without sacrificing performance or

battery life with our easy-to-use software

BECOME BETA USER bit.ly/testdeeplite

| Confidential Presentation ©2020 Deeplite, All Rights Reserved PAGE 4
TinyML for all developers
Dataset

Acquire valuable Enrich data and train

training data securely ML algorithms

Edge Device Impulse

Real sensors in real time
Open source SDK

Embedded and edge Test impulse with

compute deployment real-time device
options data flows

Test
Copyright © EdgeImpulse Inc. Get your free account at http://edgeimpulse.com
 Maxim Integrated: Enabling Edge Intelligence
www.maximintegrated.com/ai
Sensors and Signal Conditioning Low Power Cortex M4 Micros Advanced AI Acceleration

Health sensors measure PPG and ECG signals The biggest (3MB flash and 1MB SRAM) and the The new MAX78000 implements AI inferences at
critical to understanding vital signs. Signal chain smallest (256KB flash and 96KB SRAM) Cortex over 100x lower energy than other embedded
products enable measuring even the most M4 microcontrollers enable algorithms and options. Now the edge can see and hear like
sensitive signals. neural networks to run at wearable power levels never before.
Qeexo AutoML for Embedded AI
Automated Machine Learning Platform that builds tinyML solutions for the Edge using sensor data

Key Features Target Markets/Applications

 Industrial Predictive Maintenance  Automotive
 Wide range of ML methods: GBM, XGBoost, Random
 Smart Home  Mobile
Forest, Logistic Regression, Decision Tree, SVM, CNN, RNN,
CRNN, ANN, Local Outlier Factor, and Isolation Forest  Wearables  IoT

 Easy-to-use interface for labeling, recording, validating, and QEEXO AUTOML: END-TO-END MACHINE LEARNING PLATFORM
visualizing time-series sensor data

 On-device inference optimized for low latency, low power

consumption, and a small memory footprint

 Supports Arm® Cortex™- M0 to M4 class MCUs

 Automates complex and labor-intensive processes of a

typical ML workflow – no coding or ML expertise required! For a limited time, sign up to use Qeexo AutoML at automl.qeexo.com
for FREE to bring intelligence to your devices!
 Reality AI Tools® software
Automated Feature Bill-of-Materials
Exploration and Model
Generation Optimization

is for Automated Data

Assessment
Edge AI / TinyML
code for the smallest
MCUs

building products
Reality AI solutions
Automotive sound recognition & localization

https://reality.ai info@reality.ai @SensorAI Reality AI Indoor/outdoor sound event recognition

RealityCheck™ voice anti-spoofing

SynSense builds ultra-low-power (sub-mW) sensing and
inference hardware for embedded, mobile and edge
devices. We design systems for real-time always-on smart
sensing, for audio, vision, IMUs, bio-signals and more.

https://SynSense.ai
 Next tinyML Talks
Date Presenter Topic / Title
Tuesday, Vijay Janapa Reddi tinyMLPerf: Deep Learning Benchmarks for
March 16 Associate Professor, Embedded Devices
Harvard University

Webcast start time is 8 am Pacific time

Please contact talks@tinyml.org if you are interested in presenting

 Announcement

https://www.tinyml.org/event/summit-2021/
Highlights: What should I do about it:
- Keywords: Premier Quality, Interactive, LIVE … and FREE - Check out the program – you will be impressed
- 5 days, 50+ presentations - Register on-line (takes 5 min)
- 4 Tutorials - If interested: Submit nominations for Best Awards
- 2 Panel discussions: (i) VC and (ii) tinyML toolchains and/or Late News – February 28 deadline
- tinyML Research Symposium - Block out your calendar: March 22-26
- Late Breaking News - Become a sponsor (sponsorships@tinyML.org)
- 3 Best tinyML Awards (Paper, Product, Innovation) - Actively participate at the Summit
- 10+ Breakout sessions on various topics - Provide your feedback – we listen !
- tinyML Partner sessions - Don’t worry about missing some talks – all videos
- tinyAI for (Good) Life will be posted on YouTube.com/tinyML
- LIVE coverage, starting at 8am Pacific time
 tinyML is growing fast
2019 Summit 2020 Summit 2021 Summit
(March 2019) (Feb 2020) (March 2021), expected

Attendees 160 400+ 3000+

Companies 90 172 300+ (?)

Linkedin 0 798 ~ 2000

members
Meetups 0 1140 ~ 5000
members
YouTube 0 0 ~ 3000
subscribers
also started in Asia: tinyML WeChat and BiliBili

2018 2019 2020 2021

 Summit Sponsors
(as of Feb 15, 2021)

Contact: sponsorships@tinyML.org

multiple levels and benefits available

(also check www.tinyML.org)
 Reminders
Slides & Videos will be posted tomorrow Please use the Q&A window for your questions

tinyml.org/forums youtube.com/tinyml
Eben Upton

Dr Eben Upton CBE FREng DFBCS HonFIET is a

founder of the Raspberry Pi Foundation, a former
Distinguished Engineer with fabless
semiconductor manufacturer Broadcom Inc, and
founder and former CTO of mobile games
middleware developer Idea works 3d Ltd. He
holds a BA in Physics and Engineering, a PhD in
Computer Science, and an MBA, from the
University of Cambridge.
 Raspberry Pi Pico
• Raspberry Pi Pico is a new $4 board
• Built on our RP2040 microcontroller
– “Just” a break-out board
– But with a nice power chain…
– …and 2MB of QSPI Flash
• Showing promise as an ML platform
 RP2040
• Our first in-house silicon design
• Dual Cortex-M0+ @ 133MHz
• 264KB on-die SRAM
• “Flashless” architecture
• Simple, deterministic bus fabric
• Rich peripheral set
– UART, SPI, I2C
– USB 1.1
– Programmable I/O (PIO)
• Third-party boards also available
RP2040 detail
Programmable I/O
 Early ML work
• The good keyword person detect
SparkFun Edge 800ms
– High clock rate
(Cortex-M4 @ 48MHz)
– Dual core SparkFun Edge 400ms
– Large on-chip SRAM (Cortex-M4 @ 96MHz)

• The bad Arduino BLE Sense Nano

(Cortex-M4 @ 64MHz)
600ms

– No SIMD Raspberry Pi Pico 10.2ms 2200ms

– No single-cycle MAC (Cortex-M0+ @ 125MHz;
model in Flash)
– Currently limited sensor choice
Raspberry Pi Pico 3.8ms
• Initial TensorFlow Lite port (Cortex-M0+ @ 125MHz;
model in RAM)
– Stock clocks (2×)
– Single-core (2×)
– Model parameters in SPI Flash (2.7×)
 Future directions
• ML-focused third-party boards
– SparkFun MicroMod RP2040
– Arduino Nano RP2040 Connect
– ArduCam Pico4ML
• Optimised TensorFlow Lite
– 1.2V operating point
– Dual-core support
– Streaming model parameters via DMA
• Other frameworks
• Future silicon
– Lightweight (4-8MACs/clock) accelerators
 Copyright Notice
This presentation in this publication was presented as a tinyML® Talks webcast. The content reflects the
opinion of the author(s) and their respective companies. The inclusion of presentations in this
publication does not constitute an endorsement by tinyML Foundation or the sponsors.

There is no copyright protection claimed by this publication. However, each presentation is the work of
the authors and their respective companies and may contain copyrighted material. As such, it is strongly
encouraged that any use reflect proper acknowledgement to the appropriate source. Any questions
regarding the use of any materials presented should be directed to the author(s) or their companies.

tinyML is a registered trademark of the tinyML Foundation.

www.tinyML.org