AI applications at the very edge - - PowerPoint PPT Presentation

AI applications at the very edge

www.greenwaves-technologies.com loic.lietar@greenwaves-technologies.com

1

• Founded by industry veterans in November 2014
• Based in Grenoble area, France
• 16 people, going to 30
• Launched first IoT application Processor, GAP8,

in February 2018

About GreenWaves Technologies

Proprietary Information 3

Could that scale at IoT levels?

Cloud computing

+

• Installation costs $$$
• Privacy

Proprietary Information 4

Could that scale at IoT levels?

Cloud computing Edge computing

+ +

• Installation costs $$
• Bandwidth
• Privacy

Proprietary Information 5

Could that scale at IoT levels?

Cloud computing

+ +

Very edge computing Edge computing

• Installation costs $$
• Bandwidth
• Privacy

Proprietary Information 6

This can scale

Cloud computing

+ +

• Installation costs $
• Operation cost $
• Bandwidth
• Privacy

Very edge computing Edge computing

• Power supply
• Battery: 3.6V 3.6Ah A size battery, 2% loss per year

i.e. 250uW avg for 5 years

• PV: 10uW per cm2 for 50 lux indoor (ceiling) (direct sun is 100k lux)

i.e. 250uW avg for 50 cm2 12 hours a day

• Sensors
• Wireless communication
• Processing

Proprietary Information 7

Is the HW ecosystem ready?

Proprietary Information 8

Sensors

Se mW image fps # frames to snap an image uW (1 image/min) QVGA 2 30 6 6,7 VGA 230 200 1 19,2

• The numbers illustrate the potential of the technology power wise
• System wise, the sensors architecture and features only start to be

designed for IoT use e.g.

• some microphones now support wake-up-on-noise and low

resolution/lower power mode

• some image sensors offer multiple resolutions, but no image

sensor is meant for snapping one image at the time. Their use remains a bit of DIY

• Plenty of room for improvement
• QVGA, 30fps @ 2mW
• VGA global shutter with logarithmic

sensitivity, 200 fps @ 230mW

• IR 80x80, 10 fps @ 15mW
• 10Hz 2m radar @ 1mW
• microphone @ 300uW

… The devil is in the detail

Proprietary Information 9

Wireless communication, not a simple story

• As for sensors, raw computing power energy efficiency is not enough for IoT
• IoT devices have multiple states, each of which the processor shall match to

consume just the minimum necessary energy

GAP8 (up to 11GOPS)

• Deep sleep

1uA

• Data acquisition

50uA

• System control/light DSP

3 to 10 mW

• DSP like

20 to 80 mW

• Specialized computing (CNN)

+ 4 to 14 mW

Proprietary Information 10

Processing

uW (1 image/min) 10 objects recognition CNN 250 Face Detection Viola-Jones 8 Pedestrian Detection Weak predictors 47 10 objects recognition CNN 1250 GAP9 10 objects recognition CNN 313 QVGA VGA GAP8

Proprietary Information 11

In conclusion

we can support an event from once every few minutes to few times per minute,

• n average, for 5 years (battery) or “for ever” (PV cell)

uW 250

• nce a minute

image sensor 7 10 objects recognition 250 Face Detection 8 Pedestrian Detection 47 image sensor 19 10 objects recognition 1250 10 kbit/s 24 1 kbit/s 240 Power QVGA LoRa VGA

• To minimize energy consumption, we also go for hierarchical device architecture,

from always-on inaccurate low power sensor to higher power higher resolution sensor

• e.g.
• a PIR sensor detects life heat
• the camera is turned on during the day (an IR camera during the night) and captures an

image

• the processor looks for a person in the image
• a basic microphone detects a sound
• the microphone array is turned on and captures an audio sequence
• the processor calculates the direction of the sound and defines the window of interest
• the camera is turned and captures an image
• the processor looks into the window of interest for what the algorithm as been trained for

Proprietary Information 12

In real life

Proprietary Information 13

But are there uses for very edge computing?

≠

HD video stream Orientable Unlimited computing power QVGA to 1Mpixel images (and others) Fixed Limited computing power

Applications

Proprietary Information 14

Traffic classification Abnormal noises identification Counting cars, bicycles Spotting stopped cars on the road side License plate reading Counting people in meeting room Estimating occupancy in cafeteria Hotel room entrance monitoring Preventive maintenance, monitoring sound and vibrations Detecting face Face recognition (few, many) Retail, detecting pedestrians

Proprietary Information@ 15

Applications

Inattention detection Intrusion classification Window shock classification Vital signs monitoring Gesture recognition Key words spotting

Thank you

www.greenwaves-technologies.com

16